2026-05-21 08:00:00
After 24 years as a private company, SpaceX filed its S-1 yesterday. The filing reveals an AI-era conglomerate. SpaceX has three distinct segments : Space, Starlink, and AI.
In 2025, SpaceX generated $18.7 billion in consolidated revenue with $6.6 billion in Adjusted EBITDA. But the real story lies beneath those top-line numbers.
SpaceX runs three businesses with fundamentally different economics :
| Segment | 2025 Revenue | Operating Margin | CapEx |
|---|---|---|---|
| Starlink | $11.4B | +39% | $4.2B |
| Space | $4.1B | -16% | $3.8B |
| AI | $3.2B | -199% | $12.7B |
Starlink dominates at 61% of total revenue. Space launches contribute 22%. The AI segment (X platform advertising, subscriptions, and xAI compute infrastructure, consolidated through the xAI merger completed in February 2026) adds 17%.
Space enables Starlink (and eventually orbital AI). Starlink funds the enterprise. AI consumes capital today with the promise of future leverage.
Starlink is SpaceX’s largest business & profitable.
Starlink delivered $4.4 billion in operating income in 2025 ; a 39% operating margin. Meanwhile, the Space segment ran a $657M operating loss (mostly Starship R&D), and AI burned $6.4B building COLOSSUS data centers and training Grok.
The S-1 provides historical subscriber data showing rapid growth :
Starlink subscribers grew from 2.3M (2023) to 4.4M (2024) to 8.9M (2025) ; a 97% CAGR. Revenue grew 49.8% year-over-year to $11.4 billion in 2025. As of March 31, 2026, Starlink served 10.3 million subscribers across 164 countries, territories, and other markets, supported by approximately 9,600 broadband and mobile satellites in LEO (low-earth orbit) ; 75% of all active maneuverable satellites.
Amortizing more revenue over the cost has improved the economics :
Revenue grew 50% year-over-year. Operating income grew 120%. Adjusted EBITDA grew 86%. This is the leverage inherent in a satellite network ; once the constellation exists, each new subscriber adds revenue at near-zero marginal cost.
SpaceX’s Space segment lost money in 2025 despite launching more than 80% of all mass to orbit globally.
The losses stem from Starship development. The company has executed 11 Starship flight tests, with a 12th scheduled. SpaceX expects Starship to begin payload deliveries in the second half of 2026.
The economics of Falcon are already extraordinary. A Falcon 9 booster has demonstrated 34 reflights. At $67M per launch, and with 620 orbital launches completed and 99%+ mission success rate, SpaceX has cracked the code on reusability.
Starship aims to extend this further : 100 metric tons to LEO in fully reusable configuration, turnaround times comparable to commercial aviation, and a potential 99%+ reduction in launch costs relative to historical averages.
The xAI segment is where SpaceX is placing its biggest bet.
Of the $20.7 billion in total CapEx, 61% ($12.7B) went to AI infrastructure. SpaceX claims to have deployed the first coherent gigawatt-scale AI training cluster ; COLOSSUS and COLOSSUS II in Memphis and Mississippi.
The AI platform has scale :
550 million monthly active users across X, with 117 million using Grok’s AI features. That’s 21% AI penetration on the platform.
SpaceX’s $12.7B AI capex is substantial but dwarfed by Big Tech spending :
| Company | 2025 AI/Cloud CapEx ($B) | vs SpaceX AI |
|---|---|---|
| Microsoft | 80 | 6.3x |
| Amazon | 80 | 6.3x |
| 75 | 5.9x | |
| Oracle | 50 | 3.9x |
| SpaceX AI | 12.7 | — |
Big Tech collectively spent ~$285B+ on AI/cloud infrastructure in 2025. SpaceX’s AI capex represents 4.5% of this total, but generates only $3.2B revenue (25% revenue-to-capex ratio) compared to Big Tech’s more mature 2-3x ratios.
The S-1 reveals SpaceX’s long-term AI strategy : orbital AI compute. The thesis is elegant. AI infrastructure on Earth faces power constraints. The Sun contains 99.8% of solar system energy. Space-based solar arrays generate 5x more power per unit area than terrestrial installations. SpaceX plans to deploy AI compute satellites beginning 2028.
SpaceX targets a $1.75 trillion valuation at IPO on June 12, 2026, making it one of the five most valuable companies globally. The S-1 reveals the tremendous scale & nearly limitless ambition of a modern AI conglomerate.
2026-05-20 08:00:00
Google’s AI triples in price each year.
OpenAI’s flagship model was seemingly subsidized for a while, before rising again.
Anthropic’s AI has been the same price for a little bit & decreased for the most powerful models.
Those are three very different pricing strategies. If we compare the absolutes, the data completes the picture.
| Vendor | Model | Input ($/1M) | Output ($/1M) |
|---|---|---|---|
| Gemini 3.1 Pro | $2.00 | $12.00 | |
| Anthropic | Claude Opus 4.7 | $5.00 | $25.00 |
| OpenAI | GPT-5.5 | $5.00 | $30.00 |
Google remains the low-cost player, increasing the price on all its models but still less than half of the competition. Anthropic had maintained a luxe pricing until late last year.
The pricing changes indicate changes in strategy : cuts when cash is plentiful & share matters. Increases when cash is tight & margins matter. The latter is the case for all three vendors now when capex spending continues to set records.
2026-05-18 08:00:00
As I was paging through Good Writing, Anne Lamott’s new book, I wondered what AI would say about twisting cliches & finding hidden metaphors (chapters 18 & 19).
Over the last 16 years of writing, I’ve read books about writing, hired an editor, & used AI. I’ve fine-tuned models to mimic my voice, tested more than 10 AI systems, & written many post with AI, with some Hindenburgs I’ve kept public as proof despite my embarrassment.
Writing is hard for AI. First, AI has its own voice : Gemini beams sunshine ; Claude’s is languid but sharp ; & OpenAI Codex is the most dispassionate. Writing in another voice is hard for people & AI.
So writing with a single model doesn’t work. What about an AI editorial council? The concept shines with code review. Why not blog review?
At my fourth draft, I asked Gemini, Claude, & OpenAI Codex to edit my work with each other. The result wasn’t an elegant mosaic but a fingerpaint disaster. Each model had its own voice. Like three editors with three visions of the piece, the AI models couldn’t agree on a consistent tone or style.
And each is willing to deliver it directly, casually cruel in the name of being an editor.
Eg, this post:
Verdict. This is a three-beer conversation mistaken for a finished essay. Pick one angle : the Lamott meditation, or the AI choir experiment, or the vinyl-flare theory. Develop it with specifics, quotes, images. As written, it’s 500 words of intelligent observation without a single indelible sentence.
…imagine that daily farrago in triplicate.
AI’s ability to synthesize images, video, text means anything can be created. What’s authentic? Imperfection.
The pops of a vinyl record, the solar flare on Kodachrome film, the imperfect analogy and the punctuation peccadilloes (lovers of ampersands, unite!), stand out.
AI may generate digital reams of manuals & documentation, & may one day parrot the way we write authentically. But the imperfections of writing are what make it good writing.
2026-05-15 08:00:00
The fastest-growing companies in AI & software are either selling AI directly or reselling inference. At worst, they are the first derivative of inference.
Inference is the largest & fastest growing market in technology today, surpassing the database market & projected to be three times the size within seven years at $250 billion.1, 2 By selling inference or indexing a business to it, they grow at spectacular rates.
Anthropic has booked $9b & $10b in consecutive months.3 Google Cloud is growing 63% at an $80 billion run rate.4 Most businesses selling inference are exploding.
For public software & infrastructure companies that predate AI, there are two standouts so far : Twilio & Datadog.
Both of these companies are benefiting as the first derivatives of inference. They don’t sell inference primarily, but anyone building AI systems needs to understand how they perform, & agentic companies with voice use Twilio.
“The number of spans sent to our LLM Observability product nearly tripled quarter-over-quarter.” — Olivier Pomel, CEO, Datadog Q1 2026 Earnings Call
As a result of AI growing so spectacularly, there are huge power law dynamics.
“We now have over 6,500 customers sending data for one or more of our AI integrations. Though this is only 20% of total customers, they represent about 80% of our ARR.” — Olivier Pomel, CEO, Datadog Q1 2026 Earnings Call 5
This is also true for another element of core infrastructure, voice & SMS via telephone.
“Voice reimagined through the lens of AI is increasingly an entry point to the Twilio platform for AI natives & enterprises alike.” — Khozema Shipchandler, CEO, Twilio Q1 2026 Earnings Call 6
A few customers can drive tremendous gains. This level of concentration is characteristic of the current cycle.7
For any pre-AI company, the key question must be put at the board level : how do we either resell inference or benefit from our customers buying huge volumes of it?
That’s the only way out of the Saaspocalypse.
2026-05-14 08:00:00
In yesterday’s post (which an agent pushed in raw outline form via email!), I wrote about the future of AI email. What does that future cost?
If you are using state-of-the-art model ranging, it costs between $22 to $130 per month. Would you pay for that? At work, I imagine, many would. Let’s take the middle case of $26/month raw cost.
A software company seeking 75% gross margin would charge about $350 per year for that product excluding hosting & serving costs. So let’s call it a $500 per year list with a 15% discount at scale.
A Google Enterprise plan is $11-18/month. A fully agentic solution would then cost about twice as much.
Smaller models help. They cut cost by 10 to 20x, but we can do better.
By running the models locally, when the cost plummets to zero : users’ GPU does the work.
It’s this type of cost optimization that I have done crudely here that I think will define the next 12 to 24 months of AI software : determining which components can be executed deterministically, like the email filters, which are just rules. And the next is matching the model to the workload.
With some basic heuristics and techniques we can drop the overall cost by 100x. Given the tremendous shortage of GPUs, this segmentation of inference is inevitable.
2026-05-13 08:00:00
Nobody will open Gmail five times a day in five years.
The average knowledge worker receives 121 emails per day.1 That’s one every four minutes during working hours.
The inbox is a conveyor belt that keeps accelerating. You open Gmail. You read. You decide. You respond. One at a time. But the belt doesn’t wait. It just moves faster.
Today’s triage is generic : “This is from your boss. I need to work on that today. Next. Spam. Archive. Spam. Archive. Newsletter, read & archive.” Tomorrow’s is personal. User-defined skills & rules. Programming in English that encodes your priorities, your relationships, your workflow.
A receipt arrives & forwards itself to the expense platform before you see it. An inbound lead hits the CRM, gets scored, & a draft proposal waits in your outbox. The workflow starts the moment the email lands.
Then there’s the archive. Years of context about every relationship, commitment, & decision you’ve made. That history becomes a personal context layer that informs how your AI handles the next message. On-device models process sensitive messages privately.2
The inbox disappears. What remains are the 6 messages that actually matter.
