OpenAI expects to burn through almost $280 billion in cash between 2026 and 2030, according to an internal presentation described in Reuters' report on the Financial Times' findings, published September 18, 2026.
Most of that money goes toward computing power. The company's own capital spending plans reportedly reach roughly $856 billion by the end of the decade.
That number should shape how any business builds its AI vendor strategy. It matters just as much to a company routing customer workflows through OpenAI's API as it does to Wall Street reading the balance sheet.
The same report said OpenAI expects to exhaust its current cash reserves by 2028, two years before the window it's projecting against.
Sam Altman has separately ruled out an IPO this year, citing AI safety concerns, even after the company had filed confidentially to go public in June. Investors have reportedly discussed a valuation near $1.2 trillion ahead of any future listing.
OpenAI's Numbers Keep Moving
Seven months earlier, the picture looked different. In February, OpenAI told investors it expected revenue to top $280 billion by 2030, backed by a funding round expected to raise more than $100 billion.
CFO Sarah Friar credited the momentum to subscription sales of the company's AI software. Planned infrastructure spending at the time sat well under what OpenAI is reportedly telling investors now.
Notice what changed. The same $280 billion figure now describes cash burn instead of revenue.
A business that signed a multiyear contract based on February's numbers is operating against a different set of assumptions today, without ever agreeing to the change.
Where AI Vendor Lock-In Actually Bites
None of this means OpenAI is in trouble. A company at this scale can run on negative free cash flow for years if investors keep funding the gap.
But it does mean a customer's own product roadmap can end up tied to financing decisions made well outside that customer's contract.
According to TechTarget's breakdown of AI vendor lock-in, the dependency usually sits in a handful of specific places once a business has built on one provider.
- Proprietary APIs that need rewriting to switch providers
- Specialized infrastructure, like GPUs, tied to one vendor's stack
- Vector databases and logging tools embedded in daily workflows
- Multiyear contracts with step-pricing that punishes an early exit
Stack enough of those together and switching providers stops being a technical decision. It becomes a business risk nobody signed up for on purpose.
Building an AI Vendor Strategy That Survives a Shakeout
Start by separating what genuinely needs a frontier model from what doesn't. A workflow that qualifies leads or drafts a first response doesn't need the same model powering a research assistant.
Building that distinction into your AI and automation layer means a pricing change on one provider doesn't take the whole system down with it.
An abstraction layer between your product and any single model's API costs an extra week of engineering up front. Testing a second provider on a low-stakes workflow before you need it costs even less.
That way, the first time you try a fallback isn't during an actual outage. We've covered the related gap in how little OpenAI discloses about its own models, and in why AI agent oversight needs more than more AI to catch problems before a customer does.
Both point the same direction: don't build on the assumption that one vendor's roadmap will keep matching yours. If your CRM automation already runs through a single AI vendor, the cheapest time to add a second option is now, while you're the one choosing it.
Sources
- OpenAI expects to burn through almost $280 billion by 2030, FT reports — Reuters (via Investing.com)
- OpenAI forecasts its revenue will top $280 billion in 2030 — Fortune
