BUSINESS ENGLISH · C1 · 50 min

The Race Nobody Can Leave

In 2001 Eli Lilly lost its patent on Prozac and the money vanished almost overnight. In 2026 the world’s AI companies face the same cliff every few months — which may explain why the people most worried about AI keep building it faster.

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

Today: you’ll test one of the oldest ideas in business — the patent cliff — on the most expensive race in the world today, and ask why nobody in it can simply stop; and you’ll practise the formal language of scenarios: were we to stop…, should prices fall…, had we acted sooner…

Before we start. Talk these through out loud.

Discuss:

Have you ever kept doing something at work — a project, a price cut, a long-hours culture — mainly because your competitors were doing it?
Think of a product you once paid full price for that is now cheap or free. Who lost money when that happened, and who gained?
If everyone in your industry agreed to slow down, who would be the first to break the agreement — and why?

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. The drug brought in over two billion dollars a year. Analysts warned that when its protection ended in 2001, the company would face a patent cliff.

a patent cliff (n) — the sudden fall in revenue when a patent expires and competitors can legally copy the product. Most used in the pharmaceutical industry: to fall off a patent cliff, to prepare for a patent cliff.

2. Within weeks of the expiry, pharmacies were selling cheaper versions made by other firms, and most prescriptions switched to generics.

a generic (n / adj) — a copy of a medicine sold under its chemical name, not a brand, once the patent has expired. Generic competition, a generic manufacturer. The word is now used for any cheap copy that competes with an original.

3. While its model was the best on the market, the company could charge almost whatever it liked. That pricing power lasted about three months.

pricing power (n) — a company’s ability to raise or hold prices without losing customers. Warren Buffett famously called it the single most important thing when judging a business. Its opposite is being a price-taker.

B · Which term means…?

Three terms: open-weight · valuation · the frontier. Say the right one, then reveal.

1. Which term means the most advanced point that a technology has reached so far?

the frontier (n) — in AI, frontier models and frontier labs are the most capable systems and the companies that build them. To push (forward) the frontier, to be at the frontier.

2. Which term describes an AI model whose internal settings are published, so anyone can download it, run it and change it?

open-weight (adj) — an open-weight model. Close to open-source, but more precise: the trained model is shared, not always the data or code used to build it. Many of the strongest open-weight models come from Chinese companies.

3. Which term means what investors believe a whole company is worth, usually based on what they expect it to earn in future?

a valuation (n) — a sky-high valuation, to justify a valuation, a valuation of $300 billion. Verb: to value a company at. A valuation is a bet on the future, not a measure of the present.

C · Choose the meaning — a / b / c

1. Once five firms could make the same memory chip to the same standard, the chip had been commoditised, and buyers chose on price alone. A commoditised product is one that…
(a) is protected by law from copying (b) is hard to tell apart from its competitors, so it sells mainly on price (c) is produced only in very small numbers

(b) — to commoditise (v, US: commoditize), noun commoditisation. From a commodity — oil, wheat, copper — where one supplier’s product is much like another’s. Every premium product fears it.

2. The minister said the country would not cut its own fishing fleet unilaterally while its neighbours kept fishing. To act unilaterally is to act…
(a) slowly and in stages (b) in secret (c) alone, without an agreement with the other parties

(c) — unilaterally (adv), adjective unilateral: a unilateral decision. The opposites are bilateral (two parties) and multilateral (many parties).

D · Collocation & word building

1. Two superpowers, each spending more because the other is spending more, are caught in an arms ___. Complete the phrase, then say what it means when business journalists use it about technology.

an arms race (n) — originally military; in business, any competition where each side must keep spending just to keep up, whether or not the spending makes sense. A spending arms race, to escalate an arms race.

2. The bonus system rewarded speed, not safety, so the engineers had every incen___ to cut corners. Complete the noun, then give the verb and the adjective.

incentive (n) — a strong incentive to, perverse incentives (ones that reward the wrong behaviour). Verb: to incentivise. Adjective: incentivised. Economists often explain behaviour by asking one question: what are the incentives?

Now use them — out loud

Before we read:

Using the frontier, pricing power, commoditise and valuation, predict why an AI company might feel it cannot slow down.
Which of these terms would you use in a board paper, which over coffee with a colleague — and which could sound like an excuse if a company used it to explain its own behaviour?

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 — Fourteen years, then the cliff

Prozac arrived in American pharmacies at the end of the 1980s and became one of the best-known medicines in the world. For its maker, Eli Lilly, it was a remarkable business. A new drug can cost hundreds of millions of dollars to develop, but once it is approved, a patent protects it from copies. By the early 2000s Prozac was bringing in around two and a half billion dollars a year — at its peak, some six or seven hundred million a quarter, and roughly a quarter of the company’s sales.

In August 2001 the patent expired. Generic manufacturers were ready, and their versions went on sale at a fraction of the price. The effect was almost immediate: within weeks, most prescriptions had moved to the cheaper copies, and Lilly’s Prozac revenue collapsed. It never came back. The drug had enjoyed about fourteen years of protected sales; after that, it was a commodity.

For the pharmaceutical industry this was not a surprise but a law of nature. Everyone knows the date their patents end, and the whole business model is built around it: earn back the investment while the protection lasts, and have the next drug ready before the cliff. And the same story helps explain something much newer — why the companies building artificial intelligence cannot stop.

Quick check — from memory, then reveal:

1. Why could Lilly charge high prices for Prozac for so long?

2. What happened after August 2001, and how quickly?

1. Because the patent stopped other companies from making and selling copies of the drug, so Lilly had no direct competition.

2. The patent expired, cheaper generic versions went on sale, and within weeks most prescriptions had switched to them; Lilly’s Prozac revenue collapsed and never recovered.

Discuss:

What protects the prices in your industry — patents, brands, relationships, regulation, habit? How long would that protection last if a cheaper copy appeared?

Part 2 — A patent that lasts three months

The leading AI companies — the frontier labs — work a little like drug makers. They spend billions to build a product that is, for a short time, scarce: the most capable model in the world. While it is the best, they have pricing power. The difference is that their protection does not last fourteen years. It lasts three or four months.

Take Anthropic’s Claude Opus 4.7, released in April 2026 as a state-of-the-art model. On one simplified calculation, intelligence at that level cost around four dollars per million tokens (the units in which AI use is measured), and the price stayed there for three or four months. Then cheaper open-weight models, many of them from China, reached a similar level. Within a few months the price for the same level of intelligence had fallen by about three quarters, to roughly one dollar. No patent had expired; the competition simply caught up.

The effect is visible in what customers use. On OpenRouter, a website that lets developers choose between many AI models, the share of European usage going to open-weight models rose, by one estimate, from about 16% to about 60% in a single year. Most people never notice: they use familiar chat products. But underneath many popular apps are the cheap “generics”. For the frontier labs, and for investors who have given them some of the highest valuations in business history, there is only one answer — keep moving the frontier forward, so there is always something new that the copies cannot yet match.

Quick check — from memory, then reveal:

1. In the comparison with medicines, what is the AI equivalent of a patent expiring?

2. What happened to the price of Opus 4.7-level intelligence, and why?

1. Cheaper open-weight models (the “generics”) reaching the same level of capability — which happens after three or four months, not fourteen years.

2. It stayed around four dollars per million tokens for three or four months, then fell by about 75% to roughly one dollar, because open-weight competitors caught up.

Discuss:

How long does an advantage last in your field — a new product, a new process, a good idea? What do you do when it runs out?

Part 3 — The trap

Here is the problem. The frontier, where the labs must keep pushing to justify their valuations, is also where the biggest risks are. Several of the people running these companies have said publicly that advanced AI could be dangerous. So the obvious question is: why don’t they just stop?

In September 2026 Anthropic’s chief executive, Dario Amodei, published a long essay. He did not call for a pause, and he did not say his own company would slow down on its own. Instead, he argued that the pace of the whole frontier needed limits — speed limits for everyone — and called for strong economic restrictions on China, to buy time to make AI safe. But any deal with China would be negotiated by President Trump, whose view is very different. “Whoever wins AI wins,” as he put it, describing a race he wants the United States to keep leading.

There is a final complication. The race is not only dangerous; it is also enormously useful. Cheap models are what allow start-ups, and countries without a giant AI company of their own, such as the UK, to build with AI at all. They may want safety — but not a clear winner, and perhaps not even a final settlement, because they gain so much from the competition. So every player may be thinking the same thing: I may not like where this race is going, but I don’t want to lose, and I really don’t want them to win. The question is still open: can anyone leave the race — and if not, who could change its rules?

Quick check — from memory, then reveal:

1. What did Amodei call for — and what did he not call for?

2. Why might smaller countries and start-ups not want the race to end with one clear winner?

1. He called for limits on the pace of the whole frontier (speed limits for everyone) and strong economic restrictions on China; he did not call for a pause, and did not say Anthropic would slow down on its own.

2. Because the competition keeps prices low and gives them cheap, capable models to build with; a single winner could remove that benefit.

Discuss:

Have you ever been in a situation at work where everyone privately agreed something was a bad idea, but nobody could be the first to stop? What would have broken the pattern?

Grammar — formal scenarios

Strategy is built on scenarios: what happens if a competitor moves, if prices fall, if we had acted sooner. In speeches, board papers and interviews, senior speakers often drop if and put the verb first. It sounds measured and deliberate — the voice of someone who has thought the scenario through. There are only three patterns to learn.

1 · Were + subject + to-infinitive — an unlikely or imagined future
• If the regulator were to block the merger… → “Were the regulator to block the merger, the share price would fall.”
• With be alone: “Were the airline profitable, nobody would be asking these questions.”
• Negative: “Were it not for the government loan, the shipyard would have closed.”

2 · Should + subject + bare verb — a real possibility, stated politely
• If the supplier misses the deadline… → “Should the supplier miss the deadline, we will switch to the back-up factory.”
• Very common in contracts, emails and customer letters: “Should you have any questions, please contact…”
• Usually followed by will, an imperative, or a present tense — the scenario is possible, not imaginary.

3 · Had + subject + past participle — a past that did not happen
• If we had hedged our fuel costs… → “Had we hedged our fuel costs, we would not be in this position.”
• Negative: “Had it not been for the strike, the ship would have arrived on time.”
• Watch out: no contractions and no not straight after the subject in the question-like form. Say Had we not… — never ~~Hadn’t we…~~ (that is a real question).

1. Drop the if

Rewrite the beginning of each sentence without if. Write the missing words — the rest of the sentence stays the same.

1. If the patent expired early, our revenue would collapse. → early, our revenue would collapse.

2. If we had seen the price data sooner, we would have changed our plan. → the price data sooner, we would have changed our plan.

3. If a customer asks for a refund, the team will process it within two days. → for a refund, the team will process it within two days.

4. If it weren’t for cheap competitors, prices would still be high. → cheap competitors, prices would still be high.

1. Were the patent to expire 2. Had we seen 3. Should a customer ask 4. Were it not for

In 1, Should the patent expire is also possible, but it makes the scenario sound more likely than the original if … expired … would. In 4, never ~~Weren’t it for~~ — the not goes after it.

2. Which pattern?

Complete each gap with the correct form of the verb in brackets. Decide first: imagined future, real possibility, or past that didn’t happen?

1. Had Lilly (find) a replacement drug in time, its 2001 results would have looked very different.

2. Were every lab (agree) to slow down, at least one would still have a reason to break the agreement.

3. Should prices (fall) again this year, the company will need a new way to explain its valuation.

4. Had it not (be) for open-weight models, the start-up could never have afforded to launch.

1. found 2. to agree 3. fall 4. been

1 and 4 are the past that didn’t happen (had + past participle). 2 is imagined (were + to-infinitive). 3 is a real possibility (should + bare verb — no ~~falls~~, even though prices is the subject).

3. The answer on stage — complete it, then deliver it

The head of an AI company is asked at a conference: “If you think this is dangerous, why don’t you stop?” Complete the answer with the words in brackets, using inversion each time. Then deliver it aloud, slowly, as if the whole room is filming you.

“It’s a fair question, so let me answer it directly. (we / stop) tomorrow, the work would not stop with us — it would simply move to companies with fewer safeguards. (a binding international agreement / appear), we will sign it on the first day. (governments / act) five years ago, we might not be having this conversation at all. And (it / not / be) for the competition, prices would be four times higher — so let’s not pretend this race brings nobody any benefit.”

Were we to stop · Should a binding international agreement appear · Had governments acted · Were it not

Now give the same answer with if in every sentence. It is just as correct — but listen to how much more defensive it sounds. Notice too that this is one side of the argument: the Going Deeper tab asks whether “if we stop, someone worse takes over” is a reason or an excuse.

Take these into the Speak tab: a scenario stated with inversion sounds calm and considered — useful under pressure, but a listener may also hear it as a way of avoiding a straight answer.

Going deeper

A note on the figures: sales, prices and usage shares are rounded, and the AI prices and usage shares are simplified estimates from public data; this case is developing week by week.

What the case actually tests. The first reading is economic, and it is powerful. Nobody needs to be reckless for the race to continue; the incentives are enough. A frontier lab that pauses does not make the frontier pause — it just hands its customers, its staff and its investors to someone else, while the open-weight “generics” eat its prices from below. On this reading, blaming individual companies misses the point, and the only real solution is the one Amodei proposed: rules that apply to everyone at once, because nobody can afford to slow down alone. It is the classic structure economists call a prisoner’s dilemma.

The second reading is more sceptical. The Prozac comparison is neat, but medicine is one of the most regulated industries in the world: a drug cannot be sold until it is proven safe, however fierce the competition. The lesson of pharmaceuticals may be the opposite of “you can’t stop” — it may be that a race can be run safely when there is a referee. The sceptic also notices who benefits from the trap story. “We would stop, but the others won’t” is a convenient argument for anyone who wants to keep going, and asking governments to restrict a foreign competitor is not only a safety policy; it is also a commercial one.

The third point is the one most often missed. The race is not simply a danger to be managed; it is also a huge transfer of value to everyone else. Cheap models let a small company in Lisbon or Leeds do what only a giant could do two years ago. That is why the middle powers are divided: they want safety, but a single winner — American or Chinese — could end the price war that benefits them. So the case asks a harder question than “why don’t they stop?” It asks who actually wants the race to end, and on whose terms.

Talk it through:

Trap or excuse? Take one side and argue it with at least two facts from the case — then name the one fact that most weakens your side.
Is there a race in your own industry that everyone would quietly like to slow down — on price, on speed, on working hours? What kind of agreement could actually hold?
Should a country like the UK want one clear winner in AI, or a long, expensive competition? Argue it as a finance minister, then as a safety regulator.
Drug makers get about fourteen years; AI labs, by this calculation, get a few months. What happens to a business — and its people — when its advantage lasts only one quarter?

Speak

All of it out loud, against the clock. A race you cannot leave has to be explained to people with very different fears — and scenarios are the language for it.

Useful language: were we to…, should … happen, had we… · were it not for… · the incentives point one way · nobody can afford to move first · the price has fallen by roughly three quarters · and to close: the question is not whether…, but who…

Time remaining: 4:00

Task A · Brief the board (4 minutes)

Present this case to a board that is about to spend heavily on AI. Structure it: how a patent protects a drug → Prozac and the 2001 cliff → the frontier labs and their pricing power → Opus 4.7 and the three-month “patent” → open-weight models and the European shift → why the labs must keep pushing → the safety problem at the frontier → Amodei’s essay and the political reality → the middle powers → the open question.

Finish with one scenario the board should plan for. Use at least five key terms (patent cliff, generic, pricing power, the frontier, open-weight, valuation, commoditise, unilaterally, arms race, incentive) and at least three inverted conditions.

Task B · The hostile question (4 minutes)

You lead a frontier AI company and are giving evidence to a parliamentary committee. A member says: “Your own leaders say this technology could be dangerous. You keep building it faster. Either you don’t believe the warnings, or you are putting profit before safety. Which is it?”

Answer properly: accept the tension, explain the economics in numbers, say clearly what would make you slow down, and state one scenario with inversion. Then do it again in ninety seconds — the compression is the exercise.

Task C · Four rooms, one price war (4 minutes)

Take a market where a premium product is suddenly being copied cheaply — use one of the examples below or one you know — and explain the situation four ways, one minute each. Notice which room needs scenarios, and which room just wants a straight answer.

To investors: why the high price will not last, and what the company will do before the cliff arrives.
To the engineers: why the next product has to be ready in months, not years — and what that means for their workload.
To a regulator: why you would accept rules that apply to everyone, but not rules that apply only to you.
To customers: why the cheaper copy is not the same — or honestly, when it is.

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

A drug company whose best-selling medicine loses its patent next year.
A maker of premium electric cars facing much cheaper imports.
A software company whose paid tool is now copied by a free open-source version.
A designer furniture brand whose best-known chair is sold in cheap copies online.

To finish — the harder questions

Is it fair to blame a company for staying in a race that any company in its position would stay in? Argue one side properly.
The pharmaceutical industry is fiercely competitive and heavily regulated at the same time. Could the same model work for AI — and who would be the referee?
Would you rather live in a world with one dominant AI company, or many competing ones? What does each world get right and wrong?
Before you go — from memory: describe the same scenario three ways — with were … to, with should, and with had. Use the labs / agree to slow down as your first example, then choose one from your own industry.