Broadcom has raised its forecast for artificial intelligence chip sales over the next two years, offering another sign that major technology companies continue to spend heavily on computing infrastructure despite growing questions about the cost of the AI boom.
The Palo Alto, California based semiconductor company said on Wednesday that it expects AI chip revenue to reach about US$115 billion in the fiscal year ending October 2027. That is higher than its previous forecast of more than US$100 billion.
Broadcom expects the figure to rise again in fiscal 2028, reaching roughly US$230 billion. The projection would represent another major increase in demand for the company's custom AI chips and networking equipment as technology companies expand the computing capacity needed to train and operate increasingly sophisticated AI systems.
Broadcom shares fell more than 1 percent in extended trading after the announcement, although the stock recovered some of its earlier losses. The shares have gained about 6 percent this year, but that performance has lagged both major rivals and the wider semiconductor sector.
Investors have become more cautious about semiconductor companies exposed to artificial intelligence as they assess whether the enormous sums being spent on data centres and computing equipment will eventually produce enough revenue and profit to justify the investment.
Broadcom's latest forecast suggests that demand remains strong, even as competition grows and the market becomes more selective.
AI spending spreads beyond Nvidia
The artificial intelligence chip market has long been dominated by Nvidia, whose powerful processors have become a standard component of AI data centres. Broadcom's results show that other semiconductor suppliers are also benefiting as technology companies seek alternatives and build increasingly specialised systems.
Broadcom supplies custom AI chips to some of the world's largest technology companies, including Meta Platforms, Google's parent company Alphabet and OpenAI.
Its role extends beyond processors. The company also produces networking components that allow large numbers of chips to communicate inside AI computing systems. As data centres become larger and more complex, those connections have become an important part of the infrastructure required to run AI workloads.
Broadcom said bookings for its AI chips exceeded US$30 billion in the latest quarter alone. The figure gives an indication of the scale of orders being placed by technology companies preparing for future AI deployments.
Chief Executive Hock Tan told analysts that Broadcom had secured enough supply to support its increased forecast for the coming year. He also said customer demand continued to rise.
The company has visibility into additional AI infrastructure deployments extending through 2028. Those plans include more than 10 gigawatts of capacity for Anthropic, more than 5 gigawatts for OpenAI and 3 gigawatts for Meta.
Patrick Moorhead, chief executive of analyst firm Moor Insights & Strategy, said the commitments provided significant support for Broadcom's outlook.
“That is committed capacity, not aspiration, and it closes most of the gap to what the market wanted,” Moorhead said.
Competition is increasing
Broadcom's stronger forecast comes despite growing competition in the custom AI chip market.
Marvell Technology recently secured a custom chip agreement with Google, adding pressure in an area where Broadcom has built a substantial business. Custom chips allow large technology companies to design hardware around specific AI workloads rather than relying entirely on general purpose processors.
The approach can give customers greater control over performance, power consumption and operating costs. For chip suppliers, however, the market is becoming increasingly competitive as technology companies seek more ways to reduce their dependence on a small number of hardware providers.
Broadcom's position gives it an advantage because the company participates in several parts of the AI infrastructure market. It supplies custom accelerators while also providing networking technology that connects computing systems.
That combination means Broadcom can benefit from the expansion of AI data centres even when individual customers choose different approaches to processing workloads.
Strong quarterly results
Broadcom's latest financial results also showed how quickly its AI business has expanded.
AI chip sales more than tripled in the third quarter, reaching US$16.7 billion. The increase helped push total company revenue to US$29.59 billion, exceeding the average analyst estimate of US$29.36 billion, according to LSEG data.
Adjusted profit reached US$3.32 per share, also ahead of expectations. Analysts had been forecasting adjusted earnings of US$3.24 per share.
The results demonstrate that AI demand is already making a significant contribution to Broadcom's financial performance rather than representing only a future growth opportunity.
The company is now preparing for even larger orders as customers expand their infrastructure over the next several years.
Investors remain cautious
Despite the strong numbers, Broadcom's share price reaction showed that investors are demanding more than evidence of rising AI sales.
Fourth quarter revenue is expected to reach about US$34.8 billion, according to the company. That figure is below the average analyst estimate of US$35.03 billion compiled by LSEG.
The relatively modest shortfall was enough to weigh on the stock during extended trading.
Investors have spent much of the year questioning whether the enormous capital commitments being made by technology companies can generate adequate returns. Alphabet, Meta, Microsoft and other major technology groups have committed tens of billions of dollars to data centres, processors, networking equipment and energy infrastructure.
The spending reflects expectations that AI services will eventually generate substantial new revenue. But building the necessary infrastructure is expensive, and companies must continue investing before the full commercial benefits become clear.
That tension has become one of the central issues for the technology sector.
Broadcom sees demand continuing through 2028
Tan's comments indicate that Broadcom is not seeing a slowdown in customer plans.
The company already has visibility into large AI deployments extending through 2028, including major commitments from Anthropic, OpenAI and Meta. Broadcom's increased forecast for fiscal 2027 and its projection of roughly US$230 billion in AI chip revenue for fiscal 2028 reflect those expectations.
The scale of the numbers is significant. Broadcom previously expected AI chip revenue above US$100 billion in fiscal 2027. Raising that estimate to approximately US$115 billion means the company now expects another US$15 billion or more in annual AI chip sales than it previously projected.
The forecast for fiscal 2028 would take that business to approximately twice the size expected for 2027.
That growth depends on technology companies continuing to expand AI infrastructure at a rapid pace. It also depends on Broadcom maintaining access to the manufacturing capacity and other components required to meet those orders.
For now, the company says supply is sufficient to support the higher forecast.
The wider semiconductor race
Broadcom's outlook also points to a broader shift in the semiconductor industry.
The AI hardware race is no longer limited to buying the most powerful processors available. Technology companies are increasingly developing custom chips, specialised systems and high speed networking equipment to handle enormous AI workloads.
That creates opportunities for suppliers beyond Nvidia.
Broadcom's performance shows how the expansion of AI infrastructure can generate demand across several parts of the semiconductor supply chain. Every new data centre requires processors, networking equipment, memory, power systems and other specialised hardware.
The financial stakes are equally large. Customers are committing billions of dollars to infrastructure while chip companies are expanding production and forecasting unprecedented levels of demand.
Broadcom's latest figures provide strong evidence that the spending cycle remains intact. The market's response also shows that investors are asking a harder question now: not simply how much companies will spend on AI, but how efficiently that spending will translate into sustainable earnings.
Broadcom enters that debate with a sharply higher forecast, more than US$30 billion in AI chip bookings in the latest quarter and major customer deployments already planned through 2028. For the semiconductor industry, the next test will be whether that enormous demand can continue translating into sustainable profits as competition intensifies.