The men who sell the chips say the thing holding back artificial intelligence may no longer be the chips.
Broadcom chief executive Hock Tan and Intel chief executive Lip-Bu Tan, speaking October 12 at a DBS forum in Singapore, offered the same read on the industry they supply: demand for computing still runs ahead of supply, but the tightest constraint is migrating away from wafers, memory and electricity toward capital.
Hock Tan said financing may do more to set the pace of expansion than either chips or power. That is an unusual complaint from a man whose company earns its living selling the accelerators and networking gear that fill data centers.
He also offered a blunt description of the investors funding the boom, saying many are betting as though they were in a casino and do not know what they are buying. He did not place OpenAI in that category. His objection was not that the company is overvalued but that investors have set expectations so extreme that no result can satisfy them.
Lip-Bu Tan pointed to the rising cost of money, and to a question he expects to arrive with force: when does the spending turn into a return? The two men were describing the same squeeze from opposite ends of the supply chain.
Their vantage points give the warning weight. Broadcom designs the custom accelerators and switch silicon that large cloud operators use to train models. Intel builds the processors that run the rest of the data center and is trying to rebuild a foundry business on the strength of that demand. Both profit from the build-out, which makes a caution about its financing harder to dismiss as self-interest.
The industry has spent three years arguing about physical limits. First the shortage was of logic chips, then of the memory that sits beside them, then of the electricity and transformers needed to run them. Each was resolved or routed around, at rising cost, and each round of resolution produced a new round of demand.
Money is a different kind of constraint. A wafer plant can be built, a substation can be permitted, a model can be retrained with fewer parameters. Capital, by contrast, is priced by people who can change their minds, and it is priced on expectations about a future that no one can yet measure.
That is what makes the shift consequential. Physical scarcity rewards the companies with the long-term contracts and the strongest balance sheets. Financial scarcity punishes them, because a fixed obligation priced in a world of abundant capital can become a burden in a world of cautious capital.
The signs of strain are visible in how the current build-out is paid for. Data center construction and chip purchases are increasingly financed with debt and with leases that extend years past the point at which the current generation of models will be obsolete, and the lenders behind those arrangements are not betting on software. They are betting on demand for computing power continuing to rise.
Hock Tan’s casino line captures the asymmetry. If AI delivers on the most optimistic forecasts, the investors who funded the oddest deals will look like visionaries. If it delivers on a merely good forecast, some of them will discover that the asset they bought was never going to produce the return the price assumed.
Neither executive predicted a bust. Both described a market where the technical questions have been answered well enough, for now, and the financial ones have not. That is a narrower claim than the doom commentary that circulates around AI spending, and a harder one to argue with, coming from suppliers with visibility into what customers have actually ordered.
There is a practical difference between demand and funded demand, and the distinction is easy to lose in a market where order books look full. A hyperscaler that has committed to buy accelerators can defer delivery, stretch a payment schedule or renegotiate, and each of those moves shows up at a supplier before it shows up in the headlines.
The two executives also differ from the critics who argue that AI is a bubble. Hock Tan’s complaint was about the people pricing the assets, not the assets themselves, and Lip-Bu Tan’s was about the patience of the money behind them. Both men expect the technology to matter. They are questioning whether the financing structure around it can carry the weight being placed on it.
For the companies building models, the message is uncomfortable. The next obstacle to scale may not be a chip shortage or a power grid delay but the willingness of lenders and investors to keep funding a business that consumes cash faster than it produces it. Demand can exceed supply and still not be enough to justify the cost of serving it.
The test will show up in two places. Watch whether the terms on data center debt tighten, and watch whether the announcements of new capacity slow. Neither would be a verdict on AI, but together they would say that the constraint Hock Tan and Lip-Bu Tan described has arrived.
For now the two men stand closer to the money than almost anyone outside the financial sector, and both say the same thing: the binding limit on the AI build-out is becoming the price of paying for it.
