OpenAI expects a cumulative free cash flow deficit of $278 billion over the five years from 2026 to 2030, the Financial Times (FT) reported on September 18. Even if annual revenue in 2030 grows to roughly ten times its 2026 level, the forecast shows enormous spending on computing resources and related infrastructure continuing. A simple growth story, in which wider AI adoption automatically eases the cash crunch, does not fully explain the picture. Putting OpenAI's and Amazon's public disclosures side by side reveals how hard it is to run a business that secures computing capacity ahead of future demand while paying for it and building revenue at the same time.
Five years of cash outflow against single-year revenue targets
According to the FT-reported forecast, cumulative revenue through 2030 reaches $840 billion. Over the same period, spending on computing resources and related infrastructure is projected at about $856 billion. That single largest expense category exceeds cumulative revenue.
| Item | Period | Reported forecast |
|---|---|---|
| Revenue | 2026 (single year) | $36 billion |
| Revenue | 2030 (single year) | $350 billion |
| Cumulative revenue | 2026–2030 | $840 billion |
| Cumulative spending on computing resources and related infrastructure | 2026–2030 | About $856 billion |
| Cumulative free cash flow | 2026–2030 | –$278 billion |
The source is the FT report. The company documents underlying the forecast have not been made public. Bloomberg confirmed with people familiar with the matter that the materials accompanied a July computing-resources deal. They should be distinguished from any earnings outlook OpenAI has announced publicly.
Dividing by the same five-year figures, computing and infrastructure spending equals about 101.9% of revenue (8,560 ÷ 8,400 × 100), a gap of $16 billion. However, this ratio is not an accounting gross margin. Capital investment in equipment and service usage fees are recorded differently, and there are timing gaps between recognizing revenue and receiving cash. The publicly available figures are not enough to reconstruct the $278 billion outflow item by item.
Free cash flow is generally cash generated from operations minus capital expenditures and similar outlays. A deficit has to be covered by cash on hand or outside funding, but that amount does not translate directly into debt or net loss. The $278 billion is likewise not a cash shortfall that exists today; it is a forecast summed over the next five years.
Still, the fact that cash outflow continues even after factoring in a sharp rise in revenue is significant. What is at stake is not only whether demand grows, but how far the payments needed to support that demand run ahead of the cash collected from customers.
Securing capacity in advance, and the lag before revenue comes in
OpenAI Chief Financial Officer Sarah Friar said in a January 2026 post that securing computing resources requires contracts signed years in advance, and that supply capacity and usage do not grow at the same pace. There are times when demand outpaces infrastructure and times when infrastructure runs ahead of demand. The approach to handling that mismatch is to diversify partners, use flexible contracts, and invest in stages according to demand.
The scale of long-term contracts shows in the Amazon partnership. According to the companies' February 27 announcement, OpenAI will add $100 billion over eight years to its existing $38 billion AWS usage agreement. The commitment includes using about 2 gigawatts of computing capacity based on AWS's AI chip "Trainium."
Securing capacity first makes it easier to develop models and expand services. But it takes time before that capacity is fully used and converted into customer revenue. Computing resources are needed not only to train models but continuously for "inference," the processing that responds to ChatGPT queries and business AI tasks. More users mean more revenue opportunities and more operating costs at the same time.
The additional $100 billion for AWS should not simply be added to the reported $856 billion. The former is an eight-year contract; the latter is a forecast through 2030. It has also not been disclosed which portions of the contract are included in the forecast. What can be read from this is the fact that OpenAI is securing future supply over the long term, not a new total spending figure obtained by combining the two.
Amazon's $50 billion has already been invested
In its March 31 funding announcement, OpenAI said it had secured $122 billion in investment commitments, putting its post-money valuation at $852 billion. Commitments and actual deposits are different things, and checking how investments have actually been executed changes how the funding looks.
In Amazon's case, the February announcement described a structure in which $15 billion of the $50 billion would be invested first, with the remaining $35 billion invested after certain conditions were met. But Amazon's Form 10-Q for the second quarter of 2026 records how that has since played out.
| Timing of investment disclosed by Amazon | Investment in OpenAI |
|---|---|
| Q1 2026 | $15 billion |
| Q2 2026 | $13.7 billion |
| Subsequent event after June 30, 2026 | $21.3 billion |
| Total | $50 billion |
Adding up Amazon's disclosures, its investment in OpenAI comes to $15 billion + $13.7 billion + $21.3 billion = $50 billion, reaching the originally planned amount. The table converts to billions of dollars the investment in Series C preferred stock described in Note 2, "OpenAI," of the same filing. That does not mean the full amount was paid by the end of June, but the conditional investment described in February can no longer be treated as funds that remain unexecuted.
At the same time, the $50 billion investment does not tell us OpenAI's current cash balance. After funds come in, there are business payments, and there are also deposits from other investors and revenue from customers. The FT has reported that the funds secured in March are expected to be used up by 2028, but that is an estimate based on a future income-and-expense plan, not a clock counting down to bankruptcy.
Amazon provides capital to OpenAI and is also on the selling side of computing resources through AWS. These two transactions support each other, helping OpenAI secure capacity and raise funds. But an investment is not revenue from customers paying to use AI. Whether a company can raise money from investors and whether usage fees can cover its service costs must be checked separately.
What it takes to turn efficiency into profit
In an August 25 post, Friar outlined a policy of getting more output from the same computing resources through improvements in models and software, in-house chip design, and other measures. That means choosing chips and suppliers suited to each task and cutting wasteful processing and answer retries. OpenAI aims to expand capacity while improving per-task processing efficiency.
However, even if the cost per task falls, total company spending does not necessarily decline. If cheaper AI is used for work that previously did not justify the cost, total processing volume increases. Friar herself has referred to the "Jevons paradox," in which efficiency gains create new demand. This is an explanation of the company's growth strategy, not a guarantee of future profit.
Assessing profitability also requires knowing what is offered under which pricing structure. In addition to flat-rate subscriptions, OpenAI combines usage-based API billing, advertising, and other models. At the time of the March announcement, enterprise customers accounted for more than 40% of revenue. Converting all future revenue into consumer monthly fees alone would therefore not capture the actual business mix.
For example, under a flat-rate service, revenue does not rise each time a subscriber runs more processing. Usage-based billing makes it easier to convert increased usage into revenue, but the question remains whether the gap between per-unit price and cost is large enough. And if savings are passed on to users through price cuts, not all of the efficiency gains can flow to OpenAI's profit.
For this reason, showing revenue growth and improved compute efficiency separately does not prove a lasting improvement in the income statement. A business's profitability can be evaluated only when it is clear how much was collected from the increased usage and how much cash was paid out to support that usage.
What to check before the 2030 target
The 2030 revenue target of $350 billion alone does not show how much funding will be needed in which intervening year. If facilities come online late, revenue could be missed; if capacity is secured ahead of demand, the burden before recovery grows heavier. What determines whether the plan succeeds is not only total future revenue but when the facilities start being used and the order of cash in and out.
Useful indicators include how heavily the reserved computing capacity is used, how revenue and cost per unit of usage change, and whether the funds raised actually arrive by the time they are needed. The reported cumulative figures do not adequately show a year-by-year breakdown or the payment schedule for each contract. Confirming Amazon's investment execution matters because it brings the headline funding amount closer to actual cash movements.
If OpenAI can fill its expanded capacity with paying demand and turn gains in processing efficiency into higher collections, the enormous upfront investment could support business expansion. To measure that progress, it will be necessary to check not only the size of the next major fundraising but also whether annual cash outflow shrinks relative to revenue.