Gavin Baker's double shift: racks in orbit, a $1.75T SpaceX filing, and valuations that cannot all be true
Gavin Baker worked both sides of the street this week — solo on Invest Like the Best, then in the guest chair on All-In — pushing one thesis twice: SpaceX puts compute in orbit by 2028-2030, while AI's cross-sectional valuations "cannot all be true." Around him, All-In charted a $1.75 trillion SpaceX S-1 and a public souring on AI.
May 18 – May 24, 20262 episodes2 shows94 claims on the record22 new predictions
01
Orbital compute gets a clock — and a ticker
Gavin Baker booked both shows this week, and orbital compute went from reframing to timeline. On Invest Like the Best he stripped the idea of its sci-fi caricature — Blackwell-sized racks in sun-synchronous orbit, cooled by radiators, linked by lasers — noting SpaceX already operates 98-99% of all satellites in orbit. Two days later on All-In he gave a point prediction: second half of 2028 to first half of 2030. The commercial frame landed the same week: SpaceX's S-1 targets a $75 billion raise at a $1.75 trillion valuation, with Starlink at $11.4 billion in revenue, and Chamath Palihapitiya underwriting terrestrial data centers alone at $100-200 billion of revenue by 2030-2032 — with Tesla, he insists, eventually merged in.
“I would like to redefine orbital compute has racks in space, not giant, floating pentagon-sized data centers in space. That's silly. What makes a data center is you're connecting these racks with lasers.”
Baker's valuation paradox: real cash flows, impossible cross-section
Baker's other repeated thesis: AI's cash flows are real and its cross-section is not. On Invest Like the Best he argued TSMC's wafer scarcity is the only thing standing between this buildout and a bubble — supplied everything Jensen wanted, Nvidia could sell $2-3 trillion of GPUs — while SemiCap trades near 40x next-quarter annualized earnings and DRAM at mid-single digits. On All-In, after Nvidia's $81.6 billion quarter met a compressed multiple, he restated it as cross-sectional inefficiency. The revenue side he considers settled: OpenAI and Anthropic at roughly $100 billion of combined ARR with ~80% inference gross margins, GPUs at full utilization, funded from operating cash flow — unlike 2000's 99%-unutilized fiber. His year-end call: $200-400 billion of LLM ARR.
“the valuations do not make sense. They just flat out do not make sense. They cannot all be true.”
America turns on AI, and the diagnosis splits three ways
All-In's longest segment tracked the public turning against AI: booed commencement speakers, a presidential AI executive order pulled at the last minute, Cloudflare cutting more than 20% of staff despite record growth. The hosts split on why. David Friedberg reads it as structural — technology concentrates leverage in a small group, and AI epitomizes the power imbalance — yet argues slowing down is impossible once proliferation begins, invoking the post-Manhattan-Project arms race. Chamath Palihapitiya blames the messengers, savaging Matthew Prince's Cloudflare layoff memo as emblematic of how badly public tech CEOs message layoffs. Gavin Baker adds a caution against the regulatory reflex: power granted to government is almost never returned. Three diagnoses, no shared prescription — beyond Chamath's demand that executives stop writing missives.
“I think that there's like an underlying view that technology creates leverage for a small group of people which creates power imbalances”
“shut the f*** up, get behind the keyboard, just do your job, and if you need to manage something, just manage it, but don't write these missives. You're terrible at it, all of you.”
The week's quietest big story: Andrej Karpathy, 39, joined Anthropic to lead a new pre-training team focused on recursive self-improvement. On All-In, Gavin Baker named recursive self-improvement and continual learning the two final frontiers for AI — combined, they might pull the future forward. He had made the same case solo two days earlier: continual learning, a model updating its own weights in real time, is the pivotal open question, and if solved it produces a fast takeoff that insiders seem to think is close. Chamath Palihapitiya drew the compounding curve — an order of magnitude of model improvement per year, a new Moore's law — while insisting that breathless model-release commentary is wasted energy next to the end-user cases now becoming possible.
“those two combined, I think, they might pull the future forward in a very real way”
Anthropic added $11 billion of ARR in a single month, per Gavin Baker — roughly the revenue Palantir, Snowflake, and Databricks each took a decade to build, combined. ILTB →
Baker says Claude Opus now generates roughly 70% fewer tokens for the same question — intelligence effectively rationed while Anthropic stays compute-constrained. ILTB →
Cursor's Composer 2.5 went Pareto-dominant after roughly three weeks of RL on Colossus — proof, Baker argues, that proprietary coding data is decisive; Jason Calacanis thinks Cursor could be the dominant coding player within a year. ALL-IN →
A survey of professional forecasters projects US CPI hitting 6% in Q2; David Friedberg's 'Dr. Doom' monologue puts global debt-to-GDP at 310% and warns of credit-crisis risk. ALL-IN →
Baker predicts Elon Musk's Terrafab joint venture with Intel will succeed in building the world's largest fab in America — and that the Watts shortage starts easing in 2027-2028. ILTB →
Gavin Baker on Flock-style AI surveillance: 'Crime is now a choice' — municipalities now pick their crime rate and voters sort it out. ALL-IN →
The record
Every episode this week
The issue above is written from these structured records — nothing else. Each card is the full breakdown: claims, predictions, quotes, all timestamped.
Tesla will be merged into the combined Elon entity (with SpaceX), creating a single corpus of physical, learning, infrastructure, and connectivity capability.
Chamath Palihapitiya · ALL-IN · next 1-2 years · source 1:02:18
SpaceX's AI/compute revenue will be roughly $40-45B next year and then likely double the year after.
Chamath Palihapitiya · ALL-IN · by 2030-2032 · source 1:02:18
SpaceX's terrestrial data-center business alone will reach $100-200B of revenue by 2030-2032.
First mentions
New to the graph this week
Subjects no indexed episode had discussed before this week.
The week's episodes are diarized into structured records; one editorial pass writes this issue from those records when the week's ingest lands. A grounding gate greps every quote above against its episode record — a quote that isn't verbatim fails the build.
the data behind this issue
# this issue, as structured data — stories, citations, receipts
curl 'https://aipod-two.vercel.app/digest/2026-w21.json'# the records it was written from
curl 'https://aipod-two.vercel.app/api/episode/<id>'# any id in episode_ids