Qualcomm's hyperscaler ASIC win changes the chip map

Qualcomm just landed a hyperscaler ASIC deal, and Broadcom should be nervous

Qualcomm (NASDAQ:QCOM) has spent years telling investors it was serious about the data center. Most of those investors filed it under 'we'll see.' On Wednesday's Q3 2026 earnings call, the we'll-see period ended.

Cristiano Amon confirmed Qualcomm has a custom ASIC engagement with a large hyperscaler, with initial shipments starting in the December quarter and what he described as a multi-generation relationship with accretive operating margins. He paused for 3.1 seconds before describing it as 'a custom product we're working with a hyperscaler,' which is the kind of pause you hear when someone is carefully choosing what not to say rather than searching for the right words. The substance on either side of that pause was clear enough: Qualcomm has cracked open a market that Broadcom (NASDAQ:AVGO) and Marvell (NASDAQ:MRVL) have owned without serious competition.

The bull case on this is straightforward enough to state and important enough not to undersell. Custom silicon for hyperscalers is one of the highest-margin, highest-stickiness businesses in semiconductors. Once a hyperscaler designs a chip around your IP and your team, switching costs compound with every generation. Broadcom's custom ASIC franchise, which powers Google's TPUs and Meta's MTIA chips among others, is one of the reasons the stock trades where it does. Marvell built its entire re-rating story on winning hyperscaler custom work. Now Qualcomm has its first win in the door, with a client large enough to matter and a relationship structure that explicitly spans multiple chip generations.

The AlphaWave acquisition, which Qualcomm completed in 2023 largely to get SerDes IP and high-speed connectivity technology, looks much smarter today than it did when it happened. SerDes is exactly what you need to compete for custom ASIC work at the speeds hyperscalers demand. Amon clearly knew what he was building toward.

The near-term handset numbers are genuinely messy. Qualcomm guided Q4 revenue at $10.2 to $10.8 billion with QCT handset revenues down sequentially, and CFO Akash Palkawala's filler rate in Q&A climbed from roughly one per minute in prepared remarks to over four per minute once analysts started pressing on hyperscaler specifics. That asymmetry is telling: Amon stayed crisp on data center questions, Palkawala's cadence roughened. The handset softness is real, and the CFO knows the quarterly bridge there is uncomfortable.

But the handset cycle turns. Custom ASIC relationships, if you can get them, don't. The market is pricing QCOM primarily as a handset chipmaker with a promising automotive side business. That framing is now missing a third leg, and that third leg happens to be the highest-multiple business in semiconductors.

For Broadcom and Marvell, the response matters. Both have built their custom silicon franchises on the implicit assumption that Qualcomm's IP and execution weren't ready for this market. One confirmed hyperscaler win challenges that assumption publicly, and hyperscalers talk to each other about who's delivering. If Qualcomm's December shipments go cleanly, the pipeline conversation in early 2027 will look very different from anything the current consensus is modeling.

The stock is not obviously cheap at roughly 18x forward earnings, but that multiple was built for a handset business. A credible data center ASIC franchise, even in its first generation, typically commands significantly more. The re-rating, if the December ramp executes, has room to run.

✦

Quick Takes

Three more stories worth knowing about

CBRE's $700M data center quarter is the best proof the AI buildout hasn't peaked

CBRE Group (NYSE:CBRE) printed $700 million in data center services revenue in Q2 2026 alone, with CEO Bob Selentik laying out a specific, auditable path to a $10 billion infrastructure business generating over $1 billion of EBITDA by 2030. That is not marketing language. When Microsoft, Google, Amazon, and Meta are collectively pushing that volume of work through a single services provider on a quarterly basis, the argument that hyperscaler AI capex is plateauing runs directly into this number. Turner and Townsend, CBRE's project management arm, has posted 30%-plus data center growth for a decade running. The Q4 comp will be punishing given last year's data center land sale profits, and Middle Eastern capital raising came in light. But the underlying infrastructure demand is not slowing.

Boston Scientific's Watchman problem is Johnson and Johnson's opportunity

Boston Scientific (NYSE:BSX) CEO Mike Mahoney said 'I take responsibility for the guidance reduction' twice on Wednesday's call. That alone is worth paying attention to from a management team that almost never operates this way. The numbers behind it: full-year organic growth cut to 5-6%, 2027 flagged as a year of limited EPS growth, and Watchman facing mid-to-high single-digit declines in the second half with no 2027 recovery assumed. US electrophysiology is also conceding share. BSX has a genuine 2028 bull case built around seven new platform launches across a $25 billion-plus addressable market. But 2028 is two years away, and the competitive beneficiaries of the share concession BSX just made, specifically Johnson and Johnson (NYSE:JNJ) and Medtronic (NYSE:MDT) in EP, are sitting in a better position today than they were 48 hours ago.

Entergy's data center moat is written into Louisiana law now

Entergy (NYSE:ETR) reported $1.03 adjusted EPS in Q2 2026, slightly below last year on weather normalisation but with industrial load growth running at 10% on data center ramp-up. The more consequential piece of this call was regulatory, not financial. CEO Drew Marsh spent significant time explaining that the Louisiana Governor's executive order and the White House Ratepayer Protection Pledge are moving from political goodwill into enforceable standards tied to state sales tax exemptions for data center loads. A utility that has written regulatory protections into law sits in a meaningfully different competitive position than one competing on relationships and reputation alone. The 7-12 GW pipeline guidance was unchanged, but Marsh described incremental interest since Investor Day as 'significant' and 'a lot of volume.' The gap between indications of interest and signed agreements is where the next earnings revision lives.

✦

Overlooked Stock

CRDO

Credo Technology Group Holding Ltd

Credo Technology Group (NASDAQ:CRDO) makes the high-speed connectivity chips that sit between servers inside a hyperscaler data center. Specifically, active electrical cables, optical signal processors, and SerDes chiplets that let data move fast enough for AI training clusters to function at the speeds the models demand. The score here is 93, which puts it in the same tier as FICO and just behind Celestica. The business sells directly to hyperscalers, OEMs, and optical module manufacturers, which means its revenue is a direct function of AI infrastructure spending rather than a derivative of it. With CBRE printing $700 million of data center services in a single quarter and Qualcomm confirming hyperscaler ASIC engagement starting in December, the read-through for companies supplying the actual connectivity fabric inside those facilities is worth working through. Credo has been growing fast off a small base, and the competitive advantage in SerDes IP is real and hard to replicate quickly. Worth a closer look at what the revenue trajectory looks like into 2027.

The Contrarian Take

Everest Group buying back $300M of stock per quarter is a better capital allocation story than most tech buybacks right now

The reinsurance sector gets lumped in with boring institutional finance and largely ignored by growth-oriented investors. Everest Group (NYSE:EG) raised its buyback floor to $300 million per quarter on its Q2 2026 call, which CEO Jim Williamson described in plain terms: the company sees more value returning capital than deploying it into the current reinsurance market. That is a specific statement about pricing discipline and balance sheet confidence, not boilerplate. A 16.7% operating ROE and $567 million of net investment income in a single quarter suggest this is a business generating cash faster than it can find attractive paper. Compare that to most technology companies buying back stock with borrowed money to prop up EPS while the underlying business dilutes aggressively through stock compensation. Everest is doing the opposite: exiting poorly priced casualty books, rotating toward short-tail and specialty, and returning genuine surplus capital. The AIG commercial retail transfer progressing on schedule means meaningful capital release in H2 2026. The stock is priced like a commodity insurer. The underlying mechanics look more like a disciplined allocator running a clean balance sheet.

Next briefing

SpaceX's $1.75T IPO makes Rocket Lab the obvious next trade