1. Broadcom AI revenue forecast for 2027 rises to $115 billion
Broadcom now expects its artificial intelligence semiconductor business to generate roughly $115 billion in fiscal 2027, the twelve months ending in October 2027, replacing an earlier target of more than $100 billion for the same period. The chipmaker projects about $58 billion in AI revenue for fiscal 2026, meaning the Broadcom AI revenue forecast for 2027 implies sales doubling in a year.
The higher figure reflects demand for custom AI chips that continues to run ahead of what the company had planned for. Anthropic, OpenAI, Google and Meta all depend on Broadcom hardware for parts of their AI infrastructure, and their spending has repeatedly exceeded earlier assumptions. Chief Executive Officer Hock Tan told analysts on the company's most recent earnings call that Broadcom has locked in the supply chain capacity required to reach its targets.
Because Broadcom's financial year closes in October rather than December, the new projection covers a period that begins in November 2026. The timing matters when comparing the number with forecasts from companies that report on a calendar-year basis: the $115 billion goal describes revenue to be earned over a window that is already under way.
2. The quarterly numbers behind the outlook
Recent results show how quickly the AI business has grown. In the third quarter of fiscal 2026, AI semiconductor revenue reached $16.7 billion, an increase of 221% from the same quarter a year earlier.
Total company revenue for that quarter was $29.6 billion, up 86% year over year. AI semiconductors therefore accounted for more than half of everything Broadcom sold, a notable shift for a company that also has a large infrastructure software operation and a long-established business in networking, broadband and wireless chips. A segment that was once one product line among many has become the main driver of the company's results.
Guidance for the current period points to a further step up. Broadcom expects fourth-quarter AI revenue of $21.7 billion, which would represent growth of 236% from the comparable quarter of the prior year. Reaching that figure would bring full-year AI revenue to approximately $58 billion for fiscal 2026.
The growth is also showing up in cash. Free cash flow in the third quarter was $13.7 billion, and the company set its quarterly dividend at $0.65 per share. Free cash flow is the money a business generates after paying for its operations and capital spending, and it is often treated as a test of whether reported growth is real. In Broadcom's case, the figure suggests that AI orders are being converted into cash received, not merely recorded as future commitments.
3. What is an XPU and why are customers buying them?
An XPU is Broadcom's term for a custom AI accelerator: a chip designed around the workloads of one specific customer rather than sold as a general-purpose product. It is one of the two pillars of the company's AI business, alongside networking components that connect processors inside data centers.
The appeal of custom silicon comes down to efficiency and control. The largest AI developers run a narrow set of very demanding tasks, chiefly training large models and serving responses from them, at enormous scale. A chip built specifically for those tasks can deliver more performance per unit of power and cost than a general-purpose processor, and it gives the customer a say in the design roadmap. For companies spending tens of billions of dollars on computing capacity, even modest efficiency gains translate into large savings.
Broadcom's role in that arrangement is to co-design the chip with the customer and manage its manufacture. The company does not compete with its clients in building AI models, which makes it a natural partner for firms that want their own hardware without creating an in-house semiconductor operation from scratch.
The networking side of the business benefits from the same build-out. Modern AI systems link very large numbers of processors together, and the speed at which data moves between them can limit overall performance. Switches and related components that carry that traffic are therefore purchased in step with the accelerators themselves, giving Broadcom a second source of revenue from each new data center.
4. Anthropic is set to become the largest custom chip customer
Among the companies named as users of Broadcom hardware, one stands out in the outlook. Anthropic is projected to become Broadcom's biggest XPU customer in 2027.
That projection illustrates how the customer base for custom accelerators has broadened. The earliest large-scale buyers of such chips were established technology groups with years of experience designing their own hardware. The newer AI developers are now ordering at a scale that rivals or exceeds them, as they race to secure the computing power needed to train and operate their models.
For Broadcom, a wider set of large customers reduces dependence on any single buyer, although the business remains concentrated in a small number of very big accounts. Tan said on the earnings call that demand from the company's standout hyperscale customers already exceeds Broadcom's initial outlook, which indicates that the constraint on growth is the ability to supply chips rather than the willingness of customers to buy them.
5. Supply is the limiting factor
Tan's comments on the supply chain go to the heart of whether the forecast is achievable. Advanced AI chips rely on leading-edge manufacturing capacity, specialized packaging and high-performance memory, all of which have been in short supply across the industry as AI spending has surged.
A revenue target is only as credible as a company's ability to deliver the products behind it. By stating that Broadcom has secured the supply chain needed to hit its goals, the chief executive was addressing the most common doubt about aggressive AI forecasts: that orders may exist on paper but cannot be filled on schedule. His additional remark that demand is running above the outlook implies that the $115 billion figure is tied to what the company expects to be able to ship.
The scale of the increase puts that claim in perspective. Moving from roughly $58 billion in fiscal 2026 to roughly $115 billion in fiscal 2027 means finding manufacturing capacity for about twice as much product within twelve months. Quarterly AI revenue, which was $16.7 billion in the third quarter and is guided to $21.7 billion in the fourth, would need to average close to $29 billion across the four quarters of fiscal 2027 to reach the target, a level nearly equal to the revenue of the entire company in the third quarter of fiscal 2026.
The outlook reportedly stretches beyond 2027 as well, extending to $230 billion in AI revenue by fiscal 2028. That would mean doubling again from the fiscal 2027 target, a pace that would require both continued spending by the largest AI developers and a matching expansion of manufacturing capacity.
6. Why the revision matters for the AI trade
The significance of the update lies less in any single number than in its direction. Broadcom had already set a demanding goal of more than $100 billion for fiscal 2027; it then raised that goal, while its chief executive said customers want more than the company first planned for.
Suppliers of AI hardware are among the clearest sources of information about how much the industry is actually spending. Announcements from AI developers about future data center plans describe intentions, while a chip supplier's revenue and guidance describe purchases that have been made or contracted. An upward revision from a company in Broadcom's position therefore serves as evidence that the build-out of AI infrastructure is continuing at or above the pace previously expected.
The composition of the demand is informative too. The customers named as relying on Broadcom hardware include both long-established technology groups, in Google and Meta, and newer developers of AI models, in Anthropic and OpenAI. Spending is therefore not confined to one type of buyer. Companies that fund their infrastructure from large existing businesses and companies whose main product is the AI model itself are both placing orders large enough to move a supplier's forecast.
Shares of companies tied to AI infrastructure have nonetheless been volatile as a group, and analyst opinion on Broadcom, while generally positive, has been formed against that backdrop. Investors have swung between enthusiasm about the scale of spending and concern about whether it can be sustained, and strong guidance from one supplier does not settle that debate.
7. What to watch next
The next test of the forecast is Broadcom's fourth-quarter report. Delivering the $21.7 billion in AI revenue the company has guided to would keep it on course for approximately $58 billion in fiscal 2026 and would lend weight to the claim that the business can double from there.
Beyond the headline AI figure, three things deserve attention. The first is any change to the fiscal 2027 target itself, since management has already shown it is willing to raise it. The second is commentary on supply, given that Tan has identified capacity rather than demand as the factor that determines how much the company can sell. The third is cash generation: free cash flow of $13.7 billion in a single quarter set a high bar, and maintaining it would confirm that rapid growth in AI chips is translating into money that can be returned to shareholders or reinvested.
For now, the picture is of a company whose AI business has grown from a fraction of its sales to the majority in a short time, and which is telling investors it expects that business to double once more over the coming year.
