Every boom has a favourite number. For AI it has been the size of the next deal: $40 billion raised, $300 billion of cloud capacity ordered, “up to” $100 billion pledged by a chipmaker to its best customer. For two years, each new figure made the last one look sensible.
Then came last week. On Thursday Jensen Huang, Nvidia’s founder and the public face of the AI build-out, stepped down as chief executive following a board investigation and damaging reports about his private life. The shares dropped more than 13% at the open and ended the day 4.3% lower. A day later the selling reached the cloud companies that run on Nvidia’s chips.
Markets can absorb a change of chief executive. What they are struggling with is something else: under Huang, Nvidia was not only the industry’s main supplier. It had also become one of its main sources of money.
01Five days in November
02The shovel seller who became a lender
The old line about gold rushes is that the surest money is made selling shovels. Nvidia has been the shovel seller of AI, and a very successful one.
But over the past year it began helping its customers pay for the shovels. It bought stakes in AI labs and cloud start-ups. In August it joined Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR in a plan to raise more than $500 billion for AI infrastructure. Fortune reports it also discussed a guarantee of roughly $250 billion so that OpenAI could lease a data-centre campus in Ohio. The Economist had a name for this: “the central bank of AI”.
Nvidia’s own accounts show how concentrated the business is. Three customers supplied 44% of its revenue in the first half of its 2027 financial year. When doubts arise about Nvidia, they travel quickly to everyone it supports.
03A map of the money
Bloomberg drew a well-known chart of these relationships last year. Below is our own version, updated for 2026.
Look at how many arrows come in pairs. Nine companies on this map buy Nvidia’s chips, and seven of them have received Nvidia’s money. Nearly every other line passes through OpenAI. It is a system that works well while everyone is growing. It has not yet been tested by anyone shrinking.
04What Huang said about it
Huang never accepted that this was circular. Speaking at a Goldman Sachs conference in September, he said Nvidia puts in modest sums, wins far bigger orders in return, and backs only customers with real contracts. Much of that is reasonable.
A few weeks later in Berlin, though, he described the other side of the problem. Demand, he said, was “off the charts”. The shortage was money:
“The demand of the capital is much greater than the availability of the capital.”
Jensen Huang, October 2026
Customers who want a product but cannot fund it are the reason suppliers start lending in the first place.
05OpenAI: growing fast, spending faster
OpenAI’s sales are not in doubt. Revenue rose by about half in one quarter to roughly $10 billion. Over the same three months it lost $10.9 billion on operations, a little more than it took in.
Its promises are larger still. Across Oracle, Microsoft, Amazon, CoreWeave and the chipmakers, OpenAI has agreed to buy about $1.4 trillion of computing over the coming years. Its revenue is running at about $50 billion a year, which makes the commitments roughly 28 times its annual sales.
Supporters point to Amazon, which lost money for years before it paid off. The comparison has a weak spot. A warehouse stays useful for decades. An AI chip can be overtaken in two or three years.
06So who pays?
Not the investors, in the end, and not Nvidia or the banks. The bill is settled by ordinary customers: firms and individuals who decide an AI product is worth what it costs. If there are enough of them, today’s spending will be remembered as bold. If there are not, the weak points are easy to list:
- AI subscriptions that cost businesses more than they save.
- Price competition that cuts revenue faster than costs fall.
- Newer chips that make expensive hardware old before it has paid for itself.
- Investors who tire of funding losses.
07Bubble, or just a bad week?
The dot-com years offer the closest parallel. Investors then were right about the internet and still lost heavily, because they paid prices that only perfect outcomes could justify. The technology succeeded. Many of the valuations did not.
There are good reasons not to panic. Every AI company still needs Nvidia’s chips. After the sell-off its shares cost about 12 times expected earnings, the lowest since 2012. OpenAI is growing quickly. A single rough week does not make a crash.
What the week did show is how closely these companies are tied together. A shock at the chipmaker passes to the cloud providers, then to their lenders, then to the AI labs waiting for capacity.
08What to watch next
- 17 November: Nvidia reports third-quarter results, its first under Colette Kress.
- The CEO search: who takes over, and whether Nvidia goes on financing its customers.
- Funding: whether OpenAI finishes its round, and what it costs CoreWeave and others to borrow.
AI will very likely prove useful. That was never the real question. The question is whether the price paid for it was right, and last week the market started asking it out loud.
— Shubham Agrawal
This is an opinion article based on public reports as of 14 November 2026. It is for education only and is not investment advice.
Sources: Nvidia SEC filing, Nvidia board statement, CNBC, Fortune, Reuters, 13 Nov, Reuters, OpenAI valuation, Euronews, TechCrunch, Huang at GTC Berlin, The Economist, HPCwire, investingLive, Forbes, OpenAI commitments, OpenAI revenue.