Sam Altman's Bet: When Desperation Becomes Strategy
When the numbers don't work, you announce a new business line. $1.4 trillion in data center commitments. $20 billion in revenue. That's a 70:1 ratio.
Sam Altman announced on November 6 that OpenAI will launch cloud services. Same day, Ilya Sutskever’s deposition resurfaced - 52 pages documenting Altman’s “dishonesty and manipulation” from the 2023 board crisis.
Two events. One pattern.
The cloud announcement is narrative repair. Altman needs a revenue story investors will accept.
The Math That Forces This Move
OpenAI projects $28 billion on Microsoft compute in 2025 alone - triple the 2024 total. Add CoreWeave’s $12.9 billion five-year deal, Stargate’s $19 billion commitment, salaries, data - the cash depletion hits $155 billion by 2026.
Total liquidity: $16-20 billion.
Even with SoftBank’s capital, the reserves vanish in under two years. Meanwhile, OpenAI enters enterprise cloud against Microsoft, Amazon and Google - companies that spent 15 years building operational moats.
The cloud business is the only revenue lever big enough.
Three Dependencies. Zero Execution Path.
Altman is trying to manage three sequential dependencies treated as parallel workstreams:
OpenAI consumes massive compute for its own products. Altman complained in March that “GPUs are melting.” He begged publicly for capacity in “100k chunks.”
The company builds infrastructure through Stargate.
Now it plans to sell that infrastructure to enterprise customers as a cloud provider.
These are sequential dependencies. Enterprise cloud requires operational rigor that takes years - security frameworks, support, compliance, multi-year sales cycles.
AWS, Azure and Google Cloud command 75% of global market share. They built what OpenAI needs to replicate at scale.
OpenAI has neither.
The Vendor Trap
OpenAI is simultaneously a major customer of Microsoft, Amazon and Google - and now their competitor.
Microsoft: $250 billion in contractual Azure commitments remaining.
Amazon: $38 billion deal signed in November.
Both supply the infrastructure OpenAI plans to resell at lower margins.
Suppliers gain no advantage from helping OpenAI become efficient. They retain pricing power. They see OpenAI’s costs and customer strategy in real-time.
When Microsoft walked away from 1 gigawatt of data center commitments recently, it was calculation. Why build capacity your customer will use to undercut your own cloud business?
The incentives break.
Stargate: Announced Big, Delivered Small
January 2025: Announce Stargate with Trump and Masayoshi Son at the White House. $500 billion promised. $100 billion by year-end.
August 2025: SoftBank quietly admits the project is “moving slower than usual.”
Results by mid-2025: Zero new land deals beyond the Texas site. $38 billion committed - $19B from OpenAI, $19B from SoftBank. Well below the $100 billion first-year target.
The Texas facility remains incomplete. Construction delays tie to energy constraints, site selection disputes, unclear funding for the remaining $462 billion.
Altman announces cloud services predicated on capacity that’s years behind schedule.
The timeline slips before the “strategy” begins.
The Pattern Sutskever Documented
Sutskever’s memo identified a specific failure mode:
Misalignment between announced commitments and operational reality, masked by new announcements rather than diagnosed.
Announce Stargate. Admit delays. Announce cloud services.
Each announcement obscures the previous failure.
Sutskever didn’t tell Altman his concerns because he believed Altman would “make them disappear.” That level of distrust - at the chief scientist and board member level - tells you what governance looked like.
What the Board’s “Vindication” Actually Meant
The board removed Altman in November 2023. Employee threats to quit forced reinstatement within a week.
The board vindicated Altman in 2024 because removing him was too costly. The memo remained. The concerns persisted.
But vindication was premature.
The strategic errors Sutskever documented remain embedded in OpenAI’s decision-making. Altman commits to $1.4 trillion without clear profitability paths. Stargate misses targets. The response: announce another business line using the same delayed infrastructure.
Execution problems get obscured by announcements.
If Altman was untrustworthy enough to remove in 2023, what changed by 2024?
Nothing.
Altman won because he had the employees and the investors. The board capitulated.
Force. Not strategy.
Force won’t change infrastructure timelines. It won’t fix capital ratios. And it won’t resolve the vendor conflicts where your suppliers are your competitors.
The Y Combinator Playbook Applied to Infrastructure
This shows Altman’s Y Combinator pattern: announce audacious ambitions, encounter friction, reframe the narrative, announce new opportunities.
In venture capital, this works. Founder charisma moves markets.
Infrastructure operates on different constraints.
18-24 month data center construction timelines. Global TSMC capacity limits. Enterprise sales cycles that demand multi-year proof.
You can’t compress timelines through storytelling.
Altman’s response to Stargate delays was to announce a new business line - not explain how delays would be resolved.
The Question Nobody’s Answering
If OpenAI’s compute is so scarce you’re begging for GPUs, where’s the surplus to sell cloud customers?
It’s not there. Not yet. Not on announced timelines.
But buying time only works if you use it to solve execution problems.
The question:
Can OpenAI survive long enough to build the infrastructure foundation that cloud services requires?
On current timelines, that’s harder to answer every quarter.
Looking Forward: What to Watch
Stargate updates. Has Crusoe or Oracle disclosed new timelines or adjusted projections? Delays beyond mid-2026 signal Altman’s cloud narrative is in trouble.
SoftBank’s quarterly earnings and rating agency commentary. S&P warned about financial deterioration. Watch for downgrades or revisions to SoftBank’s debt outlook based on funding progress.
Microsoft’s stance on additional capacity commitments. Will Microsoft distance itself from OpenAI’s cloud plans, or quietly limit capacity? Any public statement from leadership about cloud competition is a direct signal.
OpenAI’s enterprise offerings. Altman claims one million business customers. Watch for pricing, SLAs and go-to-market strategy announcements. Vagueness is a tell.
Anthropic’s infrastructure strategy response. With Google committing massive TPU capacity, Dario Amodei will likely position how Anthropic approaches enterprise customers. Anthropic could become the “enterprise-friendly” alternative to OpenAI’s overcombined stack.
The nonprofit-to-for-profit conversion deadline. OpenAI must convert by December 2025 or lose $10 billion in funding. Legal challenges from Musk and civil society organizations are in motion. Any adverse ruling or delay signals deeper structural problems.
Here is the deal.
If you aren’t subscribed, subscribe. It takes two seconds. It costs nothing. It separates the signal from the noise.
If you have a voice, restack it. Let your network see the signal.
And if this landed for you, if it gave you the language to name the problem, upgrade to Paid.
It is the only way to support the weekly Signals and influence the next deep-dive Analysis.
Sources
TechCrunch: “Sam Altman says OpenAI has $20B ARR and about $1.4 trillion in data center commitments”
Business Insider: “Did Sam Altman just announce an OpenAI cloud service?”
The Neuron AI: “Ilya Sutskever’s secret memo and the plot to merge OpenAI with Anthropic”
Business Insider: “OpenAI debated merging with one of its biggest rivals after firing Sam Altman, court docs reveal”
Where’s Your Ed At: “OpenAI Is A Systemic Risk To The Tech Industry”
Bloomberg: “SoftBank Concedes Stargate Project with OpenAI Needs More Time”
TechCrunch: “OpenAI and Amazon ink $38B cloud computing deal”