Why Intel and Broadcom CEOs flagged capital as an AI constraint

Hock Tan described access to capital as a possible major limit. Lip-Bu Tan said, “if there’s any challenge in the front, it’s capital and the cost of capital, and then the other part is in the longer run, people want to see the ROI,” he told SiliconANGLE. ROI, or return on investment, is the financial benefit a company gets relative to what it spends.

That points to two distinct tests for an AI build-out: whether companies can finance it, and whether the resulting products or services generate enough value to justify the expense. Neither executive quantified financing conditions or said how many infrastructure plans might be affected. Their remarks do not establish that capital is the industry’s main constraint.

Could computing supply remain tight even if chip shortages ease?

Hock Tan forecast that some chip shortages could ease around 2028 or 2029, while demand for computing capacity could continue to exceed the industry’s ability to supply it for several years. The first forecast concerns particular shortages; the second describes a broader gap. Neither is a measured outcome.

SiliconANGLE also reported that Intel had disclosed it could meet 50% of its orders. The report did not specify which orders or what period the figure covered, and it did not independently verify the number. It cannot by itself establish the scale or duration of a wider supply problem.

A shortage of a particular chip and an overall shortage of computing capacity are related, but not interchangeable, constraints. And limited access to funding could affect the ability to build infrastructure even if components are available. The conference comments do not establish which factor, if any, is the binding limit on expansion.

What evidence would show that AI spending is paying off?

Hock Tan’s comment about training costs was specifically about OpenAI. He told SiliconANGLE, “For now it’s still losing money because they’re spending a lot on training.” That is his assessment of OpenAI, not an independently verified finding about AI companies as a whole.

Separately, SiliconANGLE reported that a sizable portion of AI labs’ enterprise revenue came from pilot projects. That describes a share of revenue, not how many projects remain pilots, whether they later expand or what returns they produce. The report does not establish that enterprise AI deployments generally fail to deliver value.

Hock Tan also argued that AI businesses need customers able to earn returns on inference spending. Inference is the computing used to generate outputs from a trained model. Whether the revenue from those services can cover their costs is a separate question from whether companies can afford to build the underlying infrastructure.

Lip-Bu Tan pointed to open-source development as one possible route to efficiency. “I’m a strong believer of open ​source, that’s the best ​way to drive efficiency,” he told SiliconANGLE. His support is a view about how costs might be addressed, not proof that open-source systems deliver lower costs or adequate performance in particular uses.

The CEOs’ comments frame three linked questions: whether companies can finance AI infrastructure, whether computing supply can meet demand, and whether customers will pay enough to make deployments worthwhile. Independent data on financing, supply and enterprise returns would be needed to establish the scale of those pressures.