The AI bubble bursting isn't the end of AI. It's the price cut.

In the first three months of this year, investors put $300 billion into startups.

Not the year. The quarter. Crunchbase counted it across roughly 6,000 companies, a record, and four cheques took $188 billion of it: OpenAI at $122 billion, Anthropic at $30 billion, xAI at $20 billion, Waymo at $16 billion¹.

Everyone read that as confidence. The size is not the interesting part. The interesting part is the question nobody asks in the same breath: where does the next one come from?

The arithmetic nobody says out loud

Money at that scale does not appear. It is allocated. Every dollar in those rounds was sitting somewhere else last quarter — in bonds, in property, in an index fund, in a pension member's balance in Melbourne. To fund a run rate like this, investors have to either find genuinely new money or sell something else to buy this.

Run the annualisation and the discomfort is obvious. One quarter at $300 billion is more than a trillion dollars a year of allocation, aimed at a sector where the largest companies are not yet profitable. This is not a market-timing call and I have no idea what month it turns. It is simpler than that: it is a supply question. Everybody in that queue is asking for money at the same time, from the same pool, and the pool does not care how good the technology is.

So some of them will not get what they are asking for. That is not a prediction about AI. It is a prediction about queues.

And when a few of them miss, the pricing that everybody downstream has been quietly enjoying stops making sense.

Your token price is somebody else's loss

Here is the part that lands on your invoice.

OpenAI's audited 2025 numbers: $13.07 billion of revenue against a $20.92 billion operating loss². Round it and the company spent something close to $2.60 for every dollar it earned. The headline net loss of nearly $39 billion is mostly a one-time accounting adjustment from its restructure, and quoting it would overstate the case — the operating loss is the honest number, and the honest number is bad enough.

That gap is currently your discount.

The per-token price you are budgeting against is not the cost of serving you. It is the cost of serving you, minus what an investor is willing to absorb to buy the market. It is the cheapest AI will ever be relative to its true cost of production — and every business case built on it inherits the assumption that the subsidy holds.

You already know this, in the way you know things you have not said in a meeting. You have seen the training bills in the news, and then you have looked at your own API line, and somewhere in the back of your head you have wondered how those two numbers live in the same universe. They don't. One of them is being paid by somebody else, for now.

Meanwhile, nobody can find the value

The other half of the squeeze is on your side of the ledger.

PwC asked 4,454 chief executives across 95 countries what twelve months of AI investment had returned. 56% reported neither a revenue increase nor a cost reduction. One in eight — 12% — reported both³.

The easy read is that AI doesn't work. I think that read is incomplete, and expensively so. Look at where the money went. The dominant enterprise motion of the last two years has been the seat: a licence on every desktop, a chat window beside the work, a dashboard counting how many people opened it. That motion produces exactly the result PwC measured — high adoption, no measurable P&L movement — because a person with a faster way to draft an email is not a process that costs less to run.

The value is real. It is one layer below where most companies are looking for it.

Which sets up the squeeze. Keep buying seats and you are building a cost base on subsidised prices for capability that isn't moving your numbers. Freeze everything until the market settles and you hand the compounding to whoever kept going. Both roads cost. The wrong answer is expensive in both directions.

What the last burst actually did

So let's talk about the burst, because the assumption underneath most bubble commentary is wrong.

The Nasdaq peaked at 5,048 in March 2000 and bottomed at 1,114 in October 2002 — a fall of nearly 80%⁴. Everybody remembers that. Far fewer people remember what happened underneath it.

Before the crash, telecom companies raised $1.6 trillion on Wall Street and floated another $600 billion in bonds to lay fibre across the United States — 80.2 million miles of it, 76% of all digital wiring installed in the country to that point⁴. Then the market repriced them. Companies died. Investors were wiped out. It was, by any market measure, a catastrophe.

The fibre stayed in the ground.

The glut it left behind meant that by 2004 the cost of bandwidth had fallen by more than 90%, even as internet usage kept doubling⁴. The next wave of companies — the ones running video, cloud and social at prices that would have been impossible in 1999 — were built on infrastructure someone else had already paid for and lost money on.

The crash did not delete the internet. It transferred it. It moved the capability from the people who financed it to the people who were ready to use it cheaply.

That is the actual history, and it is the opposite of the lesson usually drawn from it.

The obvious move, and why it fails

The instinctive response to everything above is to wait. Let the correction happen, see what survives, buy in at the bottom with a clear head.

Waiting until the dust settles is not a strategy. The dust is where the discount is.

Because the thing that gets cheap in a correction is compute and talent. The thing that does not get cheap — that takes the same six to eighteen months whatever the market is doing — is knowing precisely where in your own operation a machine judgment belongs. Companies that arrive at the bottom of the cycle with that map spend the cheap tokens immediately. Companies that arrive with a licence list start the mapping work then, and finish two years later, buying at whatever the price has become by then.

Small gap now. Uncatchable in three years.

There is a fair objection here, and it is worth stating properly: the price of compute was falling before any of this and will keep falling without a crash to help it. Open-weight models have already commoditised a lot of what the frontier labs charged a premium for, and underneath them the chips, the power draw, the cooling and the hardware lifespan all sit on curves that have bent downward for decades. Richard Sutton gave the dynamic its name in 2019 — the bitter lesson, that general approaches win precisely because they ride the falling cost of computation⁵. If cheap intelligence is arriving regardless, why hurry?

Because falling prices are an argument for the map, not against it. Cheap compute is only worth something to a company that knows where to point it, and that knowledge is the one input in this whole story with no downward cost curve — it takes the same months of process work whatever a token costs. Meanwhile the bills are real today: at current, subsidised prices, badly shaped usage — a model re-reading everything on every transaction, an agent looping because nothing told it to stop — produces numbers that make CFOs flinch. The waste is in the shape, not the price, and a ninety per cent price cut on a wasteful shape is still waste. Prudence now is not scepticism about where this goes. It is what keeps you solvent enough to be there when it arrives.

Judgment points, not desktops

Here is the reframe the whole piece rests on.

For almost every role in your company, the valuable form of AI is not a language model on a desktop. It is a discrete judgment point dropped into a process that was previously too hard to automate.

Think about the processes you have never automated. They were never blocked by the mechanical steps — the data movement, the form filling, the routing. Those have been automatable for twenty years. They were blocked at one or two moments in the middle where somebody had to read something and decide: does this invoice match this contract, is this complaint about the product or the delivery, does this application need a human, is this photo showing the damage the claim describes.

One judgment, mid-process, was enough to keep the whole thing manual.

That is what changed. Not "AI can do knowledge work" — that framing is what sold the seats. The specific, boring, enormous change is that a bounded judgment can now be made by a machine, cheaply, at any point in a workflow, with a confidence signal attached and a human escalation behind it.

And the economics of that shape are completely different from the seat. A judgment point is a small, predictable, measurable number of tokens per transaction — one call on one bounded question with a defined input. A chat window is an unbounded conversation with no unit of output, priced per person per month whether or not anything changes. When token prices rise to something like their real cost — and they will, because $2.60 spent per dollar earned is not a business model — a judgment point re-prices from trivial to slightly less trivial. A seat estate re-prices into a board conversation.

Which brings us to the people, because this is where most programmes lose their nerve or their soul.

When you put a judgment point into a process, hours come out. If those hours have no named destination they become invisible slack, the dashboards stay green and nothing improves. The move that works is to redeploy the freed people into the parts of the business that cannot be automated and are currently under-served: the difficult customer conversation, the bespoke work you turn away, the follow-up nobody has time for, the quality pass, the relationship. That is craft. It is the layer your competitors cannot copy from a vendor, and the only durable answer to the question of what your people are for.

Judgment points take the cost out. Craft puts the difference in. One without the other is either a redundancy programme or a hobby.

The Judgment-Point Map

So take one process. Not a department, not a transformation — one end-to-end process you could describe to a stranger in a minute. Claims triage. Onboarding a client. Month-end close. Quoting a job.

Write down its steps in order, and label every one of them with exactly one of three words.

Conveyor. Mechanical steps: moving data, filling fields, routing, chasing, formatting, reconciling against a rule. No judgment required. No AI belongs in this column — these are rules, scripts and integrations, the boring deterministic automation software has done reliably for twenty years, at a fraction of a token's cost and with the same answer every time. If you have not automated these, AI was never your blocker, and a process that is mostly conveyor has a cheaper answer than tokens.

Judgment point. The steps where somebody reads something and decides, and where the decision is bounded — a classification, a match, an exception flag, a summary that feeds the next step. The test is that you could write down what a right answer looks like, and check it after the fact.

This column is the only place AI goes, and that restriction is the entire economic argument. One bounded call, on one question, at one step — not a model sitting across the whole process reading everything and charging you for the privilege. A judgment point costs a predictable fraction of a cent per transaction because you have told it exactly what to decide. That is the difference between a token bill you can forecast per unit of work and a token bill that scales with how chatty your staff are. And each judgment point carries an escalation: when the model's confidence is low, the case goes to a person. The machine takes the ordinary volume, the human keeps the hard cases, and you pay full human attention only where it changes the answer.

Craft. Steps that carry relationship, negotiation, accountability, taste or a judgment nobody can write down. These are not "not yet automated" — they are the reason customers choose you.

And they are more than the escalation queue for the column beside them. Catching the cases the model wasn't sure about is the floor of what your people do here, not the job. The job is the work you currently cannot deliver enough of: the difficult conversation, the bespoke request you turn away, the follow-up nobody has time for, the quality pass that never happens at month-end. This is where the hours freed by the judgment points get redeployed — and if you cannot name that destination, you have not finished the map.

Then answer one question underneath the map: when this process gets faster, where do those hours go, and who moves them? A name and a step, not an aspiration.

One note on what the map is worth once you have it, because it changes where the money goes. A mapped process is a process you can now build — and building it is where your tokens should be concentrated, not spread across a licence estate. Agentic tooling has made the construction work cheap enough to be worth doing properly: reading the existing system, writing the integration, wiring the exception paths. That spend is large, deliberate and once. What you are left with afterwards runs the way processes have always run — deterministic steps at effectively no cost, calling out to a model only at the judgment points, only when it reaches one. Front-loaded build, cheap run, for years. That is a longer argument than this piece can carry, and I will make it properly another time; for now it is enough to know that the map is what makes the build possible, and the build is what makes the token price stop mattering.

The map ends in one of three words.

• Positioned — you have named judgment points and a craft destination for the freed hours. You know exactly what you buy when compute gets cheap, and exactly where the people go. Protect this map; it is worth more than your licence budget.
• Half-mapped — you have found the judgment points but not named where the hours land. You will automate successfully and capture nothing, because saved time with no destination is the most reliably wasted asset in business.
• Unmapped — no judgment points identified. Either the process is pure conveyor and wants ordinary automation at a fraction of the cost, or it is pure craft and should be left alone and resourced properly. Both are useful answers, and both save you a year of pointing AI at the wrong work.

One boundary, honestly: this map cannot tell you when the correction comes, and it cannot price your tokens. Nobody can do the first and no vendor will do the second. What it guarantees is that whenever the repricing happens — up or down — you already know which parts of your operation want compute and which parts want people. That knowledge is the thing that stays expensive when everything else gets cheap.

The winter is the opportunity

The bubble bursting will be read as a verdict on the technology. It won't be. It will be a verdict on the capital structure that funded it — on a queue where more people asked for money than could be paid, in a single year, at prices that assumed everyone got theirs.

Underneath that, the fibre stays in the ground. The models don't get worse when the share price falls. The data centres do not switch off. The chips, the weights, the tooling, the people — all of it survives the correction and most of it gets cheaper, exactly as bandwidth did, exactly as it did for every company that walked into 2003 knowing precisely what it wanted to build.

The companies that come out of this owning their markets will not be the ones that predicted the timing. They will be the ones that spent the noisy months mapping judgment points, moving people into craft, and building an operating model whose token spend is a per-transaction line they can defend, rather than a per-person line they cannot.

By the end of thirty minutes with one process, you will have every step labelled, a named destination for the hours, and one word for the meeting where somebody finally says "bubble" out loud.

Start with the middle of the process. Find the one step where somebody reads something and decides — that step is the reason the whole thing is still manual, and it is the cheapest thing you will automate this decade.

References

Footnotes

1. Crunchbase News. (2026, April 1). Q1 2026 shatters venture funding records as AI boom pushes startup investment to $300B. https://news.crunchbase.com/venture/record-breaking-funding-ai-global-q1-2026/ ↩

2. TechSpot. (2026, June 17). OpenAI made $13 billion in 2025 and lost $21 billion doing it. https://www.techspot.com/news/112794-openai-made-13-billion-2025-lost-21-billion.html ↩

3. Fortune. (2026, January 19). 56% of companies getting nothing out of AI, PwC research says; chairman blames forgetting the basics. https://fortune.com/2026/01/19/pwc-global-chairman-mohamed-kande-ai-nothing-basics-29th-ceo-survey-davos-world-economic-forum/ ↩

4. McCullough, B. (2018, December 4). An eye-opening look at the dot-com bubble of 2000 — and how it shapes our lives today. TED Ideas. https://ideas.ted.com/an-eye-opening-look-at-the-dot-com-bubble-of-2000-and-how-it-shapes-our-lives-today/ ↩ ↩² ↩³

5. Sutton, R. (2019). The bitter lesson. Summarised at https://en.wikipedia.org/wiki/Bitter_lesson — "General approaches that scale with available computational power tend to outperform ones based on domain-specific understanding because they are better at taking advantage of the falling cost of computation over time." ↩