NextFin News - OpenAI expects to burn through almost $280 billion in cash by 2030, according to a Financial Times report that reframes the debate over the world's most valuable private company. The figure is a cumulative cash-burn projection shared with investors - not a revenue target, though it coincidentally matches the $280 billion in annual revenue the company has projected for the end of the decade. The two numbers, read together, define the wager at the center of the artificial-intelligence buildout: OpenAI is asking capital markets to pre-fund losses on a scale no startup in history has attempted, on the promise that revenue will eventually absorb a computing empire that does not yet exist.
The burn disclosure landed as OpenAI held early talks with large investors about a fresh capital raise that would value the company at about $1.2 trillion, people familiar with the matter told the Financial Times. Those discussions, initiated by investors rather than the company, come less than six months after OpenAI closed a $122 billion funding round at an $852 billion valuation. The company believes it deserves at least $1.5 trillion, based on customer traction for its Codex coding tool and its latest models, GPT-6 Astra and GPT-5.6 Sol, and told investors it generated more than $40 billion in annualized revenue last month - roughly double its sales run rate at the end of last year. OpenAI declined to comment.
The $280 Billion Burn and the $280 Billion Coincidence
The headline figure requires careful unpacking, because two different $280 billion numbers have circulated around OpenAI's finances, and conflating them misstates the company's position. The $280 billion revenue projection dates to February, when OpenAI told investors it expected total revenue for 2030 to exceed $280 billion, split nearly evenly between consumer and enterprise businesses. That figure was reported by multiple outlets at the time and was always framed as a top-line target - a sum that would place OpenAI alongside Microsoft and Alphabet in annual revenue, and ahead of every technology company except Nvidia and Apple.
The Financial Times' latest report, by contrast, puts the cumulative cash the company expects to consume through 2030 at almost the same $280 billion. Read literally, the two projections imply a company that will have burned roughly one dollar for every dollar it books over a five-year stretch - a cash conversion profile that has no clean precedent among large software businesses. Whether the near-equality is coincidence or the product of a common underlying model, it is the burn figure that carries the immediate consequence: it is the number that determines how much capital OpenAI must raise, and at what valuation, before it can stand on its own cash flow.
Other estimates of the burn cluster in the same order of magnitude but differ in scope. The Information reported in September 2025 that OpenAI had raised its projected cash burn through 2029 to $115 billion, $80 billion higher than its prior forecast, with annual burns of more than $17 billion in 2026, $35 billion in 2027, and $45 billion in 2028. A Wall Street Journal review of the company's materials put the peak burn at $85 billion in 2028, with profitability deferred to 2030. Deutsche Bank's analysis of the company's cash flow put cumulative negative free cash flow through 2029 at $143 billion - more, the bank noted, than Amazon, Tesla, Uber, and Spotify burned combined before turning profitable. HSBC's semiconductor team, modeling OpenAI's contracted compute capacity, arrived at cumulative free cash flow of negative $282 billion through 2030 and a $207 billion funding gap even after counting promised investment from Nvidia, existing cash, and undrawn facilities.
"They expect to be profitable by around 2030, but between now and then they're gonna burn $600bn in cash," a Financial Times AI Labs transcript summarized the company's position. "And that's not their cash, that's cash they have to go and raise from investors, or if they go public, they can raise some of it there."
Where the Cash Goes
The composition of the burn matters as much as the total. OpenAI's most recent full-year financials, disclosed in audited documents reviewed by outside analysts and independently verified, showed revenue of $13.07 billion in 2025 - up 253% from $3.7 billion in 2024 - against $34 billion in total costs and expenses. Research and development alone consumed $19.18 billion, more than the entire top line. Sales and marketing rose more than fivefold, to $5.73 billion. The operating loss was $20.92 billion, and the net loss attributable to the company reached $38.53 billion after a $41.55 billion fair-value charge tied to the conversion from a nonprofit to a for-profit entity.
Going forward, the dominant cost is compute. OpenAI has committed to roughly $600 billion in compute spending through 2030 - a figure revised down from the $1.4 trillion and 30 gigawatts of capacity CEO Sam Altman previously touted, and presented to investors as better alignment between spending and expected revenue growth. HSBC's model translates OpenAI's long-term contracts - including a $250 billion agreement with Microsoft and a $38 billion deal with Amazon that injected no new capital - into $792 billion of cumulative data-center rental costs between late 2025 and 2030, with an annual rental bill that could reach $620 billion at full utilization. Total compute commitments may hit $1.4 trillion by 2033.
The second pressure point is margin. Inference costs - what OpenAI spends to run its models for users - quadrupled in 2025, pulling adjusted gross margin down to 33% from 40% in 2024, against a company target of 46%. That six-point miss is the fulcrum of the entire thesis. In traditional software, incremental revenue arrives with near-zero incremental cost and margins expand with scale. In inference, revenue and cost are coupled: every additional query adds expense, so gross margin expands only if the cost per query falls faster than the price per query. OpenAI is betting it can pull both levers - cheaper chips, more efficient models, and pricing power from frontier capability - but the 2025 margin print shows the bet is not yet working.
OpenAI frames the challenge as an infrastructure race, not a margin problem. "AI demand is surging across consumers, developers, and businesses," the company wrote when announcing its February funding. "Meeting that demand and providing everyone access to our products requires three things: compute, distribution, and capital." It added: "Leadership will be defined by who can scale infrastructure fast enough to meet demand, and turn that capacity into products people rely on."
Cyclical Capex or Structural Margin Problem?
This is the judgment investors must make, and the two readings lead to opposite conclusions. The cyclical reading is that the burn is front-loaded capital expenditure on a fixed capacity base. Once the data centers are rented and the GPUs deployed, each incremental inference costs a fraction of the average cost, and operating leverage kicks in hard. Under this view, the $280 billion revenue target is not aspiration - it is the volume required to absorb the capacity, and the path to 2030 profitability is a classic scale-up curve executed at a capital intensity no software company has ever attempted. The cash gap is a financing problem, and financing problems are solvable as long as the equity window stays open.
The structural reading is more damaging, and it rests on three forces that do not mean-revert on their own. First, inference cost scales with usage, so revenue growth and cost growth are coupled and gross margin does not automatically expand with volume. Second, model performance is commoditizing: one competitor's latest model reportedly matches OpenAI's flagship performance at one-tenth the cost, which caps pricing power and pushes the industry toward a price war OpenAI can ill afford while burning tens of billions a year. Third, demand is contestable: ChatGPT's share of chatbot web traffic fell from 86.7% in January 2025 to 64.5% in January 2026, with Google's Gemini capturing much of the loss, even as OpenAI maintains that 92% of Fortune 500 companies use ChatGPT and that enterprise seats grew ninefold last year.
On balance, the verdict splits by time horizon. The cash gap is cyclical - a financing problem that deep, patient capital markets can solve if they are willing to keep writing checks at trillion-dollar valuations for multiple years. The margin structure is the structural question, and it is not yet answered. A company that burned $20.9 billion in 2025 while revenue grew 253% is not failing to sell; it is selling at unit economics that only work if costs fall faster than prices. That is a condition, not a guarantee.
The Counter-Thesis: Why Believers Think the Math Works
The bullish case has real foundations, and it deserves the weight the market is giving it. Revenue more than tripled in 2025, to $13.07 billion from $3.7 billion, and the company told investors it had passed $40 billion in annualized revenue by late last month - a pace that would put the 2030 target within reach if it holds. The company has said it has 900 million weekly active users and more than 50 million consumer subscribers. It began testing advertising in 2026 and, according to a report citing a source familiar with investor presentations, crossed $100 million in annualized ad revenue within six weeks of launch, with projections of $2.5 billion for the year, rising to $11 billion in 2027, $25 billion in 2028, $53 billion in 2029, and $100 billion by 2030 if products reach 2.75 billion weekly users. Enterprise adoption is real: nine million paying business users rely on ChatGPT for work, and Codex has more than 1.6 million weekly users, tripled since the start of the year.
The valuation market is voting with capital. After the February round at a $730 billion pre-money valuation and the March round near $852 billion, investors approached the company in September about a round at $1.2 trillion, with OpenAI arguing for $1.5 trillion. That would leapfrog rival Anthropic, which raised in May at a $965 billion valuation and is itself preparing an IPO that could come as soon as October on Nasdaq, seeking to raise as much as or more than SpaceX's record $86.3 billion June offering. HSBC, despite its skeptical cash-flow model, stated it still views AI as a "megacycle" and that its forecasts "indicate a leading position for OpenAI from a revenue standpoint."
The counter-thesis, in short, is that the burn is the price of admission to a market that only one or two companies will win, and OpenAI is the most likely winner. If it captures even a fraction of the enterprise software, coding, and advertising budgets it is targeting, $280 billion in revenue by 2030 is conservative rather than aggressive - and the $280 billion burn is simply the cost of building the monopoly infrastructure before anyone else can.
The answer to the counter-thesis is not that the opportunity is fake. It is that the financing burden falls due before the opportunity matures. Anthropic, by comparison, projects profitability in 2029 - a year ahead of OpenAI - while burning far less. And the capital is not free: each successive round at a higher markup narrows the return for new investors and raises the bar for the eventual public exit. A $1.5 trillion private valuation requires the public market to assign OpenAI a multiple that leaves little room for the execution risk embedded in the burn.
The Funding Gap and the IPO Clock
Every burn projection ultimately resolves into a financing question, and the timing is awkward. The $207 billion funding gap HSBC identifies must be filled through debt, equity, or revenue that exceeds expectations - and the most natural source, an initial public offering, has been pushed out. Sam Altman said in a Fortune interview published on Saturday that the company will not go public in 2026, citing AI safety concerns that render even a 10% risk of human extinction by decade's end "unacceptable." OpenAI already filed a confidential draft prospectus with the Securities and Exchange Commission in June. Delaying the public listing while burning tens of billions per year means the private market must keep writing checks at trillion-dollar valuations for multiple years.
That is a bet on market appetite as much as on technology. The private AI funding market has shown it can absorb enormous rounds - SpaceX raised a record $86.3 billion in June - but the investor-initiated nature of the current talks is itself a signal. Investors, not the company, approached OpenAI, and the discussions have not progressed far; the $1.2 trillion figure may not hold as negotiations develop. OpenAI's desire for $1.5 trillion, anchored to Codex traction and the Astra and Sol models, sets up a negotiation in which the company is selling a growth story that requires the market to believe both that demand will compound without interruption and that margins will recover toward the 46% target the company missed in 2025.
There is also a securities-law wrinkle: having already filed confidentially, OpenAI must navigate carefully what information it can share with private investors ahead of a public offering. The burn figure itself - almost $280 billion by 2030 - is exactly the kind of forward-looking financial detail that public-market investors will scrutinize in a prospectus, and its appearance in private discussions now gives the market an early look at the mountain the company must climb.
What to Watch: The Signals That Decide the Thesis
Three specific signals will determine whether OpenAI's burn is a bridge to profitability or a hole that keeps deepening.
First, gross margin. The falsifying threshold is 46% - the margin OpenAI itself targeted. If adjusted gross margin does not recover meaningfully from the 33% reported for 2025 by the 2027 reporting cycle, the scale-economics thesis is broken: revenue growth is not producing operating leverage, and the burn is structural rather than front-loaded.
Second, utilization. OpenAI has committed to fixed capacity costs. If ChatGPT's traffic share continues to erode from the 64.5% level while those commitments remain on the books, the cost per utilized unit rises even as revenue per user falls under competitive pressure.
Third, the round itself. If the $1.2 trillion to $1.5 trillion funding does not clear, or clears at a materially lower valuation, the funding gap becomes an immediate constraint rather than a multi-year planning item. Conversely, a successful close at or near $1.5 trillion would validate the market's willingness to pre-fund the burn and push the solvency question out past the 2030 profitability target.
Short term, the story is financing: can OpenAI raise enough, at what valuation, and on what terms? Medium term, it is margin: does inference cost fall faster than price? Long term, it is whether the $280 billion revenue target is a realistic absorption of the capacity being rented, or a number that only works if AI demand compounds without interruption for five straight years.
The central judgment: OpenAI's $280 billion burn is a solvable financing problem only if its margin problem resolves in the other direction - costs falling faster than prices. The burn figure is not the story. The story is that the company must become one of the largest revenue generators on the planet while proving, year by year, that each new dollar of AI demand arrives with more margin attached than the last. If it cannot, the most valuable private company in the world is simply the most expensive way ever devised to learn that scale does not guarantee profitability.
Explore more exclusive insights at nextfin.ai.
