Broadcom Raises AI Chip Forecast to $115 Billion, Sees Demand Doubling Again by 2028

Broadcom's latest guidance offers the clearest signal yet that the AI infrastructure buildout has not peaked. The chipmaker on Wednesday raised its fiscal 2027 AI chip revenue forecast to about $115 billion, up from a prior estimate of over $100 billion, and said it expects that figure to roughly double again to $230 billion in fiscal 2028.

That is a meaningful upward revision, and it arrives at a moment when investors have grown increasingly skeptical about whether hyperscaler capital spending on AI can be justified by returns. Broadcom's numbers argue the answer is yes, at least for now.

Shares fell more than 1% in extended trading, recouping some earlier losses. The stock has gained roughly 6% year to date, badly trailing both its semiconductor peers and the broader Philadelphia Semiconductor Index. Part of that underperformance reflects intensifying competition. Marvell recently landed a custom chip deal with Google, underscoring that Broadcom no longer has the custom silicon market to itself.

The Demand Picture

The third quarter told a straightforward story: AI chip sales more than tripled year over year to $16.7 billion, pushing total revenue to $29.59 billion and beating the Street's estimate of $29.36 billion. Adjusted profit came in at $3.32 per share against expectations of $3.24. Bookings for AI chips topped $30 billion in the quarter alone, a figure that speaks to demand still outrunning supply commitments made months ago.

Broadcom's custom AI chips sit inside the infrastructure of Meta, Alphabet's Google and OpenAI. That customer roster matters. These are the companies driving the bulk of global AI capital expenditure, and their willingness to keep signing multi-year deals is the single best real-time indicator of whether the AI spending cycle is durable or overextended.

CEO Hock Tan gave analysts specifics rarely offered at this level of granularity. He said Broadcom now has visibility into AI infrastructure deployments through 2028 totaling more than 10 gigawatts for Anthropic, over 5 gigawatts for OpenAI, and 3 gigawatts for Meta. Tan also said the company has secured enough supply to support the higher forecast, addressing a lingering worry that chip shortages could cap growth regardless of demand.

Patrick Moorhead, CEO of Moor Insights & Strategy, called the disclosures substantive rather than promotional. "That is committed capacity, not aspiration, and it closes most of the gap to what the market wanted," he said. That distinction, committed versus aspirational, is exactly what institutional investors have been demanding from every AI-adjacent company reporting this earnings season. Vague enthusiasm no longer moves the needle. Signed capacity does.

Fourth-quarter guidance was the one soft spot. Broadcom expects revenue of about $34.8 billion, slightly below the average analyst estimate of $35.03 billion, according to LSEG data. The miss is modest in percentage terms, but in a market primed to punish any hint of deceleration, even a small gap between guidance and consensus can trigger an outsized reaction. The after-hours share decline reflects that sensitivity more than it reflects any real deterioration in the underlying business.

Why It Matters

The broader significance here goes beyond one company's earnings print. Broadcom's results reinforce a thesis that has been building for months: AI capital spending is no longer concentrated solely in Nvidia's GPUs. It is broadening into custom silicon, networking gear, and the connective infrastructure that ties sprawling AI data centers together. Broadcom sits at the intersection of both trends, supplying custom chips to hyperscalers while also providing the networking components that link thousands of accelerators into functioning clusters.

This diversification matters for risk assessment. A single-vendor AI supply chain is fragile; a multi-vendor one, with Broadcom, Marvell, and others competing for custom silicon contracts, is more resilient but also more competitive on margin. Investors should note that Marvell's deal with Google is not an isolated event. It signals that hyperscalers are actively hedging their chip suppliers, a rational move given the capital at stake, but one that introduces pricing pressure Broadcom will need to manage.

There is also a macro dimension worth flagging. Reports elsewhere this week noted that Texas has moved to halt new power connections for some data centers, and that state Republicans are turning more skeptical of unchecked data center growth. Power availability, not chip supply, may end up being the binding constraint on AI infrastructure expansion over the next few years. Broadcom's forecasts assume the electricity, cooling, and physical infrastructure to support 10-plus gigawatts of Anthropic capacity and similar commitments from OpenAI and Meta will actually materialize on schedule. That is a reasonable base case today, but it is not guaranteed.

Key Risks

Three risks stand out. First, customer concentration: Meta, Google, OpenAI and Anthropic represent an enormous share of Broadcom's AI revenue growth, and any pullback in their capital spending plans would flow through quickly. Second, competitive intensity: Marvell's Google win shows hyperscalers are diversifying suppliers, which could pressure Broadcom's pricing power over time even as total addressable demand grows. Third, physical infrastructure bottlenecks: power grid constraints, exemplified by Texas regulators' recent hesitancy, could delay deployment timelines that Broadcom's own guidance assumes will hold.

None of these risks undermine the core thesis that AI infrastructure spending remains robust. They do argue for tempered expectations around the pace and smoothness of that growth.

The Bottom Line

Broadcom's raised forecast and gigawatt-scale customer commitments are a credible data point supporting continued AI infrastructure investment, not just corporate optimism. The near-term revenue miss and stock underperformance reflect a market demanding perfection rather than any structural weakness in the business. Investors should watch competitive dynamics with Marvell and power infrastructure constraints as the more likely sources of disruption over the next 18 months, rather than a collapse in underlying AI chip demand.