Here’s what’s strange about Broadcom right now. The company just posted $10.8 billion in AI semiconductor revenue – up 143% from a year ago. It guided Q3 AI chip revenue to $16 billion, which would represent more than 200% year-over-year growth. It has a $73 billion AI-specific order backlog. It just unveiled a custom chip called “Jalapeño” with OpenAI.
And the stock is down roughly 25% from its all-time high.
That disconnect is either a warning sign or an opportunity. The case for the latter is worth building out carefully.
What Actually Happened at Earnings
Broadcom’s Q2 fiscal 2026 results, reported on June 3, were objectively strong. Total revenue climbed 48% year-over-year to a record $22.2 billion. Adjusted EBITDA hit $15.2 billion – 69% of revenue – up 52% from a year prior. Free cash flow jumped 60% to $10.26 billion. These are not the numbers of a business slowing down.
The selloff had almost nothing to do with what happened. It had everything to do with what was expected.
Wall Street had been modeling $17.2 billion in Q3 AI chip revenue. Broadcom guided to $16 billion. That $1.2 billion miss against an already extraordinary number triggered a 12%+ overnight drop. Some investors also read CEO Hock Tan’s commentary on key-customer concentration risk as a signal – and the broader AI/tech sector was already under pressure, which pulled AVGO down further regardless of its own results.
Slight tangent, but it matters: this pattern – where extraordinary results still fail to clear an elevated bar – is not unique to Broadcom. It’s been a recurring dynamic across the AI infrastructure trade in 2026. The question isn’t whether the business is good. It’s whether the stock had priced in perfection.
The Custom Silicon Story Nobody Fully Understands
Most investors still frame Broadcom as a networking chip company. That framing is roughly two years out of date.
Broadcom designs custom AI accelerators – application-specific integrated circuits (ASICs) – for hyperscalers who want chips built precisely for their workloads rather than general-purpose GPUs. Google’s TPU programs have historically relied on Broadcom as a key silicon partner, and Broadcom has also highlighted custom accelerator and AI networking demand from multiple hyperscale and frontier-model customers. The Jalapeño chip, unveiled with OpenAI in late June, is a purpose-built large language model inference processor – and it points directly at where the custom compute market is heading next.
This is different from NVIDIA’s game. NVIDIA dominates general-purpose training. Broadcom dominates the next layer – the bespoke, high-efficiency inference silicon that hyperscalers increasingly want for cost and performance reasons.
- Q2 AI semiconductor revenue: $10.8B, up 143% YoY
- Q3 AI semiconductor guidance: ~$16B, implying 200%+ YoY growth
- Full-year 2026 AI revenue target: Reiterated, significant acceleration
- FY2027 AI chip revenue target: Greater than $100 billion
- AI-specific backlog: $73 billion
That backlog figure matters. It means the revenue is not speculative. Broadcom has described this as an AI backlog covering roughly the next 18 months, and has discussed multi-year relationships and supply agreements with major customers in this buildout. The pipeline has visibility that most semiconductor companies can only dream about.
The Other Business Nobody Talks About
Infrastructure software – anchored by the VMware acquisition – contributes roughly a third of Broadcom’s total revenue. This segment generates recurring subscription revenue with sticky enterprise customers across large banks, government agencies, and cloud providers. It is not exciting. It is also not going away.
What the software business provides is downside cushion. Even in a hypothetical scenario where hyperscaler AI capex growth slows, Broadcom still has a massive, cash-generative software franchise that keeps the free cash flow engine running. The company also raised its dividend for the 15th consecutive year. That’s a management team that believes in the durability of the business.
Risks Worth Naming
The customer concentration issue is real. A handful of hyperscalers drive the majority of AI chip revenue, and any shift in their spending – whether from economic pressure, policy change, or in-house chip development – would show up quickly in Broadcom’s numbers.
There’s also a fresh competitive signal: reports indicate Google has tapped MediaTek to build its next-generation TPUv9 chip, codenamed Triggerfish, which is widely seen as evidence that no custom chip relationship is guaranteed to be exclusive forever.
And then there’s valuation. At roughly 27x forward earnings, AVGO isn’t cheap in absolute terms – even if it looks discounted against where it traded at its $495 peak. A broader tech sector de-rating or a miss against the $100 billion FY2027 AI revenue target would test investor patience quickly.
The Part People Keep Skipping Over
The $35 billion AI infrastructure financing platform that Broadcom launched alongside Apollo and Blackstone – the AI XPV Platform – barely made headlines. It’s designed to enable more than 20 gigawatts of AI compute capacity through 2028, initially launching with a $35 billion tranche to facilitate Anthropic’s previously announced capacity expansion expected to deploy in Fluidstack-based sites starting in mid-2026. This isn’t just a chip sale. It’s Broadcom positioning itself as a structural counterparty in the buildout of AI infrastructure at a scale most investors haven’t fully processed.
The selloff created a window. Whether that window stays open depends on what Q3 numbers look like when they land – and whether $16 billion in AI chip revenue is a floor or a ceiling.
That’s the thing investors are actually watching right now.
This article is for informational purposes only and does not constitute investment advice. Past performance is not indicative of future results. Investing involves risk, including the possible loss of principal. Always conduct your own due diligence before making any investment decisions.
