New Books Network · Pop Science Channel · Private

Interview Console

A working record of completed episodes and the books on the flight path — chosen for rigor, not reach.

Completed Interviews

Newest first. Full archive at the NBN host profile.

May 30, 2026
The Master AlgorithmPedro Domingos
Basic Books
2018
May 24, 2026
JHU Press
2025
May 22, 2026
Huge NumbersRichard Elwes
Basic Books
2026
May 5, 2026
Beyond the QuantumAntony Valentini
Oxford UP
2026
Apr 30, 2026
Becoming MartianScott Solomon
MIT Press
2026
Apr 27, 2026
New Atlantis
2026
Apr 22, 2026
Canongate
2025
Apr 20, 2026
MIT Press
2026
Apr 12, 2026
The Invisible UniverseMatthew Bothwell
Simon & Schuster
2021
Apr 8, 2026
Bloomsbury
2025
Mar 19, 2026
Radio UniverseEmma Chapman
Hachette UK
2026
Feb 27, 2026
ConsciousnessAlan J. McComas
American Scientist
2025
Feb 24, 2026
arXiv
2025
Feb 18, 2026
Addiction, IncEmily Dufton
U Chicago Press
2026
Feb 15, 2026
FSG
2022
Feb 4, 2026
The Laws of ThoughtTom Griffiths
Henry Holt
2026
Jan 30, 2026
The Tree of LifeMax Telford
W.W. Norton
2025
Jan 20, 2026
U Minnesota Press
2026
Jan 16, 2026
Columbia UP
2023
Jan 13, 2026
Harvard UP
2025
Jan 9, 2026
Light in the DarknessFalcke & Römer
HarperCollins
2021
Dec 31, 2025
Weathering SpaceWelsh, Gale, Karam
American Scientist
2026
Dec 24, 2025
Bloomsbury
2025
Dec 16, 2025
The Genius BatYossi Yovel
St. Martin's Press
2025
Dec 15, 2025
MIT Press
2025
Dec 6, 2025
SubmergedHenry Rausch
Independent
2024
Dec 3, 2025
U Chicago Press
2025
Nov 25, 2025
The Random UniverseAndrew H. Jaffe
Yale UP
2025
Nov 13, 2025
American Scientist
2025
Nov 11, 2025
Why Whales SingEduardo Mercado III
JHU Press
2025
Nov 7, 2025
SupermassiveJames Trefil & Shobita Satyapal
Smithsonian Books
2025
Nov 7, 2025
Apollo
2025
Nov 5, 2025
Springer Nature
2025
Oct 31, 2025
Basic Books
2025
Oct 30, 2025
Light on DarknessCosima Clara Gillhammer
Reaktion Books
2025
Oct 29, 2025
The Emergent MindGaurav Suri & Jay McClelland
Basic Books
2025
Oct 28, 2025
Stripe Press
2025
Oct 24, 2025
Space OdditiesHarry Cliff
Doubleday
2024
Oct 21, 2025
The Great Balancing ActJeffrey D. Sharon
Columbia UP
2025
Oct 17, 2025
The Giant LeapCaleb Scharf
Hachette UK
2025
Oct 14, 2025
Aevo UTP
2025
Oct 11, 2025
Uncharted TerritoryChris Dalla Riva
Bloomsbury
2025
Oct 10, 2025
U Chicago Press
2023
Oct 4, 2025
Diversion Books
2025
Oct 3, 2025
The Future of SeeingDaniel K. Sodickson
Columbia UP
2025
Sep 29, 2025
Imperfect OracleCass R. Sunstein
APS Press
2025
Sep 27, 2025
Everything EvolvesMark Vellend
Princeton UP
2025
Sep 25, 2025
Making Sense of ChaosJ. Doyne Farmer
Yale UP
2024
Sep 23, 2025
Facing InfinityJonas Enander
The Experiment Press
2025
Sep 13, 2025
The Magic of CodeSamuel Arbesman
PublicAffairs
2025
Sep 3, 2025
Dots and LinesAnthony Bonato
Johns Hopkins UP
2025
Sep 3, 2025
Harvard UP
2024
Aug 27, 2025
Picador
2024
Aug 21, 2025
Is Earth Exceptional?Mario Livio & Jack Szostak
Basic Books
2024
Aug 20, 2025
AI ValleyGary Rivlin
HarperCollins
2025
Aug 13, 2025
Sourcebooks
2022

Next 25

Ranked by v3 algorithm · five-year window · authors verified against completed list · Will this produce a first-rate interview?

1
Quanta and Fields
Sean Carroll · Dutton · 2024
111
Read first
2
What Is Intelligence?
Blaise Agüera y Arcas · MIT / Antikythera · Sep 2025
110
Pursue now
3
The Biggest Ideas in the Universe, Vol. 1
Sean Carroll · Dutton · 2022
109
Read first
4
The Physicist's Way
P. J. E. Peebles · Princeton UP · Oct 2026 ↑
107
Initiated
5
The Last Animal
Kyle Harper · Princeton UP · Oct 2026 ↑
105
Fall 2026
6
Earth and Life
Andrew H. Knoll · Princeton UP · Mar 2026
103
Pursue now
7
Blueprints
Marcus du Sautoy · Basic Books · 2025
102
Pursue now
8
The Man Who Broke Reality
Philip Ball · U Chicago Press · Nov 2026 ↑
102
Fall 2026
9
The Primacy of Doubt
Tim Palmer · Oxford UP · 2022
100
Pursue now
10
Proof
Adam Kucharski · Basic Books · 2025
100
Pursue now
11
The Brain, In Theory
Romain Brette · Princeton UP · Apr 2026
100
Pursue now
12
On Time
Jim Al-Khalili · Princeton UP · Sep 2026 ↑
100
Initiated
13
Battle of the Big Bang
Afshordi & Halper · U Chicago Press · May 2025
98
Sent
14
Brains, Minds, Machines
Poggio & Magrini · MIT Press · Aug 2026 ↑
98
Fall 2026
15
Six Math Essentials
Terence Tao · Quanta / Macmillan · Nov 2026 ↑
96
Fall 2026
16
Everything Is Fields
David Tong · Quanta / Macmillan · Winter 2027 ↑
96
Winter 2027
17
The Waltz of Reason
Karl Sigmund · Basic Books · 2023
95
Read first
18
Discordance
Jim Baggott · Oxford UP · Jan 2026
95
Pursue now
19
The Wilderness of the Infinite
Paolo Mancosu · Oxford UP · Apr 2026 ↑
95
Pursue now
20
Heart of Science
Jacob Stegenga · U Chicago Press · 2026 ↑
95
Pursue now
21
Emergence
David Sussillo · Grand Central · Mar 2026
90
Pursue now
22
The Man from the Future
Ananyo Bhattacharya · W.W. Norton · 2022
89
Read first
23
The Fox, the Shrew, and You
Rogier Mars · Oxford UP · 2026 ↑
89
Fall 2026
24
Life as No One Knows It
Sara Walker · Riverhead · 2024
87
Read first
25
What Are the Odds?
Mark Prell · Harvard UP · Apr 2026
85
Pursue now
v3 · June 2026 · Score = M(25) + A(20) + P(15) + R(15) + F(15) + T(10) + NSD(15) + CY(10) · max 125 · Five-year window enforced · Status: Pursue now = published, no contact · Initiated = outreach sent · Read first = queued · Fall / Winter = forthcoming

Archive Taxonomy & Flight Path Scoring

Seven intellectual buckets organizing the ~60-episode archive, plus a weighted composite score for the eleven flight path targets — algorithmic sorting to mirror the channel's editorial standards.

Seven Intellectual Buckets

Bucket I

Mathematical Foundations & Logic

Pure mathematics, formal systems, proof, completeness, and incompleteness. Books about what mathematics is before asking what it does. The Gödel thread.

Qualifier: Must engage with the structure of mathematical reasoning itself — numbers, proofs, infinity. Not applications.
Archive examples: Fantastic Numbers (Padilla) · The Laws of Thought (Griffiths) · The Necessities Underlying Reality (Franklin & Joaquin)
Bucket II

Physics & Cosmology

Theoretical physics, cosmology, quantum mechanics, and the mathematical scaffolding beneath physical law. Where the equations describe something real.

Qualifier: Must go beyond popularization — genuine argument about what physical laws mean. Preference for books willing to show the mathematics.
Archive examples: Beyond the Quantum (Valentini) · The Random Universe (Jaffe) · Radio Universe (Chapman) · The Invisible Universe (Bothwell) · Light in the Darkness (Falcke & Römer)
Bucket III

Information, Computation & AI

Information theory, computing, algorithmic thinking, neural networks, and the mathematical basis of artificial intelligence.

Qualifier: Must treat information or computation as a genuinely deep concept — not technology journalism. Shannon, Turing, entropy as a physical quantity.
Archive examples: The Laws of Thought (Griffiths) · On the Future of Species (Woolfson)
Bucket IV

Complexity, Systems & Emergence

Power laws, scaling, network theory, evolutionary dynamics, and the mathematics of how large systems behave. West, Barabási territory.

Qualifier: Must make a genuine scientific argument about structure or pattern — not "everything is connected." Quantitative claims required.
Archive examples: The Organism Is a Theory (Longo & Nocek) · Innovation and Adaptation in War (Tattar)
Bucket V

Philosophy of Science & Epistemology

What science can claim to know, the limits of mathematical modeling, the relationship between beauty and truth, and the classical questions about reason itself. The Thomistic thread lives here.

Qualifier: Must interrogate foundations, not just describe science. Hossenfelder, Sigmund, Vienna Circle territory. Aquinas is in genuine conversation here.
Archive examples: The Necessities Underlying Reality (Franklin & Joaquin) · Ripples on the Cosmic Ocean (Degroot) · The Pale Blue Data Point (Willis)
Bucket VI

Mind, Consciousness & Cognitive Science

Neuroscience, philosophy of mind, perception, memory, and imagination — the scientific study of inner experience at the rigorous edge.

Qualifier: Must bring scientific or philosophical rigor to questions about mind — not self-help, not pop psychology. Zeman, Dehaene, Chalmers territory.
Archive examples: A Sense of Things Unseen (Zeman) · The Shape of Things Unseen (Zeman) · Our Brains, Our Selves (Husain) · Consciousness (McComas) · Lectures in Neuroscience (Yuste)
Bucket VII

Biology & Life Sciences

Cellular biology, evolution, origin of life, genetics, and the mathematical structures beneath living systems. Where biology engages with mechanism, not metaphor.

Qualifier: Must engage with mechanism, structure, or principle — not popular natural history. Cellular, molecular, or evolutionary depth required.
Archive examples: Becoming Martian (Solomon) · The Tree of Life (Telford) · The Genius Bat (Yovel) · The Unreasonable Likelihood of Being (Endres)

Composite Quality Score v3.0 · NSD & CY Extension

Formula:  Score = M(25) + A(20) + P(15) + R(15) + F(15) + T(10) + NSD(15) + CY(10)  ·  max 125   |   M Aspiration · A Author · P Press · R Reception · F Freshness · T Topical Fit · NSD Narrative Structural Density · CY Conversation Yield

M Aspiration: Formula=25 · Rigorous=18 · Accessible=10 A Author: Active researcher top inst=20 · Established=15 · Elite journalist=12 · Generalist=8 P Press: Top academic=15 · Premier trade=12 · Standard trade=10 · Other=6 R Reception: GR base (4.3+=8 · 4.1-4.2=6 · 3.9-4.0=4) + prize bonuses to +7 F Freshness: Forthcoming or <12mo=15 · 1–3yr=12 · 3–5yr=8 · 5–10yr=4 · 10+yr=0 T Topical Fit: Hits 2+ buckets strongly=10 · One strong=7 · Tangential=3 NSD (new): Pacing, layering, historical sequencing, readability under rigor. Anchors: Gleick/Sigmund=15, Strogatz=14, Hossenfelder=5 CY (new): Interview elasticity, metaphor generation, ability to sustain deep conversation. Anchors: Carroll=10, Hossenfelder=9, Tao=7
RankT#TitleYrBucketTier v2 (100)NSDCYv3 (125)Primary Driver
#112What Is Intelligence?2025III · InformationRigorous89119109Top v2 score + active public communicator (TEDx, Long Now, Berggruen)
#213Quanta and Fields2024II · PhysicsFormula841310107Carroll = NSD calibration tier + Mindscape host (highest CY in corpus)
#309The Biggest Ideas in the Universe, Vol. 12022II · PhysicsFormula811210103Biggest v2→v3 mover among existing eleven (+22) — Carroll CY=10 carries it
#411The Waltz of Reason2023V · EpistemologyRigorous81138102Sigmund = NSD calibration tier ("wit and humor shine") + Thomistic thread
T#514Proof: The Uncertain Science of Certainty2025V · EpistemologyRigorous8299100Reviews note meandering structure (NSD 9 not 13), but active speaker = strong CY
T#515Blueprints2025I · FoundationsRigorous781210100du Sautoy polymath sensibility + BBC presenter (highest CY in 2025 cohort)
T#706Gödel, Escher, Bach1979I · FoundationsRigorous7515797Max NSD=15 (conceptual layering benchmark) offsets reclusive author CY
T#722The Primacy of Doubt2022V · EpistemologyRigorous7810997Oxford FRS + triple-bucket fit (V/IV/II) + active outreach line · added May 20
#910Exact Thinking in Demented Times2017V · EpistemologyRigorous7114893Sigmund anchor — Vienna Circle narrative pacing is calibration-tier
T#1016Six Math Essentials FORTHCOMING2026I · FoundationsRigorous7510792Tao prose clear in Intelligencer columns — CY moderate (less media presence)
T#1017Everything Is Fields FORTHCOMING2026II · PhysicsRigorous7511692Tong's lecture notes legendary for clarity — but limited podcast presence caps CY
T#1202Infinite Powers2019I · FoundationsRigorous6814991Strogatz NSD-calibration tier ("gold standard") + frequent media · jumps T#14→T#12
T#1218Life as No One Knows It2024IV · ComplexityRigorous7381091Polarized reception caps NSD, but Walker is highly podcast-active (Rogan, Fridman)
#1401When Einstein Walked with Gödel2018V · EpistemologyRigorous6813889Holt's essay form widely praised as "finest science writing" — strong NSD
T#1507The Man from the Future2022III · InformationRigorous6912788Biographical narrative + intellectual depth (TLS, WSJ praise)
T#1508Scale2017IV · ComplexityRigorous6613988West NSD-calibration tier ("surprising central result") + famous TED talk
#1719The Proof in the Code FORTHCOMING2026I · FoundationsRigorous7011687Hartnett's Quanta journalism consistently strong — journalist author limits CY
#1821Air-Borne2025VII · BiologyAccessible6212983Polished NYT-columnist pacing + active speaker — biggest v2→v3 mover (+21)
T#1905Lost in Math2018V · EpistemologyRigorous685982"Rigorous but narratively inert" anchor — NSD floor confirmed; YouTube CY high
T#1903The Information2011III · InformationRigorous6015782Gleick = max NSD calibration anchor — narrative partially redeems Freshness=0
#2120The Language of Mathematics2025I · FoundationsAccessible667578Short-entries format limits cumulative momentum (NSD 7); limited media (CY 5)
#2204Prime Obsession2003I · FoundationsFormula5410367Alternating-chapter conceit earns NSD; author's public reputation caps CY
Algorithm note · v3.0 NSD/CY extension · May 20, 2026 A v3 calibration spec arrived (forwarded as "from the AI siblings" — Gemini, ChatGPT, Grok) proposing two new factors: Narrative Structural Density (NSD, 15 pts) — pacing, layering, historical sequencing, readability under rigor; and Conversation Yield (CY, 10 pts) — interview elasticity, metaphor generation, ability to sustain deep conversation. Both are adopted at their proposed weights, with the composite ceiling extended to 125 rather than redistributing the v2 100-point pool, preserving backward-comparable scores. What the new factors do: Carroll Vol. 1 climbs from T#4 to #3 (Mindscape host carries CY=10); Gleick's Information rises from #20 to a tie at T#19 (max NSD partially redeems Freshness=0); Strogatz's Infinite Powers climbs T#14→T#12; Air-Borne is the biggest v2→v3 mover at +21. Hossenfelder's Lost in Math gets minimal NSD lift (NSD=5, the "rigorous but narratively inert" calibration floor) — confirming the spec's diagnostic. Spec resolutions (May 20, 2026): (a) the spec's reference to The Foundations of Reason by Sigmund — confirmed as a slip for The Waltz of Reason (2023), already Target 11; (b) Palmer's The Primacy of Doubt added as Target 22 — host has not yet read it but intends to email Palmer for interview, which is exactly Flight Path territory; book scored on its merits and lands at T#7 (97); (c) Yanofsky's Outer Limits of Reason not added — no signal from host. Top-of-list: What Is Intelligence? holds #1 at 109 — the highest unified score this system has produced.
Space Exploration Technologies Corp · Founded 2002 · Hawthorne, CA

SpaceX: The Audacity of the Possible

From a failed Russian missile deal to reusable rockets, Mars ambitions, and a vertically integrated aerospace empire.

433Total Falcon 9 Launches
99%Landing Success Rate
20+Max Booster Reuses
80%Hardware Built In-House
2002Year Founded
The Problem With Space Being Boring

Elon Musk was born in South Africa in 1971. After degrees in economics and physics at Penn, he moved to Silicon Valley in 1995 and built two companies: Zip2 (sold to Compaq for $307M in 1999) and PayPal (sold to eBay for $1.5B in 2002). He walked away with roughly $180 million and an uncomfortable realization — profitable internet companies were solving low-impact problems.

In 2001 he moved to Los Angeles, deliberately placing himself inside the world's densest concentration of aerospace talent. He joined the Mars Society board, donated to Mars research, and built relationships with engineers who shared his frustration: NASA had no credible plan to send humans to Mars, and the public had grown cynical about space entirely.

"At first I thought NASA just had a badly designed website. It turned out, NASA had no plans for Mars — and a policy that didn't even let them talk about it." — Elon Musk

He met Robert Zubrin, an aerospace engineer and Mars Society co-founder who noticed something unusual about Musk: unlike other wealthy investors who funded projects from a distance, Musk put the full force of his technical obsession into the work. In 2001 he knew almost nothing about rocket engines. By 2005 he knew everything about them.

SPACEX HAWTHORNE, CALIFORNIA · EST. 2002
SpaceX Headquarters · Hawthorne, California · Founded June 2002
The Russia Trip and the Idiot Index

Musk's first plan was not to build a rocket — it was to buy one. In early 2002 he flew to Moscow twice, attempting to purchase refurbished intercontinental ballistic missiles for $20 million. The Russians refused to deal. On the flight home, staring at a spreadsheet, Musk had the insight that would define the next two decades.

The raw materials cost of a rocket was only a tiny fraction of its retail price. He formalized this as the Idiot Index: the ratio of a finished product's cost to its raw material cost. A catastrophically high Idiot Index meant the aerospace supply chain — not physics — was the primary cost driver. His response was total vertical integration: build everything in-house.

Space Exploration Technologies Corporation was incorporated in June 2002 with $100 million of Musk's own money. That figure was chosen deliberately: large enough to be serious, and large enough that no board could seize control, as had happened at both Zip2 and PayPal. He would be CEO this time, and the company would stay private.

LOX INLET RP-1 INLET TURBOPUMP NOZZLE BELL MERLIN 1D ENGINE 934 kN THRUST · RP-1 / LOX · DESIGNED 2002
The Merlin 1D engine — designed by Tom Mueller in 2002, nine power every Falcon 9 first stage
$100MMusk's founding investment
Russia negotiations (2 Moscow, 1 California)
$20MMusk's offer for 3 ICBMs
2002SpaceX incorporated
The Silicon Valley Space Company
STAGE 1 BODY 9 MERLIN ENGINES NOSE CONE HORIZONTAL ASSEMBLY · HAWTHORNE FACTORY 50¢/SQ FT vs $12–18/SQ FT FOR VERTICAL ASSEMBLY
Manufacturing Philosophy
Rockets built horizontally — the anti-aerospace factory that changed the economics of space.

SpaceX set up in a warehouse in Hawthorne, California. The culture was deliberately anti-aerospace: no bureaucratic hierarchy, open-concept floors where PhD engineers worked alongside machinists and welders, insanely ambitious timelines, and a philosophy borrowed from software — ship fast, break things, iterate. Musk personally interviewed the first 3,000 employees, over 1,500 hours of interviews, ensuring every hire shared the mission.

The talent strategy was surgical. When Musk wanted to hire a talented engineer whose wife worked at Google in San Francisco, making relocation to LA impractical, he called Larry Page directly and arranged a Google transfer for the wife to the LA office. When he needed propulsion expertise, he found Tom Mueller — an aerospace engineer who had dreamed of building rockets since childhood despite his father's skepticism. Mueller designed SpaceX's Merlin engine in 2002 and became a cornerstone of the company's engineering identity.

SpaceX contracted with suppliers outside traditional aerospace — automotive and industrial vendors where cost discipline was far sharper — and built everything else in-house. When a supplier's tubing prices were too high, SpaceX simply learned to weld its own fuel tanks and dropped the supplier entirely.

Merlin to Raptor: Two Generations of Power
LOX RP-1 TURBOPUMP CHAMBER NOZZLE BELL
Generation 1 · 2002
Merlin Engine
Designed by Tom Mueller in 2002. Burns RP-1 kerosene and liquid oxygen. Nine Merlins are arranged in a circular "Octaweb" pattern on the Falcon 9 first stage — chosen to minimize weight and simplify maintenance. One vacuum-optimized Merlin powers the second stage. The engine that took SpaceX from startup to the world's most prolific launch provider.
934 kN thrust (sea level) · RP-1 / LOX · Turbopump-fed
FUEL PREBURNER OX PREBURNER TURBO TURBO 230 bar HIGHEST EVER
Generation 2 · Starship Era
Raptor Engine
Full-flow staged combustion engine burning liquid methane and liquid oxygen — a more complex but more efficient cycle than Merlin. Achieves the highest chamber pressure of any production rocket engine ever built. Up to 33 Raptors power the Super Heavy booster; 6 power the Starship upper stage. Designed entirely in-house at Hawthorne and Starbase.
230 bar chamber pressure — highest of any production engine · LCH4 / LOX
Three Failures and a Company on the Edge
  • March 2006 Failure Falcon 1 · Flight 1. Fails 33 seconds after launch. A fuel leak causes an engine fire. The rocket barely clears the launch tower before coming down.
  • March 2007 Failure Falcon 1 · Flight 2. Reaches space but fails to achieve orbit. Fuel slosh causes the second stage to tumble. Nearly kills the company.
  • August 2008 Failure Falcon 1 · Flight 3. Reaches Max-Q successfully but first and second stages collide during separation. SpaceX is nearly out of money. Tesla is simultaneously near bankruptcy.
  • September 28, 2008 Success Falcon 1 · Flight 4. The last launch SpaceX could afford. Reaches orbit. The first private company to do so on self-developed hardware. The team weeps.
  • December 2008 Lifeline NASA awards $1.6 billion Commercial Resupply Services contract. SpaceX survives. The commercial era of spaceflight begins.

After the third failure, Musk sent a letter to the entire company. He did not blame anyone. He emphasized resilience:

"SpaceX will not skip a beat in execution going forward. There should be absolutely zero question that SpaceX will prevail. For my part, I will never give up, and I mean never."

The fourth launch was the last one the company could afford. Musk had been sleeping on friends' couches and borrowing money for living expenses, simultaneously fighting to save Tesla. When the Falcon 1 reached orbit on September 28, 2008, it became the first privately developed liquid-fueled rocket to do so. Three months later, the NASA contract secured the company's future.

"It took six years — about four and a half more than Musk had once planned — and five hundred people to make this miracle of modern science and business happen." — Ashlee Vance, Elon Musk

The Fleet: From Falcon 1 to Starship
Design Insight · The Octaweb
Why Nine Engines?

Unlike the Falcon 1's single engine, the Falcon 9 uses nine Merlins arranged in a circular "Octaweb" pattern — one center engine surrounded by a ring of eight. This isn't just about raw thrust. It's about redundancy.

If any single engine fails mid-flight, the remaining eight can compensate and still complete the mission successfully. This engine-out capability was a deliberate design choice that lets SpaceX genuinely promise reliability to commercial customers.

Each engine also has a Kevlar debris shield protecting it from shrapnel if a neighboring engine fails — a feature absent on Falcon 1. Musk claims the Merlin achieves higher performance than any other gas-generator cycle kerosene engine ever built.

CENTER 1 2 3 4 5 6 7 8 OCTAWEB ENGINE LAYOUT 9 MERLIN 1D ENGINES · ENGINE-OUT CAPABLE
Design Insight · Two Stages
The Kestrel Problem — Why Falcon 9 Uses Merlin Everywhere

The Falcon 1 used a Kestrel engine for its second stage — a different, simpler engine than the first stage's Merlin. But the Kestrel has a lower specific impulse (a measure of fuel efficiency and thrust) than the Merlin. For the much heavier Falcon 9, that performance gap was unacceptable.

SpaceX's solution: use a vacuum-optimized Merlin on the second stage too — the same engine family as the first stage, but with a larger nozzle optimized for the vacuum of space where there's no atmospheric pressure to fight. Musk claims this Merlin achieves higher performance than any other gas-generator cycle kerosene engine ever built.

The cost advantage is hidden: using the same engine family means the same manufacturing tools, same materials, same procedures for both stages. No retooling. No separate supply chain. Pure savings through repetition.

Design Insight · Falcon Heavy
The Heavy: Three Falcon 9s Strapped Together

The Falcon 9 Heavy is conceptually simple: take a Falcon 9 core, strap two additional Falcon 9 first stages on either side as boosters, and suddenly you have 27 Merlin engines firing simultaneously at liftoff. Each side booster adds 9 engines — totaling 18 extra engines working as boosters for heavy payloads.

This modular approach — reusing proven hardware at larger scale — is signature SpaceX thinking. Rather than designing a new rocket from scratch, multiply what already works. Falcon Heavy can lift over 63,800 kg to LEO, making it the most powerful operational rocket in the world until Starship.

STAGE 1 STAGE 2 FALCON 1 1 MERLIN · 2-STAGE · 21m TALL
2006 – 2009
Falcon 1
SpaceX's first orbital rocket. Two stages, one Merlin engine. Three failed launches, one historic success. The proof of concept that made everything else possible.
2 Stages 1 Merlin 670 kg to LEO RP-1/LOX
FALCON 9 DRAGON GRID FIN LANDING LEG 9 MERLINS
2010 – Present
Falcon 9 + Dragon
The world's most frequently launched rocket. Nine Merlin engines in the Octaweb. First stage lands and reflies — single boosters have completed 20+ missions. Dragon capsule delivers cargo and crew to the ISS.
9 Merlins 22,800 kg to LEO Reusable 1st stage 430+ launches
9 9 9 = 27 MERLINS TOTAL FALCON HEAVY
2018 – Present
Falcon Heavy
Three Falcon 9 cores strapped together — 27 Merlin engines firing at liftoff. Side boosters land simultaneously. The most powerful operational rocket in the world until Starship. First mission carried Musk's Tesla Roadster toward Mars orbit.
27 Merlins 63,800 kg to LEO 3 Cores Reusable
FALCON 9 70m tall 120m STARSHIP 33 RAPTORS
2019 – Present
Starship / Super Heavy
The largest rocket ever built. Fully reusable two-stage system. Super Heavy booster: up to 33 Raptor engines. Stainless steel construction. Designed for Mars transit, lunar landing (Artemis III), and point-to-point Earth transport.
33 Raptors 150,000 kg to LEO Stainless steel LCH4/LOX
Reusability, Manufacturing, and the Starfactory

The Falcon 9 first stage was designed from day one to be recovered and reflown. After delivering its payload, the booster reignites its engines, deploys grid fins for steering, and lands vertically — either on a ground pad or an autonomous drone ship at sea with names like Of Course I Still Love You and A Shortfall of Gravitas.

The economics are transformative. A Falcon 9 first stage costs roughly $30–40 million to build. Refurbishment costs a fraction of that. By reflying a booster 20 times, SpaceX essentially flies for the cost of fuel and operations on flights 2 through 20. This is why SpaceX spent $2.5 billion to deliver four Dragon capsules to the ISS across fifteen flights — a price the rest of the industry literally cannot comprehend.

Horizontal assembly — building rockets lying on their sides — costs about 50¢ per square foot to operate. The vertical assembly buildings used by NASA and legacy contractors cost $12–18 per square foot, due to the logistics of moving people and equipment dozens of feet into the air. SpaceX's factory floor looks more like an automotive plant than an aerospace facility.

OF COURSE I STILL LOVE YOU ENTRY BURN GRID FINS LANDING! FALCON 9 BOOSTER RECOVERY SEQUENCE
The three-phase landing sequence: entry burn → grid fin steering → touchdown on drone ship.
SpaceX CATCH POINT "MECHAZILLA" LAUNCH TOWER CHOPSTICK ARMS SUPER HEAVY STARBASE · BOCA CHICA TX · THE BOOSTER CATCH SYSTEM
Mechazilla — the launch tower's "chopstick" arms catch the Super Heavy booster mid-air, eliminating landing legs entirely.

For Starship, SpaceX relocated development and production to Starbase in Boca Chica, Texas — now incorporated as its own city. What began as tents and trailers is now the Starfactory: a permanent manufacturing complex modeled on automotive assembly lines.

The production line uses linear adjacent flow: separate sections (top, middle, bottom) are built in parallel in adjacent bays, then joined in mega-bays. Target cadence: one Starship every three days. Long-term ambition: one per day.

Starship is built from a specialized stainless steel alloy — cheap, easy to weld, resilient across extreme temperatures, and polishable to reflect radiant heat during reentry. In an industry that defaults to expensive carbon fiber or aluminum-lithium, this was a first-principles choice. Physics said stainless steel worked. So SpaceX used stainless steel.

The scrappy innovation culture persists at every scale. In 2013, when SpaceX needed a Dragon docking prototype, engineers built one from mountain bike shocks and hardware catalog parts. They called it the McDocker. It worked.

The 5-Step Process: How SpaceX Builds What Others Won't Attempt

In August 2021, Tim Dodd from the Everyday Astronaut spent 2.5 hours touring Starbase with Musk. During the conversation, Musk laid out the exact design and manufacturing framework SpaceX uses — noting that manufacturing at scale is a much bigger challenge than design. The five steps must be followed in order. Musk has admitted to making the mistake of going backwards on all five multiple times, including on the Tesla Model 3 battery pack.

01
Make the Requirement Less Dumb

"The requirements are definitely dumb; it does not matter who gave them to you. It's particularly dangerous when they come from an intelligent person." Every requirement must have a named person behind it — not a department. If you can't ask a person, you can't question it.

02
Delete Part of the Process

"If parts are not being added back at least 10% of the time, not enough parts are being deleted." The bias toward adding things is overwhelming. For a rocket trying to be fully reusable, tight margins are everything. If you're not cutting, you're not serious.

03
Simplify or Optimize

"The most common error of a smart engineer is to optimize something that should not exist." This step is third — not first — deliberately. You must delete the unnecessary before you optimize what remains. Otherwise you're perfecting a mistake.

04
Accelerate Cycle Time

"You're moving too slowly, go faster. But don't go faster until you've worked out the other three things first." Speed without clarity is chaos. Speed after clarity is compounding advantage. Most organizations try to accelerate before simplifying.

05
Automate

"Then the final step is automate it." Only automate what has already been simplified and validated. Automating a flawed process just produces flaws faster. Musk admits he made the mistake of automating too early on the Tesla Model 3 battery pack line.

The Iteration Doctrine
Why SpaceX Blows Things Up on Purpose

SpaceX has three different failure thresholds for its three vehicle families, and they are deliberately asymmetric. Dragon — which carries humans to the ISS — must never fail. It is tested to extreme margins with massive redundancy built in. Falcon 9 sits in the middle: landing failures are acceptable, but ascent failures are not, because payloads are at stake.

Starship is the opposite extreme. Because no humans are on board during test flights, every flight is an opportunity to push the vehicle past its limits and collect failure data. The goal is not to succeed — it is to learn as fast as possible.

"Frankly, if you don't push the envelope, you cannot achieve the goal of a full and rapidly reusable rocket. You have to go close to the edge on margins. You just can't avoid it." — Elon Musk, Starbase Tour 2021

This is the inverse of NASA's Space Shuttle logic. The Shuttle's crewed nature meant every design change was high risk and low reward — engineers who flagged problems faced punishment if a change went wrong, and only small reward if it went right. The result: the shuttle's design froze. SpaceX's design never freezes. Every single Starship and booster has had significant iterations from the one before it.

April 20, 2023 · Integrated Flight Test 1
"Rapid Unscheduled Disassembly" — and Why It Was a Success
Max Altitude
39 km
Flight Duration
<4 min
Improvements Identified
1,000+
Milestones Cleared
2 / 3

Starship cleared the launchpad — the single most critical milestone, given the propellant capacity of 3,400 tonnes in the first stage alone. It passed Max Q (maximum aerodynamic stress). Stage 2 never separated, and the vehicle was remotely destroyed over the Gulf of Mexico.

Within 24 hours, the team had identified over 1,000 discrete improvements — including a critical launchpad redesign. The original concrete pad was heavily damaged; the solution was a water-cooled steel plate to absorb the heat from 33 Raptor engines firing simultaneously. SpaceX assigned Musk's pre-launch 50/50 odds of clearing the pad. It cleared it.

Starship in Numbers
Height (fully stacked) 390 ft / 119m
Payload to LEO vs Falcon Heavy 4–5× more
Cost per 100 MT to orbit (old) $15 million
Cost per 100 MT to orbit (Starship) $2 million
Cost reduction 80%+
Passenger capacity Up to 100
The Catch System · Mechazilla
Why Starship Doesn't Have Landing Legs

The Falcon 9 booster lands on four deployable legs. It's elegant, proven, and heavy. Landing legs that can withstand the forces of a Falcon 9 touchdown weigh thousands of kilograms — weight that could otherwise be payload. For Starship, which must carry enough propellant for a round trip to Mars, that weight penalty is unacceptable.

SpaceX's solution: don't land on legs at all. Instead, the 145-meter launch tower at Starbase is equipped with two massive mechanical arms — nicknamed Mechazilla by the SpaceX community, officially the "Chopstick" arms. As the Super Heavy booster returns from its job of lifting Starship to altitude, it falls back toward the tower, decelerates using its engines, and at the last moment is physically caught mid-air by the arms as they close around the booster's grid fins.

"The launch tower for Starship will 'catch' the booster on its return. There needs to be a lot of failures and iterations before that happens. Before both stages are fully reusable. Before daily launches. Before Mars." — Trung Phan, SatPost 2023

The Mechazilla catch has several advantages beyond weight. The booster is caught at the launch mount itself, meaning it can be immediately re-stacked with a new Starship upper stage and relaunched — potentially the same day. No ferry trips from a drone ship. No leg inspection and refurbishment cycle. The goal: a booster that flies multiple times per day, like a commercial aircraft.

There's also a Mars logic to this decision. A launchpad that doesn't require a complex permanent structure is a launchpad that could, in theory, be rebuilt on another planet. Musk has specifically noted that the catch system's relative simplicity — compared to constructing massive landing infrastructure — is a feature, not a constraint.

OLM TOWER 145 METERS SpaceX Grid fins CATCH ZONE 33 Raptors (shutdown) descending MECHAZILLA · BOOSTER CATCH SYSTEM · STARBASE TX
The Ecosystem: How the Companies Connect

SpaceX sits at the center of a constellation of Musk-founded companies, each solving a different layer of the same ultimate problem: how do you build a self-sustaining civilization on another planet? The companies share talent, manufacturing philosophy, and institutional knowledge in ways no traditional aerospace organization could replicate.

🚗⚡
Energy & Manufacturing
Tesla
Battery technology, solar systems, and energy storage essential for a Mars colony. Tesla's Gigafactory manufacturing philosophy directly mirrors SpaceX's. Talent flows both directions between companies constantly.
🛰️
Revenue Engine
Starlink
6,000+ low-Earth-orbit internet satellites generating recurring commercial revenue that funds Starship development. Now a U.S. foreign policy asset — governments encourage international adoption in trade negotiations.
🧠
Intelligence Layer
xAI / Grok
Founded 2023. Raised $12B by late 2024. AI systems Musk views as essential for autonomous spacecraft operations and the broader technological infrastructure of a multi-planetary civilization.
🧬
Human Augmentation
Neuralink
Brain-computer interface company. First Canadian human trial approved 2025. Long-term thesis: human-AI symbiosis is necessary for survival as a species, especially relevant to long-duration spaceflight.
🕳️
Underground Infrastructure
The Boring Company
68-mile tunnel network beneath Las Vegas. Mars colony habitation will require underground infrastructure to shield inhabitants from radiation — this is the R&D proxy. PruFrock 3 tunneling machine now deployed at Tesla Gigafactory.
🤖
Automated Labor
Tesla Optimus
Humanoid robot program. A Mars colony of millions requires automated labor at scale before human settlers arrive in sufficient numbers. Optimus is the advance worker. Tesla has pivoted significantly toward humanoid robotics in 2025.
First Principles, the Long Game, and the Declaration
EARTH MARS TRANSFER ORBIT ~7 MONTH TRANSIT THE MISSION: MULTI-PLANETARY SPECIES
The Mars transfer trajectory — Starship's ultimate purpose. A 7-month transit to make humanity multi-planetary.

SpaceX's success is structural, not merely personal. Musk applied a small number of reasoning principles with unusual consistency, and they compounded over twenty years.

The Idiot Index: find where process markup exceeds physics, then eliminate the process. Rockets had a catastrophic Idiot Index. SpaceX internalized the supply chain.

First principles: derive every decision from fundamental physics, not industry convention. Why vertical assembly? Because everyone did it. Why carbon fiber? Because aerospace always had. Why single-use rockets? Because no one had tried otherwise. SpaceX challenged every assumption.

"The main reason I am personally accumulating assets is to fund this. I have no other motivation except to make the biggest contribution I can to making life multi-planetary." — Elon Musk, 2017

The private company strategy: Musk will not take SpaceX public until Mars is "secure" — until a self-sustaining colony could survive without Earth resupply. This removes quarterly earnings pressure and lets SpaceX absorb failures, delays, and bets that no public company could justify to shareholders. In a 2013 email to staff: "If being a public company diminishes [the likelihood of reaching Mars], then we should not do so until Mars is secure."

Whether the Mars colony materializes in Musk's lifetime or not, the intermediate outputs — the cheapest orbital launch in history, the first reusable orbital rockets, a global satellite internet constellation, a fully reusable megarocket — represent the most significant transformation of the space industry since Apollo.

Go Deeper
Cross-Model Synthesis · Five AI Systems · v1.1 · May 2026

AI Stocks — A Cross-Model
Reading Experiment

Five AI systems. Identical prompts. Independent lists. The goal was not to produce a consensus buy list — it was to see where different models converge, where they diverge, and what the pattern of agreement reveals about the structure of the AI economy itself. These are observed response tendencies in one prompt experiment, not settled verdicts.

AI Infrastructure — The Physical Layer
Hardware, foundries, networking, and power — the layer AI runs on regardless of which model or application wins. These names appeared most consistently when five systems were asked independently to rank by expected 10-year shareholder return. Consensus here is a signal worth examining, not a guarantee.
1
NVDA
NVIDIA
~$4.5–5.1T market cap
The only name on all five model lists — the clearest signal of the exercise. The full-stack hardware-software ecosystem — GPUs, CUDA, NVLink, networking — currently has no peer at scale. Ranked cautiously on valuation grounds by the more disciplined models, first on quality grounds by the rest. The open question every investor has to answer: at $4–5T, how much of the next decade is already in the price? Interview hook: What does CUDA dominance look like when inference, not training, becomes the primary workload?
Consensus
Grok ChatGPT Gemini Perplexity Claude
2
TSM
Taiwan Semiconductor
~$1.9–2.2T market cap
Four of five models. The foundry bottleneck for all advanced AI silicon — every chip NVIDIA, AMD, Apple, and Google design gets fabricated here. Tends to trade at a discount to the design layer it enables, which most models flag as an asymmetry worth considering. Geopolitical risk is real, priced in to varying degrees, and unlikely to resolve cleanly. Interview hook: If TSMC's Arizona fabs reach full 2nm production by 2028, does the geopolitical discount permanently close?
Consensus
Grok ChatGPT Gemini Perplexity Claude
3
ANET
Arista Networks
~$130–200B market cap
Three models independently. AI clusters require high-throughput Ethernet switching at a scale that appears to favour Arista's architecture as hyperscalers explore alternatives to InfiniBand. The networking sub-thesis tends to be less crowded than the semiconductor trade — which may mean more durable returns or simply less attention. Interview hook: Does Ethernet eventually displace InfiniBand entirely in AI training clusters, and if so, at what timeline?
Consensus
Grok ChatGPT Gemini Perplexity Claude
4
AVGO
Broadcom
~$1.8–2.1T market cap
Three models. Custom ASICs and networking silicon for AI data centers — ChatGPT's top infrastructure pick on the thesis that AI compute eventually becomes a networking and interconnect problem as much as a GPU problem. More diversified than pure-play compute, with strong hyperscaler relationships and VMware recurring revenue. Interview hook: Can Broadcom's custom silicon business scale to $100B+ by 2027 without customer concentration becoming an existential risk?
Consensus
Grok ChatGPT Gemini Perplexity Claude
5
VRT
Vertiv Holdings
~$60–124B market cap
Two models — Gemini and Perplexity — identified power and cooling as the most underappreciated bottleneck in AI buildouts. The physical case is sound: high-density chips generate heat densities that traditional air-cooling cannot handle. The valuation risk is equally real: VRT is up significantly and three models flagged it as a name where the narrative may be ahead of the execution. Interview hook: At what point does liquid cooling become commoditised, and which industrial competitors are best positioned to erode Vertiv's margin?
Consensus
Grok ChatGPT Gemini Perplexity Claude
Speculative AI Plays — Physical World Transformation
AI applied to transform real industries — logistics, healthcare, drug discovery, transportation. High variance by design. Ranked by estimated 10-year upside adjusted for probability of success, not certainty of outcome. Three names achieved unanimous consensus across all five systems — a convergence none of the models predicted in advance.
1
SYM
Symbotic
~$6.8–20B market cap
All five models — the only unanimous pick across both lists. Real revenue ($2.5B+ TTM), major customer validation (Walmart, AWG), net income positive as of Q2 FY2026, and a physical-world AI/robotics moat: robots bolted into warehouse floors are not easily displaced. Gemini flagged a valuation caution — SYM frequently tops consensus "AI robotics" lists and deployment delays can trigger multiple compression sharply. Interview hook: What happens to Symbotic's growth trajectory when Walmart's rollout completes — and is the next customer cohort large enough to sustain the multiple?
Consensus
Grok ChatGPT Gemini Perplexity Claude
2
RXRX
Recursion Pharmaceuticals
~$1.6–4B market cap
All five models — unanimous alongside SYM. The Recursion OS has generated $500M+ in validated milestones from Roche, Sanofi, and Bayer; pharma partners do not pay milestones to promotional companies. The upside case is genuine: a single Phase 3 success re-rates the platform across a $200B+ R&D market. So is the risk: Gemini's probability model assigns 40% odds of permanent capital loss if clinical candidates fail. Cash runway to early 2028. Interview hook: Does AI-accelerated drug discovery compress the ten-year Phase 1–3 pipeline, or does biological complexity reassert itself at every clinical inflection?
Consensus
Grok ChatGPT Gemini Perplexity Claude
3
TEM
Tempus AI
~$6.8–13B market cap
All five models — unanimous. One of the world's largest libraries of clinical and molecular data powering AI diagnostics for oncology, cardiology, and radiology. Q1 2026 revenue up 36% YoY to $348M; 2026 guidance raised to $1.59B; data licensing revenues growing ~70% at 62%+ gross margins. The network effect compounds as more patient data improves model accuracy, which drives physician adoption, which generates more data. Reimbursement and privacy regulation are the primary risk vectors. Interview hook: At what point does a clinical AI data platform become a regulated utility — and does that cap or protect its moat?
Consensus
Grok ChatGPT Gemini Perplexity Claude
4
IOT
Samsara
~$16.9–22B market cap
Four of five models — Grok is the sole holdout, classifying it as fleet-tracking software rather than AI infrastructure. That disagreement is itself instructive: IOT's AI thesis depends on believing that proprietary physical-world data compounds into something irreplaceable, not merely useful. ARR $1.54B+, growing ~30–37% YoY, FCF-positive. Gemini's highest conviction speculative pick at 8.5/10. Interview hook: If Samsara's physical-world data flywheel is genuinely irreplaceable, what prevents a hyperscaler from acquiring it before the moat fully compounds?
Consensus
Grok ChatGPT Gemini Perplexity Claude
5
JOBY
Joby Aviation
~$10–14B market cap
Perplexity and Claude — two models. ChatGPT explicitly excluded it: "more electric aviation / autonomy / regulatory execution than a clean AI moonshot." Gemini and Grok passed entirely. Included on the strength of TAM optionality and execution lead: $2.5B cash runway, Toyota's $500M commitment, Dubai 6-year exclusive rights, inaugural JFK–Manhattan flight completed. The autonomous Superpilot AI has logged 7,000+ miles. The lowest consensus pick — which makes it either the most contrarian or the most mispriced name on the list. Interview hook: Does urban air mobility require AI autonomy to be economically viable, or does the pilot model work at scale?
Consensus
Grok ChatGPT Gemini Perplexity Claude
Cross-Model Tally
Five AI systems received identical prompts. Results tallied by ticker frequency, then filtered by valuation discipline and fundamental quality. These are observed response tendencies in one prompt experiment — not fixed model personalities or settled investment verdicts.
Infrastructure Tally
Ticker Grok GPT Gemini Perp. Claude Score
NVDA5
TSM4
ANET3
AVGO3
VRT2
AMD2
MRVL1
MSFT1
Speculative Tally
Ticker Grok GPT Gemini Perp. Claude Score
SYM5
RXRX5
TEM5
IOT4
AUR3
JOBY2
PATH2
PLTR0
Methodology · How This Works Five AI systems — Grok, ChatGPT, Gemini, Perplexity, and Claude — received identical prompts asking for a Top 5 Infrastructure and Top 5 Speculative AI list, ranked by expected 10-year shareholder return from current valuation. Results were tallied by ticker frequency, then filtered through two lenses: (1) valuation discipline — does the rationale account for current price, not just company quality? and (2) fundamental quality — does the company have real revenue, institutional validation, or a credible path to both? PATH scored 2 on the speculative tally but was excluded because enterprise software automation does not meet the physical-world transformation criterion. Three names — SYM, RXRX, and TEM — achieved unanimous 5/5 consensus across all five systems. JOBY holds its place at 2/5 on TAM optionality despite being the lowest-consensus pick. Gemini introduced probability-weighted outcome distributions for each pick — a framework that forces investor-grade thinking over narrative excitement and will be incorporated in the next iteration.
Claude's Observed Tendencies in This Experiment Across both lists, Claude tended to weight valuation discipline and moat durability more heavily than raw growth rate or narrative momentum. It was the only model to include META (cheapest quality AI compounder on a PEG basis) and FICO (quiet monopoly compounder) in earlier iterations — names that scored zero on the consensus tally but have coherent long-term theses. Claude also flagged NVDA as the name most likely to disappoint via expectation reset rather than business failure, and was more conservative than other models on PLTR's valuation. In the speculative tier, Claude and Perplexity were the only two models to include JOBY — and the only two to weight TAM optionality over near-term revenue certainty. These are tendencies observed in this prompt set, not claims about Claude's permanent analytical character.
Names examined and excluded — with reasons:

PATH (UiPath) · 2 hits · Excluded: Enterprise software automation scores on productivity metrics but does not meet the physical-world transformation criterion. The AI thesis is real but the moat is contested by Microsoft and ServiceNow.

AUR (Aurora Innovation) · 3 hits · Narrowly excluded: ChatGPT, Gemini, and Perplexity all included it. Excluded from the final five in favour of JOBY on the grounds that JOBY has stronger capital backing, a cleaner FAA pathway, and a higher-density urban TAM. AUR's per-mile SaaS model is compelling if commercialisation succeeds — it remains the strongest candidate to replace JOBY if thesis weakens.

PLTR (Palantir) · 0 speculative hits: No model included it in the speculative tier. Four models included it in Tier 2 compounder discussions — at $300B+ market cap, the speculative upside is compressed relative to the names that made this list.

LAZR (Luminar Technologies) · 0 hits · Cautionary note: Filed Chapter 11 bankruptcy December 2025. The clearest reminder that single-product risk, debt load, and customer concentration can erase a compelling autonomous vehicle thesis entirely regardless of technical merit.

MSFT, AMZN, GOOGL · Consensus compounders: All three appeared on multiple Tier 2 lists. Excluded from this tab because the full Tier 1/2/3 portfolio analysis is documented separately — this section focuses on the cross-model speculative experiment specifically.
Five AI Models · Cross-Model Experiment · May 2026

Moonshot Portfolio

Searching for Rocket Lab–style asymmetry, not safe compounders. Real revenue. Controversial narratives. Massive TAMs.

Consensus Winner
ASTS
5 / 5 models · #1 on three lists
Best LEAP Candidate
ASTS / AUR
4 / 5 chose ASTS · Grok alone chose AUR
Claude's Contrarian Outlier
RELY
1 / 5 models · $283M FCF at 1.7× revenue
★ AST SpaceMobile (ASTS) · The Consensus Moonshot · 5/5 Models
Why It Resembles Rocket Lab
Capital-intensive space infrastructure. Real tech, real telecom partners. Tiny revenue today, enormous TAM if execution succeeds.
What Must Go Right
Constellation deploys on schedule. Carrier revenue converts at the guided rate. No catastrophic satellite failures before critical mass.
What Can Kill It
A second satellite loss triggers dilutive capital raise. Starlink moves first. Regulatory friction in emerging markets extends the timeline past the capital runway.
Vehicle & Five-Year Case
Common stock (small) + LEAP calls. Global coverage = software economics on satellite hardware. Marginal cost per subscriber approaches zero.
I  ·  Portfolio
Rocket Lab Asymmetry vs. Value Dislocation
The brief asked for Rocket Lab–style returns. That produces two different answers depending on whether you want execution optionality or market mispricing. Both are valid. They are not the same bet.
Rocket Lab Portfolio
Execution Optionality
These bets pay off when something technically difficult goes right. The upside is binary and large. The downside is real.
ASTS AST SpaceMobile
30%
TEM Tempus AI
25%
NBIS Nebius Group
20%
RXRX Recursion Pharma
15%
IONQ IonQ
10%
Claude's Value-Dislocation Portfolio
Market Mispricing
These companies are already executing. The market is wrong about the category, the narrative, or the valuation. The upside is a re-rate, not a breakthrough.
RELY Remitly Global
40%
ODD ODDITY Tech
25%
IOT Samsara
20%
SOFI SoFi Technologies
10%
BRZE Braze
5%
Sizing Note These are not retirement holdings. Size each position so that a total loss would hurt but not impair the plan. The Rocket Lab Portfolio in particular contains binary-outcome bets where zero is a realistic scenario.
II  ·  Risk Matrix
Risk, Upside, and Vehicle at a Glance
Claude's estimates. Not predictions — frameworks for sizing positions.
Ticker Company 5× Prob 10× Prob Zero Risk Best Catalyst LEAP?
RELY Remitly Global 45% 20% Low Market re-rate + CEO execution Avoid
ODD ODDITY Tech 30% 18% Low Second brand launch proves platform Avoid
TEM Tempus AI 25% 12% Medium Data biz margin expansion Small
IOT Samsara 22% 10% Low International expansion Avoid
BRZE Braze 20% 8% Medium AI-native messaging wins enterprise Yes
SOFI SoFi Technologies 18% 7% Medium Galileo B2B scale Avoid
ASTS ★ AST SpaceMobile 18% 10% High Constellation deployment + revenue ramp Yes — 4/5
RXRX Recursion Pharma 15% 7% High Partnership compound in Phase 2+ Avoid
NBIS Nebius Group 15% 6% High EU AI sovereignty demand Avoid
IONQ IonQ 12% 8% Very High Fault-tolerant QC demonstrated commercially Tiny only
* MXL was not in the original 14-stock brief — added from the CRDO peer analysis. Grade C reflects current RSI >90 after 84% single-quarter move; becomes B on a pullback to $47–53.
Low zero-risk = real FCF, low dilution exposure  ·  High = binary execution dependency, significant dilution history  ·  ★ = 5/5 model consensus
III  ·  Ranked Candidates
10 Candidates by Conviction
Valuation-anchored throughout. Color badges indicate primary thesis type.
1RELY
Remitly Global
Cap ~$2.8B · Rev $1.63B TTM · P/S ~1.7× · FCF $283M
45%
10× 20%
The most glaring valuation anomaly on this list. A company growing at 29% annually with $283M in free cash flow trading at 1.7× revenue — less than 4% of the global remittance market captured. The immigration-fear discount is real; the revenue growth disproves it every quarter. New CEO from Amazon and Apple.
Why This Could Fail
A meaningful expansion of the current U.S. remittance tax would structurally impair unit economics. If Congress moves from 1% to 10%, the entire remittance corridor re-prices. This is the single regulatory risk that could break the thesis completely.
Real FCF Claude only — 1/5 models Common stock Resembles: MercadoLibre 2016
Why the market is wrong Pricing in immigration-driven remittance collapse that hasn't shown up in the revenue. The fear is real and persistent enough to hold the stock down indefinitely — which is precisely the opportunity.
NBN · No fit The remittance story is economic and demographic, not scientific. If the global payments angle interests you, it belongs in an economics channel, not Pop Science.
2TEM
Tempus AI
Cap ~$9.2B · Rev $1.27B (2025) · P/S ~7× · NRR 126%
25%
10× 12%
83% revenue growth in 2025. Proprietary multimodal clinical dataset built over a decade. Q4 EBITDA positive. 2026 guidance: $1.59–1.6B revenue, $65M EBITDA. NRR of 126% — existing pharma customers spend more every year. Perplexity ranked it #1 and chose it as the single-stock pick.
Why This Could Fail
A HIPAA enforcement action or data breach would be catastrophic to the trust the platform is built on. Also: health systems building their own data infrastructure could undercut the moat before monetization fully scales.
AI · Healthcare 3/5 models Common stock Resembles: Veeva 2015
Why the market is wrong High absolute valuation and uncertainty about reimbursement paths keep most generalist investors away. The data moat is invisible until it compounds — and 126% NRR is the compounding signal most are missing.
NBN · Strong fit AI and precision oncology is exactly the kind of mathematically serious, clinically grounded science the Pop Science channel is built for. An oncologist or data scientist who built the diagnostic layer would be a compelling guest.
3IOT
Samsara
Cap ~$15B · Rev ~$1.25B · P/S ~12× · 29% growth
22%
10× 10%
Operating system for physical operations — trucking, construction, food distribution, municipalities. Switching costs are extraordinary: once routes, compliance, and insurance programs are built on Samsara, replacement is a multi-year project. Physical operations ≈ 40% of global GDP, almost entirely uninstrumented.
Why This Could Fail
Cisco, Trimble, or a hyperscaler decides physical operations is worth owning and competes on price. The moat is integration depth, but it can be bought or built by deep-pocketed competitors with patience.
Physical-world AI 2/5 models Common stock Resembles: Datadog 2021
Why the market is wrong Persistently misread as an IoT hardware company rather than a software platform with extraordinary switching costs. The physical-world OS thesis hasn't been priced in.
NBN · Moderate fit Physical operations intelligence is adjacent to complexity science and emergence themes. A conversation about instrumenting the analog world — sensors, logistics, the mathematics of flow — could work.
4ODD
ODDITY Tech
Cap ~$4.2B · Rev ~$700M · Profitable · AI-native
30%
10× 18%
The only profitable, scaling AI-native consumer company on this list. Builds brands using proprietary skin-analysis algorithms trained on 100M+ consumer interactions. The market prices it as a beauty company. It is a consumer AI platform that happens to sell cosmetics. No other model surfaced it.
Why This Could Fail
Beauty trends are fickle and brand fatigue is real. If IL MAKIAGE loses its cultural moment, the underlying AI infrastructure doesn't save you. An Israeli-based company also carries geopolitical headline risk.
Consumer AI Profitable Claude only — 1/5 Resembles: Lululemon 2012
Why the market is wrong The market sees an Israeli beauty brand with geopolitical headline risk. The AI consumer platform underneath — 100M+ interactions, skin-analysis algorithms, multi-brand architecture — is entirely invisible to most analysts.
NBN · No fit Consumer AI is not a natural Pop Science subject. The underlying algorithm is interesting, but the guest would need to be a scientist, not a marketer.
5ASTS ★
AST SpaceMobile
Cap ~$20–27B (volatile) · 2026 guidance $150–200M · $3.5B cash
18%
10× 10%
First space-based cellular broadband connecting directly to standard smartphones. FCC approval secured. $1.2B+ in contracted carrier commitments. $3.5B cash. Universal 5/5 model consensus — and the only name where even Claude, which ranked it #5, didn't exclude it. The SpaceX IPO filing in late May drove a ~17% single-day spike. Entry price matters here more than any other name on this list.
Why This Could Fail
Space hardware is unforgiving. BlueBird 7 was already lost in a launch anomaly. A second major satellite failure before the constellation reaches operational scale could trigger a capital raise that dilutes equity holders severely. The TAM is real; execution is the entire question.
Space · Infrastructure Dilution risk 5/5 consensus ★ LEAP: 4/5 models Resembles: Early Rocket Lab
Why it's controversial Near-universal consensus that it is overvalued relative to current revenue. The 5/5 model consensus is itself a mild contrarian warning. Entry price matters here more than any other name on this list.
NBN · Strong fit Direct-to-device satellite connectivity involves orbital mechanics, RF physics, and deployable structures. The science story is legitimate and has not been well told at the popular level.
6BRZE
Braze
Cap ~$3.5B · Rev ~$600M · P/S ~6× · NRR 115%+
20%
10× 8%
Customer engagement platform at the intersection of AI and customer data. LLMs are making its product dramatically more powerful. NRR above 115%. Re-rated down from 2021 highs, creating a genuine entry point. Primary risk: Salesforce, HubSpot, and Adobe all want this market.
Why This Could Fail
Adobe or Salesforce acquires a close competitor and bundles it into an existing enterprise relationship. Braze's moat is integration depth, but enterprise software markets can be disrupted by bundling strategies that make switching cost irrelevant.
AI · MarTech Claude only LEAP viable Resembles: HubSpot 2018
Why the market is wrong Dismissed as a 'nice-to-have' in enterprise budgets — the first cut in a downturn. The AI-native messaging thesis, which makes the product dramatically more powerful, has not been priced in.
NBN · No fit Marketing automation is not a Pop Science subject regardless of the AI involved.
7IONQ
IonQ
Cap ~$10–21B (volatile) · Rev ~$50–80M · Trapped-ion QC
12%
10× 8%
Most credible pure-play quantum computing company. The lottery ticket on the list. Over $20B in market cap on less than $100M in revenue — you are paying for a future that may not arrive in the five-year window. Included because the asymmetry is real, not because the probability is high.
Why This Could Fail
Quantum computing timelines have slipped repeatedly. Error correction at commercial scale may be a decade away. Cash burn continues and a dilutive capital raise at a lower price than today is a plausible scenario before any commercial milestone is reached.
Quantum · Deep tech Very high zero risk 2/5 models Tiny LEAP only
Why it's controversial $20B+ market cap on $80M in revenue. The market is pricing a technology that may not commercially arrive in this decade. This is a bet on timing as much as on the technology.
NBN · Strong fit Quantum computing is the highest-CY subject on this list. Trapped-ion architecture, error correction, the relationship between quantum and classical computation — a natural interview. Several strong books on QC have appeared in the last three years.
8RXRX
Recursion Pharmaceuticals
Cap ~$2B · Rev ~$50–80M · Roche + Bayer partnerships
15%
10× 7%
Using AI and high-throughput biology to industrialize drug discovery. Not trying to develop one drug — trying to rebuild how drugs are found. Proprietary dataset of billions of cellular images. Platform thesis: the output is a pipeline engine the entire pharma industry needs.
Why This Could Fail
Biology is hard and AI doesn't change the fundamental failure rate of drug candidates. If the first wave of partnership molecules fail in Phase 2 trials, the platform thesis collapses faster than the science would suggest. Cash runway is finite.
AI · Drug discovery 2/5 models Common stock Resembles: Early Illumina
Why the market is wrong Three years of missed drug discovery timelines have exhausted investor patience. The platform thesis requires faith that biology can be predicted at scale — and that patience is now at a low.
NBN · Strong fit AI drug discovery at the biology-computation interface is exactly the interdisciplinary science the channel is designed for. What does it mean to industrialize the search through chemical space?
9SOFI
SoFi Technologies
Cap ~$8B · Rev ~$2.6B · P/S ~3× · Banking charter
18%
10× 7%
Only digital financial services company building banking, lending, and investing simultaneously. Banking charter gives a cost-of-funds advantage neo-banks can't match. Consistently misread as a lender when the real asset is a growing deposit base and cross-sell engine.
Why This Could Fail
Banking is slow and regulated. Credit losses in a recession, student loan policy shifts, or a regulatory action on the banking charter could each independently impair the thesis. Being a bank while the market prices you as a tech company is a permanent tension.
Fintech · Banking 2/5 models Common stock Resembles: Square 2018
Why the market is wrong Persistently misclassified as a lender rather than a banking platform. Rate sensitivity fears and student loan policy noise override the membership flywheel story every quarter.
NBN · No fit Fintech and digital banking do not map naturally to the Pop Science channel.
10NBIS
Nebius Group
Cap ~$8–15B · Rev ~$250M (rapid growth) · EU-native AI cloud
15%
10× 6%
European AI infrastructure — GPU cloud, AI tooling, data services — spun out of former Yandex. Non-American GPU cloud at a moment when European enterprise AI demand is growing and sovereignty concerns push customers away from AWS/Azure/GCP. Trades at a fraction of CoreWeave's valuation for similar positioning.
Why This Could Fail
GPU cloud is a commodity unless you layer software on top. CoreWeave, AWS, and Azure will compete on price. The Yandex heritage creates reputational friction in some markets. Geopolitical headline risk is permanent and unpredictable.
AI · Infrastructure 3/5 models Common stock Resembles: Equinix 2010
Why it's controversial The Yandex heritage creates a reflexive avoidance reaction in many institutional investors. European AI sovereignty is real but underappreciated as a commercial driver.
NBN · Weak fit AI infrastructure is a technology story, not a science story. Unless the conversation is about the physics of computation or thermodynamics of data centers, this belongs in a business channel.
IV  ·  Model Comparison
Methodology Differences
The prompt was identical. The interpretations — and the willingness to violate or honor the brief — were not.
Claude (Anthropic)
Most financially rigorous. Led with P/S ratios and FCF throughout. Only model to pick a profitable, FCF-positive company (RELY) as #1. Excluded pre-revenue companies by brief criteria. Pushed back on IONQ's valuation explicitly.
Unique: RELY, BRZE, ODD  ·  1 stock: RELY  ·  LEAP: ASTS
Grok (xAI)
Most concise. Telegraph-style delivery. Closest to the Rocket Lab analogy — concentrated on physical-world infrastructure (space, trucking, eVTOL, lunar). Only model to run AUR at #2. No valuation engagement at all.
Unique: AUR, ACHR, LUNR  ·  1 stock: ASTS  ·  LEAP: AUR
ChatGPT (OpenAI)
Most data-heavy, least disciplined on the brief. Included Palantir at $300B+ — directly violating "avoid mega-caps." Cited ASTS at $36–44B market cap during a spike-day. The only model to include CRWV, SOFI, DUOL, and ZETA.
Unique: PLTR, CRWV, DUOL, ZETA  ·  1 stock: ASTS  ·  LEAP: ASTS
Gemini (Google)
Most space-heavy and most idiosyncratic. Five of ten picks are aerospace. Only model to include Redwire (RDW, $650M micro-cap), Origin Materials, Desktop Metal (in restructuring), STEM Inc., and QuantumScape. Richest list for discovery value.
Unique: RDW, ORGN, DM, STEM, QS  ·  1 stock: ASTS  ·  LEAP: JOBY
Perplexity
Most brief-compliant. Only model to explicitly flag Oklo for missing the "real revenue" filter. Most heterodox picks: VSEC (aviation aftermarket), ETON (specialty pharma), ICHR (semiconductor equipment). Only model besides Claude not to choose ASTS as single-stock pick.
Unique: VSEC, ETON, ICHR  ·  1 stock: TEM  ·  LEAP: ASTS
Ticker Company Hits Cl · Gr · GP · Ge · Px Signal
ASTS ★AST SpaceMobile5/5
Perfect consensus. The only name no model excluded.
TEMTempus AI3/5
Claude, Grok, Perplexity. Perplexity's #1 and single-stock pick.
SYMSymbotic3/5
Grok, ChatGPT, Perplexity. $22.7B backlog. Not in Claude's list.
JOBYJoby Aviation3/5
Grok, Gemini, Perplexity. Claude excluded on pre-revenue grounds.
NBISNebius Group3/5
Claude, Grok, ChatGPT. European AI cloud.
AURAurora Innovation2/5
Grok + Perplexity. Perplexity also named it most likely to fail. That tension is honest.
RELYRemitly Global1/5
Claude only. The value anomaly. No other model surfaced a $283M FCF company at 1.7× revenue.
PLTRPalantir1/5
ChatGPT only. $300B+ cap — the brief asked to avoid mega-caps.
Cl = Claude · Gr = Grok · GP = ChatGPT · Ge = Gemini · Px = Perplexity  ·  Navy = included   Gold = included
V  ·  Conviction Questions
Where They Agreed. Where They Didn't.
One Stock · Maximum 5-Year Upside
ClaudeRELY
GrokASTS
ChatGPTASTS
GeminiASTS
PerplexityTEM
ASTS wins 3/5. Claude chose value, Perplexity chose proven platform. The two most financially rigorous models broke from consensus.
One LEAP · Maximum Leverage
ClaudeASTS
GrokAUR
ChatGPTASTS
GeminiJOBY
PerplexityASTS
4/5 chose ASTS LEAPs. Grok (AUR) and Gemini (JOBY) argued binary execution milestones make those better LEAP candidates than a TAM re-rating story.
Highest Probability Life-Changing Return
ClaudeRELY
GrokASTS
ChatGPTPLTR
GeminiJOBY
PerplexityASTS
No consensus. ChatGPT's PLTR answer is the most defensible and the least exciting. Claude's RELY answer is the most contrarian.
Most Likely to Fail Completely
ClaudeIONQ
GrokLUNR
ChatGPTIONQ / RGTI
GeminiRGTI
PerplexityAUR
Quantum gets 3/5 failure votes. Perplexity named AUR — which it also included at #5. The most honest tension on any list.
Not investment advice. This is a cross-model intellectual exercise — five language models, one brief, one comparison. None of these outputs constitute a recommendation to buy or sell any security. All figures approximate as of May 30, 2026. Do your own diligence.
Five AI Models · Cross-Model Synthesis · May 2026

Put Lab

Assignment is a feature, not a bug. Four AI models. One brief. Find the 15 stocks worth selling weekly puts against — and separate the great put candidates from the moonshots that will hurt you if assigned.

Consensus Grade A
IOT · ODD · BRZE
4 / 4 models · unanimous
Universal Avoid
SPCE · RKLB
F / D across all 4 models
Critical Data Error
RELY
Gemini stale data · Perplexity unavailable · ChatGPT corrected → #1
⚠ Data Error — Two Models Got RELY Wrong Three of five models initially identified RELY as Reliance Global Group — a broken $15M micro-cap with a going concern. The correct RELY is Remitly Global: $1.63B revenue, $283M FCF, 1.7× P/S. Claude and Grok identified Remitly correctly and ranked it #1. ChatGPT was corrected via a follow-up prompt and also ranked it #1. Gemini ranked it F (wrong company). Perplexity was unavailable. All three correct evaluations agree: RELY is the top put-selling candidate in this universe.
14 Stocks — Best to Worst for Put Selling
Synthesizing Claude, Grok, ChatGPT, Gemini, and Perplexity. Assignment quality, valuation, and downside protection weighted above premium income.
Rank Grade Ticker Company Key Metrics 5× Prob Impairment 5-Model Agreement
1
A
RELY Remitly Global
Correctly identified by 3/5 models · Ranked #1 by all three
$2.8B cap · $1.63B rev · 1.7× P/S · $283M FCF 45% 5%
2
A
IOT Samsara
Universal A across all 5 models
$15B cap · $1.3B rev · 115%+ NRR · FCF positive 22% 5%
3
A
ODD ODDITY Tech
Profitable AI platform, 3–4× revenue
$4.2B cap · $700M+ rev · Profitable · No debt 30% 10%
4
A
BRZE Braze
SaaS with NRR 110%, improving FCF
$3.5B cap · $840M rev · 30% growth · Net cash 22% 6%
5
B
RDDT Reddit
$311M FCF, 91.5% gross margins, AI data moat
$28B cap · $2.5B rev · 69% growth · $2.77B cash 20% 5%
6
B
TEM Tempus AI
Clinical data moat, EBITDA turning positive
$9B cap · $1.6B guided · 36% growth · NRR 126% 25% 15%
7
B
NBIS Nebius Group
684% revenue growth, but $57B market cap
$57B cap · $1.6B ann. rev · European AI cloud 15% 20%
8
C
RXRX Recursion Pharmaceuticals
AI drug discovery — real platform, long timeline
$2B cap · $66M rev · Roche + Bayer · heavy burn 20% 35%
9
C
LUNR Intuitive Machines
Real contracts, 16% gross margins, mission binary risk
$7B cap · $334M rev · $1.1B backlog · NASA 15% 25%
10
C
ASTS AST SpaceMobile
Blue Origin launch pad destroyed — timeline pushed to 2027
$25B cap · $60M ann. rev · $3.5B cash · 400×+ EV/S 18% 30%
11
D
AUR Aurora Innovation
Pre-revenue autonomous trucking at $14B
$14B cap · ~$15M rev · $150M+/qtr burn 20% 40%
12
D
RKLB Rocket Lab
Great business — wrong price for put selling
$83B cap · $800M ann. rev · ~100× sales · Fully rerated 10% 12%
13
F
POET POET Technologies
$1.4M revenue — not a put-selling vehicle
$2.1B cap · $1.4M TTM rev · Pre-commercial 20% 45%
15*
C
MXL MaxLinear
CRDO pre-rerating analog · Added from separate analysis
$8.3B cap · $549M ann. rev · +136% infra YoY · 17× P/S 25% 10%
14
F
SPCE Virgin Galactic
Going concern disclosure — universal F
$272M cap · $1.5M rev · -$438M FCF · Dilution machine 5% 55%
Gold = strong model agreement  ·  Navy = general agreement  ·  Order: Claude · Grok · ChatGPT · Gemini · Perplexity
What Would You Actually Own?
For each name: thesis, assignment comfort, the strongest argument against.
A
Sell
RELY · Remitly Global
Value Anomaly — The Consensus Miss
$2.8B cap · $1.63B revenue · 1.7× P/S · $283M free cash flow
45%
Impairment 5%
A company doing $1.63B in revenue with $283M in free cash flow trading at 1.7× revenue. Holds under 4% of the global remittance market. The immigration-fear discount is real; the revenue growth disproves it every quarter. New CEO from Amazon and Apple. Profitable at a scale most SaaS companies five times its valuation can't match.
The case againstA meaningful expansion of the U.S. remittance tax from 1% to 10%+ would structurally impair unit economics. Political risk is real and unpredictable. Also: two of five AI models didn't know which company this was — confirming the market hasn't discovered it either.
Profitable · FCFClaude #1 · Grok #1Common stock if assignedMercadoLibre 2016 analog
A
Sell
IOT · Samsara
The Unanimous Choice
$15B cap · $1.3B revenue · 29% growth · NRR 115%+
22%
Impairment 5%
Physical operations OS for trucking, construction, municipalities. Extraordinary switching costs — once fleet compliance and insurance workflows are built on Samsara, replacement is a multi-year project. FCF positive. No binary risk. Physical operations represent ~40% of global GDP and are almost entirely uninstrumented. Every model gave this an A.
The case against12–14× forward revenue is not cheap. A deep-pocketed competitor (Cisco, Trimble, or a hyperscaler) deciding this market is worth owning could compress growth. And at 29% growth, if it decelerates to 20%, the multiple compresses hard.
FCF Positive5/5 Model Consensus ABest 10-year holdDatadog 2021 analog
A
Sell
ODD · ODDITY Tech
Profitable AI Platform, Mislabeled as Beauty
$4.2B cap · $700M+ revenue · Profitable · No debt · 3–4× P/S
30%
Impairment 10%
The only profitable, scaling AI-native consumer platform in this universe. Skin-analysis algorithms trained on 100M+ interactions. The market prices it as a beauty brand. The real asset is a consumer AI platform that happens to sell cosmetics. Pre-rerating. No other model in the Moonshots exercise surfaced it. That's the point.
The case againstBeauty trends are fickle. If IL MAKIAGE loses its cultural moment, the AI infrastructure doesn't save you. Israeli-based company carries geopolitical headline risk. Some models flagged accounting/legal optics as a concern worth monitoring.
Profitable5/5 Model Consensus APre-reratingLululemon 2012 analog
A
Sell
BRZE · Braze
The Defensive SaaS Candidate
$3.5B cap · $840M revenue · NRR 110% · Net cash positive
22%
Impairment 6%
Customer engagement platform at the AI/customer-data intersection. NRR rising (109% → 110%). $1.08B RPO. Fourth consecutive quarter of organic acceleration. Not exciting — which is exactly why it's a good put-selling candidate. Clean balance sheet, improving FCF, 6–7× forward revenue.
The case againstAdobe, Salesforce, and HubSpot all want this market with significantly more resources. Gross margins declining (69.3% → 67.4%). If a competitor bundles a comparable product, Braze's switching cost argument weakens faster than the NRR implies.
5/5 Model Consensus AImproving FCFHubSpot 2018 analog
B
Select
RDDT · Reddit
Best Business, Valuation Debate
$28B cap · $2.5B revenue · 69% growth · 91.5% gross margin · $2.77B cash
20%
Impairment 5%
$311M free cash flow. 91.5% gross margins. $2.77B cash with 0.2% capex as percentage of revenue. One-of-one internet community platform. AI data licensing is a durable, barely-started revenue layer. The business is exceptional. The debate is entirely about whether $28B already reflects it.
The case againstThe stock has rerated hard. At 11× forward revenue on a platform where AI search threatens to reduce the referral discovery that drives user acquisition, the multiple leaves little margin for error. Sell puts at meaningful discounts to current price — not aggressively at current levels.
Profitable · FCFAI data moatMid-rerating
B
Select
TEM · Tempus AI
Clinical Data Moat, GAAP Losses Widening
$9B cap · $1.6B guided · 36% growth · NRR 126% · $644M cash
25%
Impairment 15%
Decade-built clinical genomic dataset. NRR of 126% means existing pharma customers spend more every year. Data licensing growing 44% with improving margins. EBITDA turning positive. The data moat is the kind that took a decade to build and cannot be replicated cheaply. Adjusted EBITDA positive in 2026.
The case againstGAAP net loss widened 85% to $125.9M in Q1 even as revenue grew 36%. Operating leverage has not appeared yet. High SBC dilution. If assigned at today's price and GAAP losses continue to widen for two more years, the position is uncomfortable. Sell puts at conservative strikes — not aggressively.
NRR 126%Mid-reratingGAAP losses wideningVeeva 2015 analog
B
Select
NBIS · Nebius Group
684% Revenue Growth, Valuation Already Reflects It
$57B cap · $1.6B ann. revenue · European AI cloud · Yandex heritage
15%
Impairment 20%
Revenue growth is real — 841% YoY in AI cloud. European AI sovereignty is a genuine commercial driver. Nvidia partnership validates the infrastructure. The problem is that $57B market cap at 35× forward revenue already prices in near-perfect execution for years. GPU cloud is a commodity business without a software layer.
The case againstAWS, Azure, and CoreWeave will compete on price. The Yandex heritage creates permanent reputational friction in some markets. A 35× revenue multiple on infrastructure that could commoditize gives very little margin for error. Sell puts only at deeply conservative strikes — the downside from current valuation is asymmetrically large.
35× forward sales3/5 models BGeopolitical risk
C
Conservative
RXRX · Recursion Pharmaceuticals
AI Drug Discovery — Long Validation Ahead
$2B cap · $66M revenue · Roche + Bayer · heavy burn
20%
Impairment 35%
AI drug discovery at the biology-computation interface. Proprietary dataset of billions of cellular images. Roche and Bayer partnerships validate the platform. If the platform works, the output isn't one drug — it's a pipeline engine. Gemini's pick as "most like Rocket Lab before the market believed."
The case againstBiology is hard and AI doesn't change the fundamental failure rate of drug candidates. $559M in trailing net losses on $66M in revenue. If assigned, you own a company burning ~$140M/year with no approved drugs and a validation event (Phase 2 success) that could be 3–5 years away.
Deep OTM only$559M trailing lossesEarly Illumina analog
D
Avoid
RKLB · Rocket Lab
Great Company — Wrong Price for Put Selling
$83B cap · $800M ann. revenue · ~100× sales · Fully rerated
10%
Impairment 12%
The benchmark and the cautionary tale simultaneously. A great space company that has fully rerated — $83B on $800M in annualized revenue, trading 27% above analyst consensus. The business is real. The valuation makes put-selling uncomfortable. If assigned you own a fantastic company at a price the fundamentals won't catch up to for years.
The one real model disagreementChatGPT gives RKLB a B — would selectively sell puts — citing the business quality. Claude and Grok give it D, citing the valuation: 100× sales makes assignment at current prices difficult to defend in a strategy where assignment is a feature. Both are correct about the company. The disagreement is about whether 100× revenue is a price you'd want to pay involuntarily through assignment. It isn't.
100× revenue5/5 Model DWait for pullback
F
Never
SPCE · Virgin Galactic
Going Concern — Universal F
$272M cap · $1.5M TTM revenue · -$438M FCF · Going concern disclosure
5%
Impairment 55%
The going concern disclosure ends the analysis. The company relies on ATM capacity, warrant structures, and expected customer payments at commercial launch to survive. FY2025 FCF was -$438M. Revenue is $1.5M. Every model gave this an F. No further analysis needed.
The unanimous verdictDo not sell puts. If assigned, you own equity in a company that may require continuous dilutive capital raises before generating meaningful revenue. The going concern disclosure is the answer to every question about whether to participate.
Going concern5/5 Model F-$438M FCF
MaxLinear (MXL) — AI Connectivity Inflection
Not in the original 14-stock brief. Added because of the CRDO peer analysis earlier in this session — this is the name five models identified as the closest to where CRDO was 12 months ago.
C
OTM Only
MXL · MaxLinear
AI Connectivity — CRDO Before the Rerating
~$8.3B cap · ~$549M annualized revenue · 43% growth · Infrastructure segment +136% YoY · 17× P/S vs CRDO's 38×
25%
Impairment 10%
MaxLinear makes PAM4 DSPs — high-speed signal processors that sit inside the optical transceivers connecting AI cluster racks at 800G and 1.6T speeds. Infrastructure segment grew 136% YoY in Q1 2026, becoming the largest revenue segment. The market still prices MXL as a cyclical broadband recovery story rather than an AI structural winner. Q2 guidance: $160–170M revenue with 56–61% gross margins — the mix shift to high-margin AI optical revenue is visible in the numbers. CRDO made an identical journey from "niche SerDes designer" to "AI connectivity platform" — and went from a $3B to a $43B market cap in 18 months. MXL has run 7× off its lows but trades at 17× P/S vs CRDO's 38×. The re-rating has started. It is not finished.
Why C and not B RSI above 90 after an 84% single-quarter move. The stock has run too hard for aggressive weekly put-selling at current levels — if assigned after a gap-down on a semi cycle concern, you own a company mid-transition with legacy broadband drag. Broadcom and Marvell own the high-speed DSP market and can compete aggressively. The right vehicle here is common stock or LEAPs on a pullback — not weekly puts at 20-30 delta. A pullback to $47–53 would change this to a B.
5/5 Models: CRDO Analog Mid-run · Pre-full rerating Keystone 5nm/4nm PAM4 DSP RSI >90 — wait for pullback Common stock preferred
The CRDO Parallel CRDO was "a niche SerDes chip designer" until Active Electrical Cable revenue made the label irrelevant. MXL is "a broken broadband recovery play" — until the Keystone PAM4 DSP ramp with a tier-1 hyperscaler makes that label irrelevant. The tell: when the financial media stops calling it a broadband recovery and starts calculating its AI pure-play revenue run-rate, the multiple adjusts from 17× toward 38×. That re-rating is the entire trade. Q2 2026 margin guidance (56–61%) is the first piece of evidence the mix shift is real.
Claude · Grok · ChatGPT · Gemini — Complete Synthesis
Perplexity was unavailable for this round. All four completed models correctly identified RELY as Remitly Global. Gemini's data reflects an earlier snapshot — the MXL valuation discrepancy is flagged below.
Consensus Top Five · All Four Models Agree
#1
RELY
Grade A
#2
IOT
Grade A-
#3
BRZE
Grade A-
#4
ODD
Grade B+
#5
TEM
Grade B+
"Put-selling rarely rewards excitement." — ChatGPT's framing on why the moonshots (ASTS, AUR, RXRX, POET) belong in your common stock allocation, not your weekly put rotation.
Claude · Anthropic
1. RELY  ·  2. IOT  ·  3. ODD
4. BRZE  ·  5. RDDT
Led with valuation and FCF. Most skeptical on RKLB (D). MXL as "next CRDO before belief." Only model to put RDDT in top 5.
Grok · xAI
1. RELY  ·  2. MXL  ·  3. TEM
4. IOT  ·  5. BRZE
Most bullish on MXL (A, #2). RKLB F — most negative of all models. RELY conviction identical to Claude. Best on hardware infrastructure framing.
ChatGPT · OpenAI
1. RELY  ·  2. IOT  ·  3. BRZE
4. TEM  ·  5. ODD
Most forgiving on RKLB (B-). Clearest moonshot/put distinction. Best single line: "Put-selling rarely rewards excitement." MXL ranked B.
Gemini · Google (stale data — see note)
1. RELY  ·  2. MXL  ·  3. TEM
4. IOT  ·  5. BRZE
Used pre-rally data: MXL at $1.79B (now $8.3B+), RKLB at $6.4B (now $83B+). Grades are valid for those prices but not today's. IOT flagged as most overvalued. AUR as "next RKLB before belief."
⚠ Gemini Data Currency Note Gemini's MXL grade (A, #2) reflects a $1.79B market cap and 3.1× sales — prices from before MXL's 84% single-quarter rally. At today's $8.3B market cap (17× sales), Claude grades it C. Both analyses are correct for their respective entry points. The thesis is identical; the price changed. Gemini's RKLB ($6.4B) and ASTS ($5.2B) share the same issue — both reflect pre–SpaceX IPO rally data. Where Gemini's grades seem surprisingly optimistic on space names, this is the explanation.
Where They Agreed. Where They Diverged.
Perplexity unavailable. Gemini data caveated for stale pricing on space and MXL names.
One Put to Sell This Week · Unanimous 4/4
RELY
The only unanimous answer across all nine questions. Q1: net income up 332%, EBITDA $101.6M up 74%, guidance raised to $1.96B. Every model, independently: RELY.
Best Risk-Adjusted Return Today
RELY / MXL
Claude: ODD · Grok: RELY or MXL · ChatGPT: RELY · Gemini: MXL (stale price). RELY and MXL split the vote. At current prices, RELY is the cleaner answer — no binary risk. MXL's endorsement reflects the thesis correctly; Claude's C grade reflects today's price after the 84% rally.
Highest Probability of Next Rocket Lab
ASTS (3/4)
Claude, Grok, ChatGPT: ASTS · Gemini: LUNR. ASTS wins 3/4. Gemini's LUNR answer reflects the NSNS anchor contract thesis — same "institutional infrastructure" pattern as early RKLB. Neither is a good put candidate. Both are right as common stock moonshots.
Most Overvalued Relative to Fundamentals
RKLB / ASTS
Claude: ASTS · Grok: RKLB · ChatGPT: RKLB · Gemini: IOT (stale data). ASTS vs. RKLB splits evenly — both are correct. ASTS at 400× sales after launch pad destruction. RKLB at 100× sales above analyst consensus. Gemini's IOT answer reflects a period when space names were cheaper.
Best Combo: Valuation + Growth + Quality + Optionality
IOT (2/4)
Claude: IOT · Grok: RELY or MXL · ChatGPT: IOT · Gemini: TEM. IOT wins 2/4 with strong seconds. Gemini's TEM answer credits the data optionality layer. Grok's MXL answer is the AI hardware thesis. IOT remains the most balanced answer across all four criteria simultaneously.
Highest Probability Permanent Impairment · Unanimous 4/4
SPCE
The second unanimous answer. Going concern disclosure. -$438M FCF. $1.5M in annual revenue. Do not sell puts. Do not want assignment. This answer required no deliberation from any model.
Most Confident 10-Year Hold If Assigned
IOT (3/4)
Claude: IOT · Grok: RELY or IOT · ChatGPT: IOT · Gemini: IOT. IOT wins 3/4. Once embedded in compliance, insurance, and routing workflows, Samsara switching is a multi-year project. That moat compounds for a decade regardless of what the multiple does in the short term.
Most Like RKLB Before the Market Believed
ASTS / MXL / AUR
Claude: MXL · Grok: ASTS · ChatGPT: ASTS · Gemini: AUR. Three different answers, all defensible. ASTS: binary setup, enormous TAM, market still debating. MXL: business structurally mislabeled. AUR: quiet validation phase, deep skepticism. None are good put candidates — all are worth owning directly in small size.
Most Like RKLB After the Market Recognized the Story
RKLB (3/4)
Claude: RKLB · Grok: RKLB or NBIS · ChatGPT: RKLB · Gemini: ASTS. RKLB wins 3/4. The debate has shifted from "will this work" to "how much future success is already priced in" — that's post-recognition. Great company. D grade for put-selling at 100× sales.
Four Buckets — Not Always the Same Stock
Great Businesses
RDDT
IOT
BRZE
ODD
RKLB
Great Stocks Today
RELY
ODD
BRZE
IOT
Great Put-Selling Candidates
RELY · IOT · ODD · BRZE
Selectively: RDDT · TEM
5x–10x Moonshots
ASTS · AUR · RXRX · POET
Buy stock — not puts
Not investment advice. This is a cross-model analytical exercise — five language models, one brief, one synthesis. None of this constitutes a recommendation to buy, sell, or write options on any security. All figures approximate as of May 30, 2026. Options involve risk of loss including the full amount secured. Do your own diligence and consult a licensed advisor for your specific circumstances.
Cash-Secured Puts · Decision Cockpit
Puts for Assignment
Gate on quality, rank on fear, remember every trade. A name earns the right to be ranked only if you'd be happy to own it — then the discount, not the hype, sets the order. Assignment price is the true entry price.
Step 1 — The Gate
Quality is a hurdle, not a score.

Passes only if Changed? = No/Resolved and Assignment ≥ A. ROIC, growth and management earn admission — nothing more.

Step 2 — The Rank
Contrarian = Valuation + Range.

Range is measured: (price − 52w low) ÷ (high − low) → ≤20% = 5, ≤70% = 3, else 1. Yield is shown, never folded into the rank.

Account size (optional — flags any single put over 50%):

Cleared the Gate

Rankable
Sorted by contrarian score (Valuation + Distance-from-low) among gate-passing names only.

Resolve Thesis First

Gated out · open question

Below the Gate

Excluded

Postmortem Ledger

The learning system
DateTickerStrikePremBasisRisk*Buy?*StatusLesson
*Risk and Buy? are snapshotted from the scorecard the moment you log — that's what lets the panel below find what you misprice.

How the cage scores

  1. Business Changed?No: price moved, thesis intact. Resolved: the existential question got answered favorably (Google's AI-search worry). Key Question: unsettled — form a view before selling.
  2. The gate — pass requires Changed ≠ Key Question and Assignment grade A or better. Failures drop out of the ranking entirely.
  3. Valuation (1–5) — your conviction on the discount, a thesis you set. Distance-from-low is measured from the real 52-week range. Contrarian score = the two added; only this orders survivors, which is why a quality name near its low outranks the same quality near its high.
  4. Assignment price, sizing, earnings — strike − premium is your true entry; collateral = strike × 100; an earnings date inside 7 days flags red, because a weekly put shouldn't straddle a binary event unintentionally.
  5. The ledger — every put becomes a row carrying its risk and buy? tags, so the patterns panel can tell you whether losses cluster in High-risk names or in puts you wouldn't have bought outright. That's the edge.
MELI footnote. ~$169K collateral per contract — best thesis, worst vehicle. If the credit-book thesis turns bullish, express it with shares or a defined-risk put spread, not a naked weekly CSP.  ·  Estimated 52-week highs are marked ~est — replace them with real numbers, as they move the ranking. The ledger saves to this browser and persists on the hosted site; use Copy all data to back up.  ·  Not investment advice. Prices an approximate Jun 1, 2026 snapshot; verify the chain before trading.