Broadcom CEO Says Capital Could Become Bigger AI Bottleneck Than Memory Chips

Broadcom chief executive Hock Tan says the next major obstacle to artificial intelligence growth may not be a shortage of chips or memory.

It could simply be money.

Speaking at a panel in Singapore on Monday, Tan said Broadcom is gradually overcoming shortages of semiconductor wafers and memory, but the enormous amount of capital required to build AI infrastructure could become the industry’s biggest constraint.

His comments come as technology companies, cloud providers and AI developers prepare increasingly expensive data-centre projects to support larger models and growing demand for computing power.

FCRF Launches CP-FRM to Build India’s Next Generation of Fraud Risk Professionals

Chip Shortages May No Longer Be the Biggest Problem

For much of the AI boom, the industry’s most visible bottleneck has been hardware.

Companies have struggled to secure advanced processors, high-bandwidth memory and enough semiconductor manufacturing capacity to meet demand.

Tan suggested that Broadcom is making progress in dealing with those supply constraints.

At the Singapore panel, he said the company was overcoming both wafer and memory shortages.

The larger question is now whether companies can continue raising enough money to pay for the next wave of AI infrastructure.

AI Infrastructure Is Becoming Extremely Expensive

Building modern AI systems involves far more than buying processors.

Companies need data centres, networking equipment, memory, power supplies, cooling systems and access to large amounts of electricity.

Those investments often have to be made before the infrastructure begins generating meaningful revenue.

The financing requirement is becoming so large that major AI-related companies are increasingly turning to banks, private-credit firms and debt markets rather than relying only on their own cash flows.

Reuters reported last week that Morgan Stanley estimates AI infrastructure could require about $1.5 trillion in external financing by 2028.

That raises a different kind of risk for the AI boom.

Even if companies can obtain the necessary chips, expensive borrowing or weaker investor appetite could slow how quickly new computing capacity is built.

Broadcom Is Already Exposed to the Financing Challenge

Broadcom sits at the centre of the AI infrastructure buildout through its semiconductor and networking businesses.

It designs custom AI chips and supplies other components used by large technology companies.

But recent financing plans show how capital-intensive those projects have become.

The Financial Times reported that banks have been syndicating a roughly $60 billion financing package connected with Anthropic’s use of advanced Google chips, with Broadcom providing support for part of the debt structure.

Separately, reports have said Broadcom is seeking tens of billions of dollars in financing connected with custom AI-chip projects.

These arrangements illustrate why Tan increasingly sees funding capacity as a potential limit on expansion.

OpenAI and Anthropic Revenue Matters

Tan also pointed to OpenAI and Anthropic as evidence that frontier AI companies can generate substantial revenue from newly deployed models.

According to Reuters, he specifically referred to AI personal assistants as an area where advanced models are producing meaningful commercial returns.

That matters because investors financing expensive AI infrastructure ultimately need evidence that the technology can produce enough revenue to justify the spending.

The debate around AI has therefore started shifting.

The question is no longer only whether the technology works.

Investors increasingly want to know whether revenues can grow quickly enough to support the billions of dollars being spent on chips and data centres.

Returns Are Becoming as Important as Computing Power

Separate industry analysis suggests the economics are becoming more closely scrutinised.

At the same Singapore event, Tan reportedly said the world is spending roughly $200 billion annually to operate AI models while generating about $150 billion in associated revenue.

Many companies using AI have also not yet demonstrated a clear return on their investment.

Those figures do not mean the AI industry is collectively losing exactly $50 billion.

They illustrate a broader concern about whether revenues are keeping pace with infrastructure and operating costs.

The answer will become increasingly important as spending moves from hundreds of billions of dollars into trillions.

Rising Borrowing Costs Add Another Layer of Pressure

The financing challenge is becoming more serious because borrowing money itself has become more expensive.

Global bond yields have risen, increasing the cost of long-term debt for companies building capital-intensive infrastructure.

That creates a difficult combination.

AI companies want to expand computing capacity rapidly, while the cost of financing those projects is moving higher.

Investors are therefore becoming more selective about which projects they are willing to fund and what returns they expect.

The result could be a shift from a race for chips to a race for affordable capital.

What this means for you

The next stage of the AI boom may depend less on whether companies can obtain chips and more on whether they can afford the data centres, electricity and financing needed to use them. If capital becomes harder or more expensive to raise, even strong demand for AI could translate into slower infrastructure expansion.