Broadcom CEO Warns: Most AI Investors Are "Like Gamblers" — Capital Will Become the Biggest Bottleneck

Broadcom (AVGO.US) President and CEO Hock Tan issued a rare double-barreled warning about the artificial intelligence investment boom at a conference in Singapore on Monday: he asserted that capital will replace chip supply as the single biggest constraint on the AI industry's next phase, while unsparingly criticizing most AI investors for having "no idea what they're investing in" and behaving "almost like they're in a casino."

Hock Tan made the remarks during a semiconductor roundtable at the inaugural DBS Asia Future Forward Conference. He was joined on stage by Intel (INTC.US) CEO Lip-Bu Tan, with DBS Group CEO Tan Su Shan serving as moderator.

His criticism of investors was blunt: "A lot of investors have no idea what they're investing in." He went further, saying these investors "set extremely exaggerated expectations, and then eventually discover they're almost like they're in a casino."

Bottlenecks Keep Moving, Ultimately Landing on Capital

Hock Tan's core argument is that supply chain constraints will be resolved one after another, but the capital constraint will not disappear on its own. He reflected on the past two years, noting "we've seen a series of bottlenecks emerge one after another, but we managed to overcome them," and specifically predicted that memory shortages "may no longer exist a year from now," while wafer capacity tightness could ease by 2028 or 2029.

However, after the sequence of bottlenecks spanning "wafers, memory, power, and land," he believes the biggest constraint will be capital itself. "I think capital is probably going to be the most severe constraint," he said on stage.

Hock Tan pointed out that frontier AI labs OpenAI and Anthropic have already demonstrated that their latest models deployed in applications like AI personal assistants can generate substantial revenue, but financing will become the primary bottleneck in the next phase. He put it even more plainly: the real driver is not "selling shovels," but whether frontier AI companies can build models valuable enough to generate returns that justify the current buildout. Value is migrating up the industry chain, and hardware supply alone cannot sustain this boom.

Praise and Reservations on OpenAI

On the valuation question, Hock Tan's stance runs counter to market chatter — he does not believe OpenAI is overvalued. The company is seeking to raise at least $30 billion at a $1.4 trillion valuation.

But his interpretation of that number is notably restrained. He acknowledged that OpenAI achieved roughly $50 billion in annualized revenue in September, which does point to a viable business model, yet there is "almost no guarantee of long-term profitability." "Right now it's still losing money because they're spending too much on training," he said. "Frankly, at this stage, it doesn't prove much — it only shows there's substantial potential revenue growth in providing frontier generative AI models."

He delivered an even more cutting assessment of enterprise AI spending, using a vivid analogy: "When you were a kid and went to the arcade, you played, your parents gave you $10 in tokens, you blew through them in 10 minutes and learned nothing. That's what a lot of companies are doing today." He used this to question whether a swath of corporate AI spending is being wasted, and raised doubts about the long-term economics of persistently loss-making AI labs.

The Man Calling Capital a Bottleneck Is Raising Money for His Clients

Hock Tan is no bystander. Broadcom is deeply involved in the AI infrastructure financing chain, having turned itself into one of the largest capital intermediaries in that chain.

According to an October 7 report, Broadcom is in early-stage discussions to arrange more than $50 billion in debt financing for OpenAI, to support OpenAI's purchase of custom AI chips the two companies are jointly developing. Apollo Global Management and Blackstone have approached Broadcom about becoming lenders on the deal. Negotiations remain at an early stage and the final size could change; the financing could cover multiple gigawatts of compute capacity and is expected to close before year-end.

Broadcom is also a funding provider to Anthropic. Anthropic's confidential IPO prospectus shows Broadcom has agreed to provide up to $42 billion in financing to Anthropic to support its AI infrastructure spending; the funding corresponds to roughly one-third of Anthropic's five-year, $125.2 billion TPU compute lease. The filing also disclosed that Broadcom can designate financing partners, and the related debt instruments are convertible into Anthropic shares.

Based on market reporting, the Anthropic-related financing package has been reported at over $60 billion, of which $42 billion is senior debt guaranteed by Broadcom and $18 billion is subordinated debt without a Broadcom guarantee, led by Blackstone, with syndication launched in October. Combined with the $35 billion financing partnership Broadcom completed with Apollo and Blackstone in June, plus the proposed OpenAI financing, Broadcom's total funding arrangements now exceed $145 billion.

In other words, the man saying "capital is the biggest bottleneck" is precisely the one using his balance sheet to solve his clients' capital problems. That is also the most concentrated source of market skepticism about the AI trade.

Counter-Evidence: IPO Withdrawal and the Revenue Recognition Debate

Hock Tan's warning has a ready-made footnote. Firmus Grid, an Australian data center company backed by Nvidia (NVDA.US), Blackstone, and Coatue, withdrew its $5.5 billion IPO plan last week.

Firmus had planned to list at A$11 per share, implying an equity value of approximately $30.4 billion — nearly triple the $10.5 billion valuation the company commanded less than three months earlier. The company announced on October 9 that it was withdrawing its listing application from the Australian Securities Exchange, citing recent market volatility, and would instead pursue private financing, with a round of $2 billion to $3 billion under discussion. Yet the company's fiscal 2026 revenue was just $51 million, and only 46 megawatts of its 912-megawatt development pipeline has been completed — investors were being asked to price a future that remains largely on paper in the form of project plans and partnership agreements.

Just a week earlier, the valuation anchor for the AI trade experienced another jolt. Reports indicated OpenAI's annualized revenue at the end of September was "close to $50 billion," roughly $20 billion below the approximately $70 billion figure previously hinted to investors. The news triggered a broad selloff in U.S. tech stocks on October 8, with the Nasdaq Composite falling 1.3% to 27,193.34. However, the gap largely stems from accounting methodology — the higher figure was recalculated by OpenAI's investors using Anthropic's approach, under which Anthropic counts customer spending on its models through cloud platforms like AWS and Google Cloud as revenue, while OpenAI does not.

Hock Tan's remarks at the same event formed a subtle echo of this "methodology dispute": he did not deny the growth, but refused to equate growth with profitability.

High Rates and Broadcom's Financial Reality

What makes "capital as a bottleneck" more tangible is the price of money. The 10-year U.S. Treasury yield is currently hovering around 5.25%, not far from the 24-year high touched early last week. Morgan Stanley estimates AI infrastructure will require approximately $1.5 trillion in external financing by 2028 — making "access to capital" a leading indicator for chip demand.

Broadcom's own financials have yet to show strain. The company reported fiscal 2026 third-quarter revenue (quarter ended August 2) of $29.6 billion, up 86% year over year, with AI semiconductor revenue of $16.7 billion, up 221%, accounting for 56% of total revenue. The company guided fourth-quarter revenue of $34.8 billion and AI semiconductor revenue of $21.7 billion, and raised its fiscal 2026 AI revenue guidance from $56 billion to $58 billion.

Notably, Hock Tan forecast that Broadcom's annual AI chip revenue will grow from roughly $58 billion this fiscal year to $230 billion by fiscal 2028 — nearly a fourfold increase. As of the October 9 close, Broadcom traded at $361.54, with a market capitalization of approximately $1.73 trillion. In Monday's intraday trading, Broadcom shares fell 1% to $357, down roughly 28% from the 52-week high of $495.

Institutional Views Are Divided

Market assessments of Broadcom's role are sharply polarized. Seaport Research analyst Jay Goldberg said: "Nvidia has deployed a massive balance sheet, and Broadcom has had to follow suit." He views Broadcom's financing arrangements as a reactive move to a competitor's playbook.

Rothschild & Co. managing partner Robert Leitao's concern goes a step further: the market is "right now highly concentrated on a single bet — whether these two companies can generate enough revenue to support all the financing." Anthropic itself acknowledged in its prospectus that Broadcom's dual role as both hardware supplier and financier constitutes a potential conflict of interest that could affect the pricing and availability of its compute capacity.

Some institutions believe the market is overreading the risk. One sell-side view assigns Broadcom a "Buy" rating with a $450 price target, implying roughly 25% upside, arguing that the market has overstated the risks associated with Broadcom's $29 billion gap guarantee and $42 billion convertible debt financing, given the company's strong cash flow, limited exposure, and AI revenue guidance still pointing to $230 billion in fiscal 2028 and EPS above $30.

Based on the October 9 closing price, Broadcom's trailing-twelve-month P/E ratio is approaching 70x. For a company that keeps saying "capital is the biggest bottleneck" while simultaneously raising money for its clients, the premium the market is willing to pay for that statement may be harder to answer than the statement itself.

Another voice at the conference came from Intel. Lip-Bu Tan argued that AI is bigger than the internet: the first wave was training, and the next wave will be inference and physical AI. He listed memory, supply chain, and power as constraints, and predicted the future will not be a "one chip to rule them all" scenario, but rather an "XPU" architecture combining GPUs, CPUs, and specialized accelerators. He also said hardware will remain in short supply for the next two to three years, with memory chips potentially the first segment to lose momentum. Conference moderator Tan Su Shan opened the event by noting that the market is repricing the risks of AI financing.

On prediction market Polymarket, traders are pricing a 6% probability that a "qualifying AI drawdown" occurs by December 31. The contract requires three of six conditions to be met within 90 days, including a 50% decline in Broadcom or Nvidia shares from all-time highs. Trading volume on the contract stands at $2.44 million.