Broadcom reported record third-quarter results Tuesday, with AI semiconductor revenue surging 221% year-over-year to $16.7 billion and CEO Hock Tan projecting the company will double that figure to $115 billion in fiscal 2027.
The results confirmed Broadcom’s position as the primary beneficiary of the custom AI chip boom, supplying custom accelerators to six major customers including Google, Anthropic, OpenAI, and Meta. AI semiconductor revenue now represents 56% of total company revenue, up from 49% in the prior quarter, and the company expects that share to climb further in Q4 as shipments of next-generation chips accelerate across all of its major customers simultaneously.
The Numbers
Consolidated revenue hit a record $29.6 billion for the quarter ended August 2, up 86% from a year ago. Semiconductor solutions revenue reached $20.8 billion, up 127% year-over-year, while infrastructure software contributed $8.8 billion, up 29%.
| Metric | Q3 FY2026 | Q3 FY2025 | Change |
|---|---|---|---|
| Consolidated Revenue | $29.6B | $15.95B | +86% YoY |
| AI Semiconductor Revenue | $16.7B | ~$5.2B | +221% YoY |
| Operating Income | $20.1B | $10.5B | +92% YoY |
| Free Cash Flow | $13.7B | ~$7.2B | +90% YoY |
| Gross Margin | 75% | 78% | -300 bps |
| AI Revenue as % of Total | 56% | ~33% | +23 pp |
Operating income reached a record $20.1 billion, up 92% from a year ago. Free cash flow was $13.7 billion, representing 46% of revenue. Gross margin came in at 75%, down 210 basis points sequentially as AI chips, with their higher memory content, made up a growing share of the product mix. The company spent $532 million on capital expenditures during the quarter.
The Custom Accelerator Machine
Broadcom’s growth is driven by custom chip programs, known internally as XPUs, built for individual cloud customers rather than the general-purpose GPUs that Nvidia sells. These chips are tailored to each customer’s specific AI workloads, offering better performance per watt for inference and training tasks.
During the quarter, Broadcom shipped Ironwood TPU v7 in high volume to both Anthropic and Google. It also began production shipments of TPU v8i, a next-generation Google accelerator designed with more memory and bandwidth than Ironwood and optimized for inference workloads.
“In performance it is comparable, if not surpasses, the Vera Rubin GPU.” – Hock Tan, Broadcom CEO, on TPU v8i
For OpenAI, Broadcom shipped Jalapeño, the AI lab’s first custom accelerator, which the company said outperforms Nvidia’s Grace Blackwell GPU for inference workloads. Overall XPU shipments rose 3.5 times year-over-year and represented 73% of AI revenue during the quarter.
AI networking revenue, which covers the high-speed switches and optical interconnects that link chips together in large clusters, grew more than 2.5 times year-over-year. Broadcom dominates this market with its Tomahawk switch family, which all six of its XPU customers deploy at scale across their data centers.
The Customer Map
Google remains Broadcom’s largest XPU customer. The company is shipping TPU v8i in production and has planned successive generations of increasingly complex TPUs through 2029. Broadcom expects to deliver an incremental 10 gigawatts of Google TPU capacity in 2028, even as Google’s overall demand continues to grow.
Anthropic is on track to become Broadcom’s largest XPU customer in 2027, surpassing Google. The AI lab deployed 1 gigawatt of Ironwood capacity in 2026 and plans to add 5 gigawatts of TPU v8i in 2027. “Anthropic is on track to become our largest XPU customer in 2027 and sustain that in 2028,” Tan said on the earnings call.
OpenAI’s Jalapeño deployment is planned at 1.3 gigawatts in 2027, with more than 5 gigawatts expected in 2028 across Jalapeño and its successor generation. Broadcom is also in development with OpenAI on a third-generation XPU that is approaching tape-out. The ability to deliver custom silicon at this pace underscores the depth of the engineering partnership between the two companies.
Meta will see production shipments of its custom MTIA accelerator, optimized for inference and recommendation workloads, in the fourth quarter. The company’s six XPU customers collectively represent the vast majority of frontier AI compute demand worldwide, according to Broadcom management.
Looking Ahead: The Path to $115 Billion
Broadcom raised its full-year fiscal 2026 AI semiconductor revenue guidance to $58 billion, up 186% from the prior year and above its earlier forecast of $56 billion. For the fourth quarter, the company expects consolidated revenue of $34.8 billion, up 93% year-over-year, with AI semiconductor revenue accelerating to $21.7 billion, up 236%.
The forward projections are even more striking. Tan said Broadcom has “secured the supply” to double AI revenue again in fiscal 2027 to approximately $115 billion, with demand exceeding that outlook. For fiscal 2028, the company projects AI semiconductor revenue will double once more to $230 billion.
| Fiscal Year | AI Semiconductor Revenue (Projected) | Growth |
|---|---|---|
| FY2025 | ~$20B | Baseline |
| FY2026 | $58B | +186% YoY |
| FY2027 | $115B | +98% YoY |
| FY2028 | $230B | +100% YoY |
The projections rest on the assumption that frontier AI labs will continue scaling their compute infrastructure aggressively. Tan described the current dynamic as “exponential growth in demand from a concentrated group who develops state-of-the-art frontier models.”
The Nvidia Question
Broadcom’s custom chip strategy positions it as the primary alternative to Nvidia in the AI accelerator market. While Nvidia sells general-purpose GPUs used across thousands of customers, Broadcom builds bespoke chips for a handful of hyperscale buyers who want to optimize for their specific workloads.
The two approaches are not mutually exclusive. Several of Broadcom’s XPU customers also buy Nvidia GPUs, and the total addressable market for AI compute is expanding fast enough to accommodate both. But Broadcom’s 221% growth rate, compared to Nvidia’s more modest expansion, suggests that the largest AI labs are increasingly investing in custom silicon to reduce long-term dependence on a single supplier and optimize cost per inference token.
Broadcom’s Q4 guidance calls for non-AI semiconductor revenue of approximately $4.3 billion, up 5% sequentially, and infrastructure software revenue of $8.7 billion, up 25% year-over-year. The software segment, which includes VMware, posted a gross margin of 94% and an operating margin of 84%, providing a stable cash flow base alongside the volatile AI chip business.
The company paid $3.1 billion in cash dividends during the quarter and reduced long-term debt by $5.6 billion. Broadcom carried $59.6 billion in gross principal fixed-rate debt at a weighted average coupon of 4% with 7.4 years remaining to maturity.
Analysts expect Broadcom’s stock to remain under pressure on margin concerns as AI chips dilute overall gross margins, but the sheer scale of revenue growth is likely to overshadow near-term profitability questions. The company’s ability to secure multi-year supply agreements with the world’s largest AI labs gives it visibility into demand that few semiconductor firms can match.
discussion