Broadcom (AVGO) Q1 FY2026 Earnings — Research Note

March 4, 2026 | After-Hours Report

The Headlines

Broadcom reported record Q1 revenue of $19.3 billion, up 29% year-over-year, with AI revenue of $8.4 billion growing 106% year-over-year — above the company's own forecast. The stock moved roughly 2–5% higher in after-hours trading.

Key numbers at a glance:

Q2 Guidance — The Real Story

The forward guidance is where things got exciting:

The 47% YoY total revenue growth expected in Q2 represents a significant acceleration from the 29% in Q1.

Custom AI Accelerator (XPU) Deep Dive — The Crown Jewel

This is where the call delivered the most material new information. CEO Hock Tan laid out a customer-by-customer roadmap that directly maps to the AI infrastructure capex cycle:

Google: Continued growth trajectory in 2026 with strong demand for the seventh-generation Ironwood TPU, with even stronger demand expected from next-gen TPUs in 2027 and beyond.

Anthropic: Off to a strong start in 2026 targeting 1 gigawatt of TPU compute, with demand expected to surge past 3 gigawatts in 2027.

Meta: Tan pushed back on recent analyst reports questioning the future of Meta's custom silicon, saying the MTIA custom accelerator is shipping now and targeting multiple gigawatts of capacity in 2027 and beyond.

Customers 4 & 5: Strong shipments this year, expected to more than double in 2027.

OpenAI (6th customer — NEW): Broadcom disclosed OpenAI as a sixth customer, expecting volume deployment of its first-generation XPU in 2027 with over 1 gigawatt of compute capacity.

The headline number: Tan stated Broadcom has "line of sight to achieve AI revenue from chips, just chips, in excess of $100 billion in 2027."

Shareholder Returns

Broadcom returned $10.9 billion in Q1 through $3.1 billion in dividends and $7.8 billion in stock repurchases. The board authorized up to $10 billion in new share buybacks through 2026. The quarterly dividend remains $0.65/share.

Connecting to the AI Infrastructure Capex Thesis

This is where Broadcom's results fit into the bigger picture across the AI capex cycle, optical networking (Nokia/Infinera), and semiconductor supply chain dynamics.

1. The $600–700B Hyperscaler Capex Wave Is Real — and Broadcom Is a Primary Beneficiary

The four major hyperscalers are projected to increase capex by more than 60% from 2025 levels, with combined spending approaching $700 billion in 2026. Roughly 75%, or about $450 billion, of that spend is directly tied to AI infrastructure — servers, GPUs, data centers, and networking equipment.

Broadcom sits at the intersection of this spend in two critical ways: custom AI accelerators (XPUs) and high-speed networking (Tomahawk switches, optics, interconnects). The customer-by-customer roadmap Tan provided tonight shows Broadcom touching virtually every major hyperscaler's AI buildout.

2. Custom Silicon Is Displacing Some GPU Spend — the Nvidia Counter-Narrative

By early 2026, more than 70% of AI data center revenue has shifted from training to inference, a transition that favors Broadcom's custom ASICs, which offer better total cost of ownership and power efficiency compared to general-purpose GPUs.

The fact that OpenAI — historically Nvidia's most high-profile customer — is now Broadcom's 6th XPU customer is a major strategic signal. This doesn't mean Nvidia loses, but it does confirm that the hyperscalers are pursuing a dual-track strategy: Nvidia GPUs for general training, custom ASICs for optimized inference and proprietary workloads.

3. Connecting to the Nokia/Infinera Optical Networking Research

The scale of compute being described — measured in gigawatts per customer — has enormous implications for the optical networking and interconnect layers. When Tan talks about 1 GW of TPU compute for Anthropic in 2026 and 3+ GW in 2027, that compute needs to be connected via massive data center interconnect (DCI) fabrics. This is precisely the market that Nokia (post-Infinera acquisition) is targeting with 400G/800G+ coherent optical and open DCI architectures.

Broadcom's own Tomahawk 6 switches (102 Tbps) and its AI networking backlog (previously reported at >$10B) are the electrical switching layer, while Nokia/Infinera's optical transport sits at the next layer connecting clusters and data centers. Both benefit from the same capex wave.

4. The Power Constraint Is Becoming the Binding Variable

Tan's consistent use of "gigawatts" as the unit of measurement for customer deployments is telling. Power infrastructure development is increasingly becoming a binding constraint on the pace of AI data center deployment, particularly in the US and Europe. The fact that hyperscalers are planning multi-gigawatt deployments per customer suggests the power infrastructure buildout (generation, transmission, cooling) will become a bottleneck — and an investment opportunity — alongside the silicon itself.

5. The Financing Question

Hyperscalers are increasingly leaning on debt markets to fund the gap between rising AI capex budgets and internal free cash flow, with aggregate capex now exceeding projected cash flows. The top five hyperscalers raised a record $108B in debt during 2025, with projections suggesting $1.5 trillion in debt issuance over the coming years.

This is the sustainability question investors should watch. Broadcom itself generates tremendous free cash flow ($8B in Q1 alone), but its customers are the ones funding the buildout increasingly through leverage. Any wobble in AI monetization timelines could slow the capex cycle.

Software Segment — The Quiet Concern

The one soft spot: infrastructure software revenue of $6.80B came in below the $7.02B consensus. Tan emphasized that Broadcom's infrastructure software (primarily VMware) is "not disrupted by AI," addressing growing investor concern that generative AI models could threaten mature software businesses. This segment bears watching — the VMware integration has been a success story, but if the software business decelerates while AI semiconductor margin mix shifts lower (due to system-level sales), it could pressure overall profitability.

Investment Implications

Bull case: Broadcom is increasingly the picks-and-shovels play for the custom silicon wave. With 6 confirmed hyperscaler XPU customers, a $100B+ AI chip revenue line of sight for 2027, and a massive networking backlog, the company is a direct proxy for the AI capex cycle without the concentration risk of a single customer. The Q2 guidance substantially above consensus suggests the acceleration is real.

Bear case: Customer concentration remains a risk — 6 customers for the vast majority of AI revenue. Gross margin pressure from the shift to system-level AI deliveries is a concern, as the 100 bps sequential margin decline flagged for Q1 reflects higher content but lower-margin rack solutions. The software miss, while small, is worth watching. And the broader question of whether $600–700B in annual hyperscaler capex is sustainable long-term remains open.

Connection to portfolio research: If building a thesis around the AI infrastructure supply chain — as the Nokia/Infinera optical networking research and semiconductor analyst comparison work suggests — Broadcom sits at the center of that supply chain alongside TSMC (manufacturing the XPUs), Nokia/Ciena (optical interconnect), and the power/cooling infrastructure layer. Tonight's results and guidance reinforce that the capex cycle is accelerating, not decelerating.

Disclaimer: This is research commentary, not financial advice. Always do your own due diligence before making investment decisions.