SAN JOSE — When OpenAI unveiled Jalapeño, its first custom inference processor, in June, as TechCrunch reported, the announcement confirmed something Broadcom Inc.’s investors already knew: the company that built the chip was not Nvidia Corp. It was Broadcom. Custom silicon for the world’s leading AI laboratory, engineered to run faster and cheaper than a graphics processing unit. That is the business Broadcom has built, and the stock should reflect it. At $358.58 on Monday, it sits $136 below where it traded at its 52-week peak.
Broadcom Inc. (AVGO) closed Monday at $358.58, a decline of $0.18, or 0.05 percent, from Friday’s close of $358.76. The session opened at $360.99, reached an intraday high of $362.99, and found a low of $356.09. Volume was 8.19 million shares, against a daily average of 26.2 million, less than a third of typical activity. That compression is meaningful: on a day when the broader market was thinly traded and the macro focus was squarely on Federal Reserve Chair Kevin Warsh’s upcoming Jackson Hole remarks Friday, Broadcom’s stock moved very little. The market cap stands at approximately $1.706 trillion.
The 52-week range tells the fuller story. The low is $287.17; the high is $495.00. At $358.58, the stock is 28 percent below its annual peak and 25 percent above its trough, a position that places it squarely in the middle of the range, which is unusual for a company reporting artificial intelligence revenue growth of 143 percent year-over-year. The gap between what Broadcom’s AI business is doing and where its equity is trading reflects a question the market has not finished answering.
That question starts with Marvell Technology Group. Alphabet Inc., which has been one of Broadcom’s largest ASIC customers for its Tensor Processing Units, began deepening its engagement with Marvell for a new generation of custom chips. Alphabet still works with Broadcom, but the possibility that its largest hyperscaler customer is developing an alternative ASIC supply chain introduces a concentration risk that did not exist twelve months ago. Broadcom currently counts Google, Meta Platforms Inc., ByteDance Ltd., and OpenAI among its AI chip partners. Losing meaningful share at even one of those relationships would compress the forward AI revenue trajectory in ways that the current multiple cannot absorb easily.
The valuation picture is layered. Broadcom’s trailing price-to-earnings ratio is 59.47 on earnings per share of $6.03. Its forward P/E, at 19.19, reflects the market’s expectation of substantial earnings growth ahead, growth that the AI revenue line, up 143 percent year over year, is clearly positioned to deliver assuming the customer base holds. That compression from 59x trailing to 19x forward is the bet. BMO Capital Markets formalized its own view on August 21, initiating coverage with an Outperform rating and a price target of $455. The consensus analyst target across the coverage universe sits at $526.30, with a range running from $215.88 to $675.00. The spread in that range reflects genuine uncertainty, not disagreement about the company’s near-term results.
What makes Broadcom’s ASIC business structurally different from selling standard chips is the depth of the engineering integration required. Each custom chip is co-designed with the hyperscaler to fit its specific model architecture, inference latency requirements, and power envelope. That process takes years and, once completed, creates a switching cost that is not easily unwound. OpenAI’s Jalapeño chip and Apple Inc.’s forthcoming wireless silicon, part of a $30 billion multiyear commitment Apple made to produce chips with Broadcom in the United States, as TechCrunch reported, represent the kind of multi-year, high-specificity relationships that are not easily replicated by a competitor. The Marvell risk is real. The switching cost is also real. Both are simultaneously true.
The macro overlay complicates the near-term read. Warsh’s remarks at Jackson Hole on Friday will be the first significant signal of the Federal Reserve’s posture since his nomination was confirmed. His reputation as a policy hawk drove a ten percent single-session drop in gold when his appointment was announced in February. Equity markets have since recalibrated, but a Friday speech that suggests higher-for-longer rates would disadvantage rate-sensitive growth equities. Technology and semiconductor names carry elevated multiples. Broadcom, at 59x trailing earnings, is not immune to a repricing if duration risk re-enters the conversation. The S&P 500 today gained 0.32 percent, but the session’s lightness suggested institutional investors are waiting rather than positioning.
The broader semiconductor sector is navigating a period in which AI infrastructure spending is enormous but concentrated. Alphabet, Microsoft Corp., Meta, and Amazon.com Inc. are committing capital to custom silicon in amounts that benefit Broadcom today and may benefit Marvell tomorrow. The Alibaba AI infrastructure raise of $10.2 billion in August reflects the scale of investment flowing through the sector, investment that ultimately finds its way into custom ASIC procurement. Whether Broadcom captures a disproportionate share of the next wave, as it did the first, is not settled.
The Nasdaq Composite climbed 0.57 percent on Tuesday, with technology components outperforming the broader index. Broadcom’s muted session was not a reflection of sector weakness. It was a reflection of a stock that has priced in substantial AI revenue growth, absorbed the Marvell competitive headline, and is now watching to see what Friday brings before making its next directional move.
What Monday’s close left open: whether Warsh’s Friday remarks reset duration assumptions for high-multiple growth names, whether Alphabet’s engagement with Marvell deepens or remains a secondary relationship, and whether the AI infrastructure spending cycle has enough runway at current chip prices to justify the forward earnings estimates on which Broadcom’s 19x forward P/E depends. Those questions will not close before the Jackson Hole speech. They may not close after it.
Related Posts
