- The 2 cases, defined
- The clock test
- Startup culture shows the same split
- If it breaks, what recovers
- The 2 lines to write this week
The founder of the AI hardware boom left his job this week. Jensen Huang resigned. Nvidia now has an avatar problem covers the exit itself. The exit starts a harder question, and the question lands on every owner’s desk at once. Is the AI boom a bubble, and does the answer change what a business should do this quarter? This piece lines up both cases with their best facts. It picks neither. The decision it asks for is smaller, and closer to home.
The 2 cases, defined
A bubble is a price story that runs ahead of its customers. Prices climb because the next buyer will pay more, and the product underneath stays thin. A boom is a build-out where customers arrive while the concrete is still wet. Prices can still fall inside a boom. The work underneath keeps its value after they fall. The AI fight is over which of those 2 stories fits the facts. Here is the honest scoreboard.
The table below sets the bubble case against the boom case on the same 5 tests. It is the comparison every AI bubble argument turns on.
| Test | Reads as a bubble | Reads as a boom |
|---|---|---|
| Prices | AI stocks trade at 31 times forward earnings against 19 for the wider index, near dot-com levels. | The central bank study behind the warning also notes the prices can hold if the forecast growth lands. |
| The money | The largest headline deal in the sector, a $100 billion Nvidia pledge to OpenAI, stayed a letter of intent and never became a signed contract. | The build-out so far ran mostly on the cash flow of the builders, not on borrowed money. |
| Revenue | Estimates of the yearly gap between AI spending and AI revenue run into the hundreds of billions. | OpenAI told investors its run rate neared $50 billion a year by September 2026, and the 2 largest clouds grew 39% and 48% in that quarter. |
| Customers | AI stocks now make up 44% of the S&P 500, so a handful of firms carries the whole index. | Use is broad and climbing: 19.2% of Canadian businesses ran AI in production in 2026, triple the 2024 share. |
| The debt | The Bank of Canada names AI 1 of its 3 key financial risks, on concentrated value and debt-funded building. | Debt is growing from a small base, and the big builders held or raised their 2026 spending plans at the autumn results. |
Read the table twice and a pattern shows. The bubble case argues about prices and promises. The boom case argues about customers and cash. Both can be true in the same market, in the same quarter, at different layers of the stack.
The clock test
Booms and manias also differ on the clock, and the AI bubble debate keeps skipping that test. The NFT run shows the mania pattern at full speed. Trading peaked in January 2022 at about $6 billion in 1 month. By August 2022, the largest marketplace traded 99% below its best day. Inside 20 months, the bulk of the fall was done. Trading for the full year 2025 came to about $5.5 billion, near a tenth of the peak year.
AI runs on a longer clock. The boom dates from the public launch of ChatGPT on November 30, 2022, which makes it 48 months old this month. Corporate AI investment reached $581.7 billion in 2025, more than double the year before. Manias burn fast because the product is the price. Build-outs burn slow because the product takes years to pour, wire and fill. At 48 months, the spending still rises. That fact fits the boom story. It also fits a bubble in its late chapters, and the clock alone cannot settle the argument.
Startup culture shows the same split
Startup culture shows the split inside 1 market. AI startups took more than half of all venture funding in 2025, the first time 1 field has held the majority. In the first half of 2026, 2 companies, OpenAI and Anthropic, raised 43% of all startup funding globally. At the same time, startup shutdowns ran at a record pace in the first half of 2026. The median AI startup that closed still held $30,000 in the bank.
Character.AI counts 163 million current and past users in its court filings, and it raised more than $1 billion. It filed for court protection in October 2026 and sits in a supervised sale. The top of this market has never held more money. The middle of it has rarely been thinner. Both sentences describe the same culture in the same year.
If it breaks, what recovers
History gives 1 clean test for the recovery question. In the dot-com bust, Amazon’s shares fell about 94% from peak to trough. The shares needed about 10 years to pass the old high. The business grew revenue through every year of that fall, and it runs the largest cloud service today. Even the NFT market kept a working residue. A major museum now holds its best-known collection, and tokenized trading cards still clear real volume on the rails the bust left behind. The pattern across busts is steady. Prices recover last. Firms with paying customers recover first, and firms priced on the next buyer do not recover at all.
Canada holds a concrete stake in the outcome. Alberta’s grid operator counts about 19,565 MW of data centre requests. Canada’s data centre map: what is built, what is blocked, what is waiting tracks the projects behind the number. Concrete poured for a real customer base keeps its value in either story. The AI buildout runs on private credit, so ask your vendor how long its money lasts makes the same point from the buyer’s side. The length of a vendor’s money is a fact you can check today.
The 2 lines to write this week
A leader cannot settle the bubble argument, and does not need to. Settle the smaller question instead. This week, take every AI tool your firm pays for and write 2 lines beside each one. Line 1 names the customer task the tool serves. Line 2 names your cost if that vendor closed inside 90 days. A tool with a clear line 1 survives any market, because your customers fund it. A tool with a blank line 1 is a position, and positions are what a bust punishes first. Which of your tools would you rebuild first if the market broke next quarter? That answer is your real AI strategy, in boom or in bust.
