Broadcom Inc. (AVGO)

News Summary

• Broadcom's Q3 AI semiconductor revenue hit $16.7B, up 221% YoY and 54% sequentially, now representing 56% of total revenue; Q4 AI semiconductor guidance is $21.7B (up 236% YoY), with management guiding fiscal 2027 AI revenue to $115B and fiscal 2028 to $230B—numbers Hock Tan described as deliberately conservative given demand exceeds supply-constrained outlooks.
• Jalapeno, Broadcom's custom XPU co-designed with OpenAI, was explicitly positioned as outperforming NVIDIA Grace Blackwell Ultra on performance-per-watt, latency, and throughput, and comparable to Vera Rubin GPUs at roughly half the cost—the most direct public competitive claim against NVIDIA inference silicon to date; OpenAI is preparing to tape out a second-generation chip with a third already in planning.
• Anthropic is on track to be Broadcom's largest XPU customer in 2027 with 5 gigawatts of TPU v8i planned, scaling to 10 additional gigawatts in 2028; Tomahawk Ultra, Broadcom's Ethernet-based scale-up networking switch for XPU and GPU clusters, entered production deployments this quarter with adoption pace described as surprising even to management—signaling accelerating custom AI infrastructure buildout across multiple hyperscalers and labs.
• Q3 revenue was $29.6B (+86% YoY), non-GAAP EPS $3.32 (beat consensus of $3.24), non-GAAP operating income $20.1B (+92% YoY), and free cash flow $13.7B (46% of revenue); Q4 guidance of $34.8B revenue (+93% YoY) came in slightly below the $35.03B analyst consensus, with non-GAAP operating margin guided at 66%.
• AVGO fell roughly 6% in after-hours trading despite the beat, with the Q4 revenue guide missing consensus as the proximate trigger; sentiment was broadly constructive on the multi-year AI demand thesis—analysts probed supply constraints and financing structures rather than questioning demand—but the stock's year-to-date underperformance versus the S&P 500 (roughly 6% vs. 12%) reflects lingering valuation sensitivity to any guidance shortfall.

Financial Highlights

Earnings Call Highlights

Broadcom delivered a blowout quarter with AI semiconductor revenue tripling year-over-year to $16.7 billion, and management's posture was one of barely contained confidence bordering on triumphalism, with Hock Tan repeatedly framing the company's XPU momentum as a structural, multi-year displacement of GPU incumbents. The dominant theme was the articulation of an extraordinary forward revenue ramp—$115 billion in AI revenue for fiscal 2027 and $230 billion for fiscal 2028—numbers so large they reframed the competitive landscape entirely. Analysts were largely in validation mode, probing supply constraints and financing structures rather than questioning the demand thesis, signaling broad buy-side acceptance of Broadcom's AI narrative. The call was notable for the specificity of customer-level gigawatt disclosures and the explicit performance comparisons positioning XPUs as superior to NVIDIA's Vera Rubin and Grace Blackwell Ultra for inference workloads.

• Broadcom guided fiscal 2027 AI semiconductor revenue to approximately $115 billion and fiscal 2028 to $230 billion, with Hock Tan stating supply has been secured for both figures and that demand actually exceeds these outlooks—implying the numbers are deliberately conservative.
• Jalapeno, OpenAI's first custom accelerator designed with Broadcom, was disclosed to outperform Grace Blackwell Ultra in performance per watt, latency, and throughput, and to be 'comparable to Vera Rubin GPUs' for OpenAI workloads at roughly half the cost of a GPU—a direct and pointed competitive claim against NVIDIA.
• Anthropic is on track to become Broadcom's largest XPU customer in 2027, with 5 gigawatts of TPU v8i planned for that year and 10 gigawatts of incremental capacity in 2028; OpenAI is projected to deploy over 5 gigawatts of Jalapeno and its successor in 2028, making it the second largest XPU customer.
• Broadcom's Tomahawk Ultra, enabling low-latency Ethernet-based scale-up networking within XPU and GPU clusters, is beginning production deployments this quarter and is seeing adoption across both XPU and GPU cluster customers, with Charlie Kawwas noting the ramp 'surprised us.'
• Supply constraints are multidimensional—leading-edge wafers, substrates (addressed by Broadcom's own Singapore fab coming online in fiscal 2027), HBM memory, and land/power/shell for data centers—with Hock Tan framing the $115B and $230B outlooks as already accounting for these bottlenecks through joint analysis with each of the six customers.

Sources

• Broadcom delivers strong earnings view as CEO touts growth with AI labs
• Broadcom Inc. Announces Third Quarter Fiscal Year 2026 ...
• Broadcom Inc. Announces Third Quarter Fiscal Year 2026 Financial ...