Hey there, bargain hunter.
Broadcom just had one of the busiest months in its history. Record earnings. A landmark chip deal with OpenAI. A $35 billion AI compute financing platform backed by Apollo and Blackstone. And the stock is sitting roughly 25% below its peak.
That gap is worth understanding.
What actually happened
On June 3, Broadcom (AVGO) reported fiscal Q2 2026 results. Total revenue came in at $22.2 billion, up 48% year over year. AI semiconductor revenue hit a record $10.8 billion, up 143% year over year. The company previously announced a $10 billion share repurchase authorization, and it announced another quarterly dividend increase.
The stock dropped anyway.
Two things spooked the market. First, the infrastructure software division posted $7.18 billion in revenue, missing analyst expectations of $7.32 billion. Second, the Q3 AI semiconductor revenue guide of $16 billion came in below some analyst estimates. Never mind that $16 billion would still represent over 200% growth year over year. The market had priced in perfection, and anything short of perfection was punished.
Then came the Jalapeño announcement.
The OpenAI chip changes the story
On June 24, OpenAI and Broadcom unveiled Jalapeño, OpenAI’s first custom-built AI inference chip, designed and brought to production in collaboration with Broadcom, with Celestica helping industrialize the platform through board and rack system integration and scalable production systems. The chip was built from scratch in nine months. Engineering samples are already running machine-learning workloads in the lab. Early testing shows performance per watt substantially better than current state-of-the-art alternatives.
Inference is where the real volume is. Training gets the headlines, but inference is what happens billions of times a day every time someone queries ChatGPT or runs a Codex task. Making inference cheaper and faster is the actual economic lever, and Broadcom is now OpenAI’s partner on that problem.
OpenAI has said it plans to start using Jalapeño to handle customer queries later in 2026.
Slight tangent, but it matters here: the real question is not whether Broadcom can build the chip. It already did. The question is whether OpenAI keeps this chip exclusive to its own inference workload or eventually opens it to third-party customers. I could not verify the claim that Reuters reported OpenAI intends to use the chips only for internal use, so treat that as unconfirmed. If the chips remain primarily internal, it limits the near-term revenue ceiling for Jalapeño specifically, even as it locks in a massive, growing customer.
The numbers that matter
- Q2 revenue: $22.2 billion (+48% YoY)
- AI semiconductor revenue: $10.8 billion (+143% YoY)
- Q3 AI semiconductor revenue guide: $16 billion
- Gross margin (LTM): approximately 68% (not ~76%)
- Market cap: approximately $1.8 trillion
- 52-week range: $262.66 to $495.00
- Current price (as of June 30): $372.45
- Consensus 12-month analyst price target: not verified (varies by source)
- Custom chip customers confirmed: Broadcom has said it has six major custom AI chip customers; OpenAI, Anthropic, Google, and Meta have been publicly named, while the remaining customers have not been consistently identified in primary sources.
The valuation question
At roughly 62 times trailing earnings, Broadcom is not cheap by traditional metrics. That is the honest answer. But it is cheaper than it was at $495. Analysts at JPMorgan maintain a $580 price target. The Wall Street consensus sits around $493 to $524, depending on the source. That implies somewhere between 30% and 40% upside from current levels.
What justifies the premium? The VMware software layer, which Broadcom absorbed through its 2023 acquisition, carries structurally higher margins than silicon. Every enterprise that renews a VMware Cloud Foundation deal converts legacy infrastructure spend into recurring software revenue. I could not verify the specific claim that LSEG signed a five-year VMware renewal, so treat that detail as unconfirmed. The broader point stands: Broadcom’s infrastructure software segment is a large, recurring-revenue base alongside the AI semiconductor ramp.
The bear case is real though. AI chips can carry slimmer margins than parts of Broadcom’s networking portfolio because they depend heavily on costly memory and advanced packaging. As AI semiconductors climb toward a larger share of total revenue, margin pressure is real. I could not verify the claim that Google is shifting its next-generation TPU chip to MediaTek, so treat that as unconfirmed. And at 60-plus times earnings, any guidance softness is expensive.
What I’m watching
The next earnings report is expected on September 3, 2026. Between now and then, the key variables are whether the VMware integration keeps expanding margins, whether any of the six custom chip customers announce delays, and whether Jalapeño’s technical performance report (expected in coming months) confirms the early efficiency claims.
The stock peaked near $495 on June 3. It closed June 30 at $372.45. At that level, the market is pricing in continued AI revenue growth but demanding execution. Broadcom has executed. The question is whether it can keep doing it at a premium multiple when competition from Marvell, and increasingly from in-house hyperscaler silicon efforts, is intensifying.
If the Q3 guide converts cleanly and the Jalapeño timeline holds, this is a different stock by late 2027. If hyperscaler capex softens or Google keeps diversifying chip suppliers, the multiple has more room to compress from here.
The business has never been better. The stock is cheaper than it was a month ago. Those two facts are worth sitting with.
