BUSINESS ENGLISH · C1 · 50 min

The Trillion-Dollar Maybe

OpenAI has a product nearly a billion people use every week, revenue that more than tripled in a single year — and operating losses larger than that revenue. It has raised well over $100 billion on one promise: that the next, bigger model will be worth it. By 2030 we will know. Until then, everyone is guessing.

Easier version: read this lesson at A2/B1 level →

Today: you’ll follow the financial and structural bet behind OpenAI — a research lab that became one of the most valuable private companies in the world while losing more money each year — and practise the language of projections: by 2030 it will have spent…, it will still be burning cash…, it is expected to…, it may well have…

Before we start. Talk these through out loud.

Discuss:

Would you invest in a company that loses more than it earns, if its revenue were tripling every year? What would you need to see?
Do you use ChatGPT or a similar tool? Would you pay more for it — and how much more?
Can a company be run for “the benefit of humanity” and for its shareholders at the same time?

The vocabulary of this case

Ten items. Work through the four activities — say your answer out loud before you reveal it.

A · What do they mean here? Work it out from the context

1. OpenAI expects to burn through $25 billion of cash in 2026. Investors watch the cash burn closely, because it tells them how long the money will last.

cash burn (n) / to burn (through) cash (v) — the rate at which a loss-making company spends its money. Related: burn rate, runway (how long until the cash runs out).

2. Training and running AI models needs enormous compute — chips, data centres and the electricity to power them.

compute (n, uncountable) — in AI, computing power as a resource you buy, like fuel. Collocations: secure / buy / run out of compute; compute capacity; compute deals.

3. Altman said the company would end 2025 above a $20 billion annualised run rate, even though its revenue for the whole year was about $13 billion.

a run rate (n) — current revenue projected over a full year (e.g. this month × 12). Useful for fast-growing companies; also easy to flatter. Compare reported annual revenue.

B · Which term means…?

Three terms from the case: a valuation · a capped return · governance. Say the right term, then reveal.

1. Which term means the price investors put on a whole company when they buy part of it?

a valuation — a post-money valuation of $852bn (= value including the new money). Collocations: at a valuation of; justify / slash a valuation.

2. Which term means a limit on how much profit an investor can ever make from an investment?

a capped return — in 2019 OpenAI capped its first investors’ returns at 100 times their money; anything above went to the non-profit. Also a cap, capped-profit.

3. Which term means the system of rules and people — especially the board — that controls a company?

governance (n, uncountable) — corporate governance; a governance crisis / failure / structure.

C · Choose the meaning — a / b / c

1. Altman said OpenAI had commitments of about $1.4 trillion for computing over eight years. Commitments here means…
(a) money already spent (b) promises to pay in the future, usually in contracts (c) donations from supporters

(b) — make / honour / meet / walk away from a commitment; a funding commitment; committed capital (money promised to an investment round, not all paid at once).

2. OpenAI has told investors it does not expect to break even on cash flow until about 2030. This means…
(a) to stop growing (b) to bring in as much cash as it spends (c) to pay back all its investors

(b) — break even (v); the break-even point (n). After it, the company is cash-flow positive.

D · Collocation & word building

1. In 2025 OpenAI rebuilt its ownership — who holds which shares — in a process called a ___. (from capital)

recapitalisation (US recapitalization) — changing a company’s mix of equity and debt or its ownership structure. Verb: to recapitalise.

2. The loss from running the business, before tax and one-off accounting items, is the ___ loss. Which adjective?

operating loss — compare net loss (the bottom line, after everything). OpenAI’s 2025 net loss was much bigger than its operating loss because of a one-off charge.

Now use them — out loud

Talk it through before we read:

Using cash burn, compute and commitments, explain in thirty seconds why a fast-growing AI company might still lose money.
Why might a run rate and a reported annual figure tell different stories? When would you trust each?

The case, in three parts

Read each part. After it, there's a quick check (answer from memory first), then one question to discuss.

Part 1 — A lab that needed billions (2015–2022)

OpenAI began in December 2015 as a non-profit research lab, with $1 billion pledged by its backers and a mission “to ensure that artificial general intelligence (AGI) benefits all of humanity”. By 2019 it had concluded that the most powerful AI systems needed the most computing power, and that it would have to “invest billions of dollars in upcoming years” to compete. Donations would not be enough.

So in March 2019 it invented an unusual structure: a “capped-profit” company, controlled by the non-profit’s board, in which the first investors could earn at most 100 times their money. Anything above that would belong to the non-profit. That July, Microsoft invested $1 billion. In June 2020 OpenAI launched its first commercial product, an API that let other companies pay to use its GPT-3 model. It was a business — but a small one, built around research.

Then, on 30 November 2022, it released ChatGPT. By January 2023, according to an estimate by analysts at UBS, it had about 100 million monthly users — the fastest-growing consumer app they had seen. That same month Microsoft announced a further “multiyear, multibillion-dollar” investment; by late 2025 its total funding commitment had reached $13 billion.

Quick check — from memory, then reveal:

1. Why did OpenAI create a “capped-profit” company in 2019?

2. Did OpenAI have products and revenue before ChatGPT?

1. Because it believed it would need billions of dollars for computing power and people — more than a non-profit could raise — while keeping the non-profit board in control and limiting investors’ returns.

2. Yes — its API, launched in June 2020, let businesses pay to use GPT-3. ChatGPT turned a small business into a mass-market one.

Discuss:

A cap of 100 times your money: is that a real limit or a symbolic one? What does it tell investors about how big OpenAI expected to become?

Part 2 — The weekend the structure was tested (2023–2025)

On Friday 17 November 2023 the non-profit board fired Sam Altman, OpenAI’s chief executive, saying he had not been “consistently candid in his communications with the board”. Within days, nearly all of the company’s roughly 800 employees signed a letter demanding the board resign and Altman return. On 21 November he did, with a new board chaired by Bret Taylor. The structure designed to keep the mission in charge had used its power — and lost.

The spending, meanwhile, became the story. In January 2025 Altman stood at the White House to announce Stargate, a $500 billion data-centre project with SoftBank and other partners. More deals followed with Oracle, Nvidia, AMD, CoreWeave, Google, Amazon and Microsoft. By November 2025 Altman said OpenAI was looking at commitments of about $1.4 trillion over eight years, because “the risk to OpenAI of not having enough computing power is more significant and more likely than the risk of having too much.”

To raise that kind of money, the structure had to change again. On 28 October 2025 OpenAI completed a recapitalisation. The non-profit became the OpenAI Foundation, holding 26% of a new public benefit corporation, OpenAI Group PBC — a company legally required to pursue its mission as well as profit. Microsoft held about 27%; employees and other investors the remaining 47%. Reuters described it as a $500 billion company. Altman said a stock-market listing was the most likely path.

Quick check — from memory, then reveal:

1. What happened between 17 and 21 November 2023?

2. What did the October 2025 recapitalisation create, and who owned what?

1. The board fired Altman for not being “consistently candid”; nearly all staff threatened to leave; he returned as CEO with a new board.

2. OpenAI Group PBC, controlled by the OpenAI Foundation (26%); Microsoft about 27%; employees and investors 47%.

Discuss:

Altman’s argument is that running short of compute is a bigger risk than having too much. When is “too little capacity” really the bigger danger — and when is that just what a builder wants to believe?

Part 3 — The numbers (2025–today)

OpenAI’s audited accounts for 2025, reported in June 2026 by the Financial Times and others, showed revenue of $13.07 billion, up from $3.7 billion in 2024. Research and development alone cost $19.18 billion. The operating loss was $20.92 billion. The net loss was nearly $39 billion, though about $30 billion of that was reportedly a one-off accounting charge linked to the restructuring.

Investors kept paying anyway. On 31 March 2026 OpenAI closed the largest funding round in Silicon Valley history: $122 billion of committed capital, at a valuation of $852 billion, anchored by Amazon (up to $50 billion), Nvidia and SoftBank ($30 billion each), and including more than $3 billion from individual investors through banks. At the same time it was cutting back — it shut its Sora video app — and in April the Wall Street Journal reported that it had missed an internal revenue target, with rivals Anthropic and Google gaining ground.

The forecasts are the bet in numbers. According to company documents reported by The Information and the Financial Times, OpenAI expects to burn about $25 billion in 2026 and around $278 billion in total between 2026 and 2030, and to become cash-flow positive only in 2030. In September 2026 Altman said there would be no stock-market listing this year; later that month Bloomberg reported that OpenAI was seeking new money at a valuation of about $1.4 trillion. Whether the next models will earn back what is being spent on them is still an open question.

Quick check — from memory, then reveal:

1. Give three figures from OpenAI’s 2025 accounts.

2. What does OpenAI reportedly expect for 2026–2030?

1. Revenue $13.07bn (from $3.7bn); R&D $19.18bn; operating loss $20.92bn; net loss nearly $39bn including a one-off charge of about $30bn.

2. About $25bn of cash burn in 2026, around $278bn of negative cash flow over 2026–2030, and positive cash flow only in 2030.

Discuss:

Every new round has been at a higher valuation, while the losses have also grown. What has to be true for both of those things to make sense at once?

Grammar — the future perfect & the future continuous

OpenAI’s story is told almost entirely in forecasts: what it will have spent, what it will be doing, where it will be by 2030. Two forms do most of this work — and at C1 you combine them with the language of probability.

1 · Future perfect — looking back from a point in the future
will have + past participle: an action completed before a future time.
• “By 2030 OpenAI will have burned through hundreds of billions of dollars.”
• “By the time it lists, investors will have put in more than any start-up in history.”
Continuous version — duration up to that point: will have been + -ing: “By 2030 it will have been building AI for fifteen years.”
Passive: “By 2030 the new data centres will have been built.”

2 · Future continuous — in progress at a future time
will be + -ing: an action happening at a future moment, or a future arrangement that will happen as a matter of course.
• “In 2028 the company will still be burning cash.”
• “Over the next eight years OpenAI will be paying for data centres it has already committed to.”
• Polite questions about plans: “Will you be raising more money this year?” (softer than Will you raise…?)

3 · Adding speculation — how sure are you?
Replace will with a modal or add an adverb: may / might / could have + past participle · should have (= expected) · will probably / almost certainly have.
• “By 2027 it may well have overtaken rivals — or they might have caught up.”
• “If the forecasts hold, it should be generating positive cash flow by 2030.”
Form watch: after by the time / when / before, use a present tense, not will: By the time it breaks even… (not ~~will break~~).

1. Perfect or continuous?

Put the verb in brackets into the future perfect (will have done) or the future continuous (will be doing).

1. By the end of 2030, OpenAI (spend) roughly $278 billion more than it earns over five years — if the forecasts are right.

2. This time next year, engineers (train) the next generation of models.

3. By the time the company lists, Microsoft (be) a shareholder for at least eight years.

4. Don’t call the CFO at nine tomorrow — she (present) to investors.

1. will have spent 2. will be training 3. will have been 4. will be presenting

2. Add the right degree of doubt

Rewrite each sentence with the word in brackets, keeping the future time.

1. By 2028 rivals will have caught up. (may well) → By 2028 rivals caught up.

2. In 2030 the company will be making money. (should) → In 2030 the company making money.

3. By then OpenAI will have listed on the stock market. (almost certainly) → By then OpenAI listed.

4. By the time it (break) even, it will have raised hundreds of billions.

1. may well have 2. should be 3. will almost certainly have 4. breaks (present after by the time)

3. The CFO’s statement — complete it, then deliver it

An investor asks: “Where will OpenAI be in 2030?” Complete the answer, then deliver it aloud in under a minute.

“Let me be careful. By 2030 we have built the computing capacity we have already committed to. We still be investing heavily in 2028 and 2029 — that is in the plan. If revenue keeps growing at anything like today’s pace, by the end of the decade we be generating positive cash flow. What I can’t tell you is whether, by then, a competitor have built something better.”

will · will · should · might / may / could

Notice the shape: a commitment, an arrangement, an expectation, an open possibility. Each future form carries a different level of promise.

Take these into the Speak tab: when you are spending money now for results later, the future perfect is how you make the promise — and the modal is how you protect yourself.

Going deeper

A note on the figures: OpenAI is a private company. Its 2025 accounts and its forecasts were not published by the company; they come from documents reported by the Financial Times, The Information, the Wall Street Journal and others, and the forecasts have been revised several times. Treat every future number here as a projection, not a fact. The case is still moving.

What the case actually tests. The first reading sees a rational scale bet. OpenAI has repeatedly found that bigger models, trained on more compute, are more capable — and that customers pay for capability. Revenue went from a small API business to $13 billion a year in three years; ChatGPT reached hundreds of millions of weekly users. On this reading the losses are what it costs to stay at the frontier, the compute contracts are the only way to secure scarce chips and power, and the investors who keep paying higher valuations are reading the same curve. Altman’s own logic: having too little capacity is the bigger risk.

The second reading sees a financing spiral. The spending commitments — around $1.4 trillion — are a hundred times the 2025 revenue. Each round is priced on growth that has not happened yet, the forecasts of cash burn keep rising, and some internal targets have already been missed while competitors close the gap. On this reading the company must keep raising at higher valuations to fund contracts it has already signed, and the bet is less on AGI than on the next round always being available.

The third point complicates both: the structure. OpenAI is still controlled by a foundation whose mission is AGI that benefits humanity, not returns. In 2023 the board used that power and was overruled within days. In September 2026 Altman cited safety to delay the stock-market listing, and said the company needed to be able to make decisions “not obviously in the interest of our business and our shareholders”. Is that a safeguard that makes the bet more responsible, or a source of uncertainty that investors are pricing badly? The case leaves that open too.

Talk it through:

Rational scale bet or financing spiral? Argue one side with at least two facts from the case — then name the fact that most weakens your side.
If you were a Microsoft shareholder, would you be pleased or nervous about your company’s 27% stake? Why?
Should a company whose stated goal is the benefit of humanity be allowed to raise money from retail investors through banks and funds?
What would you need to see in 2027 to change your mind about this bet?

Sources for the figures in this case (all read in October 2026):
• Founding, mission and $1bn pledge (December 2015): OpenAI, “Introducing OpenAI”. Capped-profit structure, 100x cap, “billions of dollars” quote (March 2019): OpenAI, “OpenAI LP”. Microsoft’s $1bn (July 2019) and the GPT-3 API (June 2020): OpenAI announcements.
• ChatGPT at about 100m monthly users (January 2023): UBS analyst note, reported by Reuters. Microsoft’s multiyear, multibillion-dollar investment (23 January 2023): Microsoft blog; $13bn total commitment: Microsoft 10-Q (October 2025); CNBC.
• Altman’s removal and return (17–21 November 2023): The Guardian; ABC News; AP.
• Stargate (January 2025) and the $1.4tn / “not having enough computing power” statements (October–November 2025): Reuters; TechCrunch; DatacenterDynamics.
• Recapitalisation (28 October 2025): OpenAI, “Our Structure” and “The Next Chapter of the Microsoft–OpenAI Partnership”; Reuters; CNBC.
• 2025 audited accounts: Ars Technica (16 June 2026), citing documents also reviewed by the Financial Times.
• $122bn round at $852bn (31 March 2026): OpenAI; CNBC; WSJ. Missed revenue target and $25bn 2026 burn: The Decoder (28 April 2026), citing the WSJ and The Information. 2030 cash-flow forecast: The Decoder, citing The Information.
• $278bn cash burn 2026–2030: Reuters (18 September 2026), citing the Financial Times. No listing in 2026: Fortune (12 September 2026). $1.4tn valuation talks: Reuters (29 September 2026), citing Bloomberg.

Speak

All of it out loud, against the clock. When the money is spent now and the results arrive later, every audience wants a different promise about the future.

Useful language: by 2030 we will have…, by then we will still be…, we will have been investing for… · we should be…, we may well have…, we might not have… · it’s a projection, not a promise · the risk of too little is greater than the risk of too much · and to close: the question is not how much we will have spent, but what we will have built with it.

Time remaining: 4:00

Task A · Brief the board (4 minutes)

You are advising a board that is considering a huge long-term investment in an unproven technology. Use OpenAI as the case study: the non-profit and its mission → the capped-profit structure → the API → ChatGPT → Microsoft → the November 2023 crisis → Stargate and the compute deals → the recapitalisation → the 2025 accounts → the $122bn round → the forecasts to 2030.

End with where you think OpenAI will be in 2030. Use at least five key terms (cash burn, compute, run rate, valuation, capped return, governance, commitments, break even, recapitalisation, operating loss) and at least two future perfect and two future continuous sentences.

Task B · The hostile question (4 minutes)

You are OpenAI’s chief financial officer at an investor meeting. A fund manager says: “You lost twenty billion dollars on operations last year, you’ve signed up for over a trillion in compute, and your own forecasts keep getting worse. By the time you break even, how much of my money will you have burned — and what happens if the next round isn’t there?”

Answer properly: acknowledge the numbers, explain what the spending buys, give one forecast with the future perfect and one with the future continuous, and mark your confidence clearly. Then do it again in ninety seconds — the compression is the exercise.

Task C · Four rooms, one contract (4 minutes)

Your organisation has just signed a ten-year contract that commits it to spending far more than its current annual revenue. Explain the decision four ways, one minute each:

To the board: what you will have achieved by the end of the contract — and what you will be watching each year.
To the market: why the commitment is a strength, not a risk — in under a minute.
To the workforce: what they will be doing differently next year, and what you cannot promise.
To the press: one sentence about 2030 that you will still be comfortable with in 2030.

Prefer not to use your own organisation? Take one of these instead:

A biotech company that has signed a long manufacturing contract for a drug still in trials.
A fusion-energy start-up that has booked a power-plant site before proving its reactor works.
A streaming service that has committed billions to content before reaching profitability.
A university that has borrowed heavily to build a research campus for a field that does not exist yet.

To finish — the harder questions

Is it possible to price a company whose main product might change the economy — or might be overtaken in two years? What would a sensible valuation method look like?
OpenAI’s board tried to use its power in 2023 and failed. Does that make the mission-based structure meaningless, or did the 2025 restructuring fix the problem?
Who carries the risk if the bet fails: OpenAI’s investors, its partners who built the data centres, or someone else?
Before you go — from memory: make three sentences about OpenAI in 2030: one with will have, one with will be + -ing, one with may well have. Then make the same three about yourself.