Broadcom’s AI Supercycle Test: Will Thursday’s Earnings Prove It’s the Next Nvidia?

Broadcom heads into Thursday's print with as much AI momentum – and as much pressure – as any name in the semiconductor complex. The stock has surged to all-time highs, up more than 70% year to date, on the back of explosive AI demand, VMware integration, and a drumbeat of positive datapoints from cloud service providers. That rally leaves very little room for disappointment: consensus is looking for roughly 24–25% revenue growth and low-30s EPS growth, and the buy side is effectively anchored to another step-up in AI guidance for 2026. This is less an ordinary quarter and more a referendum on whether Broadcom can keep compounding into Nvidia’s slipstream as the second pillar of AI infrastructure.

Expectations for fiscal Q4 2025 are straightforward but demanding. Wall Street is modeling about $17.4–17.5 billion in revenue, up roughly 24–25% year on year, and adjusted EPS around $1.87, up about 32%. Management has already guided to $17.4 billion, including $10.7 billion from semiconductors and $6.7 billion from infrastructure software. Within that semi bucket, the star of the show is AI: Broadcom has flagged AI semiconductor revenue of about $6.2 billion, which would be up roughly 66% year on year and represent well over half of semiconductor sales. Investors will want to see that trajectory not only confirmed, but rolled forward via a higher fiscal 2026 AI revenue target.

That AI engine is built around custom accelerators (XPUs) and networking silicon that sit at the heart of hyperscale data centers. Broadcom is the design and manufacturing partner behind Google’s TPU line, which now powers Gemini and underpins Google’s plan to offer TPU capacity to external customers like Anthropic and potentially Meta. Analysts estimate Google’s TPUs contributed on the order of $8 billion of Broadcom revenue in FY24, with unit volumes expected to climb from roughly 2 million in 2025 to 3 million or more in 2026, alongside a meaningful ASP step-up as newer generations roll out. On top of Google, Broadcom is ramping bespoke chips at Meta and ByteDance, and most recently signed a 10-gigawatt custom accelerator and networking deal with OpenAI that should start contributing in the back half of 2026 and scale through 2029.

Layered onto that is a potentially significant share shift in custom silicon away from Marvell. Reports that Microsoft is in talks to move parts of its custom chip program from Marvell to Broadcom have reinforced the sense that AVGO is becoming the default co-design partner for cloud giants who want something tailored rather than an off-the-shelf GPU. Marvell’s selloff on the headlines – and disappointment at being left out of the latest S&P 500 additions – highlight the contrast: while Marvell is still fighting for sockets, Broadcom is increasingly seen as the safe pair of hands for massive, multi-year ASIC ramps. If management confirms deeper ties with Microsoft or quantifies incremental revenue from these shifts, it would go a long way toward justifying the premium multiple.

Last quarter set a high bar. Broadcom reported record Q3 revenue of $16 billion, up 22% year on year, with semiconductor revenue of $9.2 billion and infrastructure software of $6.8 billion. AI semiconductor revenue reached $5.2 billion, up roughly 63%, marking a tenth consecutive quarter of strong AI growth. Non-AI semis were sluggish but stabilizing, with management explicitly saying they did not expect a V-shaped recovery, while VMware drove software gross margins into the 90s and helped lift consolidated gross margin to 78.4%. Perhaps most importantly, backlog swelled to around $110 billion, largely driven by AI-related bookings, and a fourth qualified XPU customer placed more than $10 billion in orders, prompting management to raise its 2026 AI outlook.

Against that backdrop, Thursday’s call will be all about forward guidance and mix. Several firms – including Oppenheimer, Mizuho, Goldman, BofA and Jefferies – have recently raised price targets into the $420–$480 range and highlighted Broadcom as a top AI pick, with some more aggressive voices modeling AI revenue to more than double again in 2026. Many on the Street expect updated 2026 AI guidance to exceed the prior “100% growth” framework. The split between XPUs and networking will matter too: XPUs and HBM-heavy inference workloads tend to dilute gross margins, but ASIC operating margins are accretive and high-speed networking content per watt of compute is rising as clusters scale from hundreds of thousands to over a million connected GPUs/XPUs.

Key things to watch on Thursday’s print and call:

• Fiscal 2026 AI revenue guide: does management formally raise the bar above prior 100% growth commentary, and how much of that uplift is tied to Google versus newer deals with OpenAI, Meta and others?
• Customer concentration and new wins: any concrete color on Microsoft’s custom chip plans, Anthropic TPUs, or additional XPU customers beyond the fourth already disclosed.
• Gross margin trajectory: with management already flagging about 70 bps of sequential compression on higher XPU and wireless mix, investors will be parsing whether software and networking can keep consolidated margins in the high-70s over time.
• Non-AI semi recovery: signs that broadband, wireless and other legacy segments are moving from “not getting worse” to something more like modest growth would bolster the durability story.
• VMware and software integration: updates on VMware Cloud Foundation adoption, subscription renewals and long-term software growth, which help smooth the inherently cyclical semi business.

Valuation is the other piece of the puzzle. By most metrics, Broadcom is expensive: the stock is trading near all-time highs around the $400 level, with forward P/E estimates in the stratosphere and EV/EBITDA multiples well above its historical range. Bulls argue that those multiples are backward-looking, and that operations will grow into the valuation as AI and custom silicon deals ramp in 2026–2028, compressing the multiple over time. Bears counter that AI capex is cyclical too, and that any pause in hyperscaler spending or wobble in AI monetization could leave investors holding a very richly priced bag.

For now, though, Broadcom occupies a rare space in the AI ecosystem: less speculative than pure-play model providers, more diversified than GPU-only vendors, and increasingly indispensable to any cloud provider that wants to own its own silicon destiny. Thursday night’s report won’t settle the “AI bubble” debate, but it will tell investors whether Broadcom is still gaining share in the arms race – or merely keeping up with expectations that have already run far ahead of the fundamentals.