Broadcom’s NASDAQ:AVGO numbers out late on Thursday offered a textbook case of how to turn AI hype into hard cash. Fiscal Q3 revenue rose 22% year on year to $15.95bn, with non-GAAP EPS at $1.69 and adjusted EBITDA at $10.7bn (67% margin).
Management guided Q4 revenue to about $17.4bn and kept the EBITDA margin at 67%, while declaring a $0.59 quarterly dividend and flagging record free cash flow of $7.0bn (44% of sales).
Beneath the headline figures sits the engine now driving the valuation: custom AI silicon and the plumbing that connects it. AI semiconductor revenue climbed 63% to $5.2bn in the quarter and is expected to accelerate to $6.2bn in Q4, even as management acknowledged softness across non-AI enterprise networking and storage.
New customer on the radar for Broadcom
The company also said a “new customer” had placed more than $10bn of AI-infrastructure orders, a catalyst for meaningfully stronger AI growth in fiscal 2026. Market reporting on Friday widely linked the unnamed buyer to OpenAI and a Broadcom partnership to design and manufacture accelerators — a potential wedge against Nvidia’s NASDAQ:NVDA dominance.
Wall Street duly sharpened its pencils. Bernstein’s Stacy Rasgon lifted his target to $400, arguing the AI trajectory could quicken beyond prior expectations after Hock Tan sketched a more expansive outlook on the call.
- Pictet brings trio of active ETFs to the US market
- Which stocks will be next to reach the $4 trillion club?
- TSMC profits surge 39% as AI boom drives record results
Jefferies pushed its target to $350, highlighting the “fourth” AI ASIC customer arriving faster than anticipated. Morgan Stanley marked its target up to $382 and reiterated Overweight, leaning into custom-silicon mix gains as hyperscalers seek cost-efficient, task-specific computing power. The tone across the Street: Broadcom’s AI narrative is re-accelerating, with custom accelerators and Ethernet-centric networking giving it leverage that is orthogonal — rather than merely derivative — to GPU cycles.
Any road bumps for Broadcom?
Two caveats temper the euphoria. First, the gap between booked AI racks and shipped revenue remains. Management’s $10bn-plus order will ship largely from the second half of fiscal 2026; execution risk — supply, yields, and customer roadmaps — is therefore non-trivial. Second, the non-AI semiconductor businesses were down sequentially, a reminder that Broadcom is not immune to enterprise spending lulls. Still, with VMware bedding in and infrastructure software revenue up 17% year on year to $6.79bn, the software leg continues to provide diversification and cash.
What distinguishes Broadcom’s setup is less a single blockbuster chip than the compounding effect of three flywheels:
- custom AI ASIC wins that convert capex line-items at the biggest buyers into multi-year programmes;
- an Ethernet roadmap (Tomahawk Ultra, Jericho) tuned to AI clusters as they scale out; and
- VMware-driven software cash flows that fund both R&D and shareholder returns.
In that context, Tan’s pledge to stay on “at least five years” matters: capital allocation discipline — and an iron stomach for large integrations — is central to the story.
The market reaction — a 9%+ gain on the stock on Friday morning in the US — looks less like a single-print squeeze and more like a reset of medium-term AI assumptions. If the $6.2bn Q4 AI run-rate lands as guided and the fourth customer is indeed OpenAI, consensus models will likely pull forward AI revenue ramps for FY26–27 and edge operating leverage higher, even with mix headwinds. Against that, investors should watch delivery timing, any signs of customer concentration risk, and whether Ethernet retains its share of AI networking as alternatives mature.
On balance, Broadcom has earned the Street’s benefit of the doubt — not for promising an AI future, but for selling it, at premium margins.
