Marvell Jumps 5% as Google's Custom Chip Work Opens Up
Marvell climbed 5.23% to $241.29 while Broadcom slipped 0.58% to $356.68, as a second supplier for Google's custom AI silicon reframes what had been a single-vendor story.
Marvell shares rose 5.23% to $241.29 on 25 August 2026 as the company's push into custom accelerator design challenged Broadcom's position as the near-exclusive partner on Google's in-house AI chips, with Broadcom down 0.58% to $356.68.
The market spent Tuesday repricing one of the tighter relationships in artificial-intelligence hardware. Marvell Technology (NASDAQ: MRVL) rose 5.23% to $241.29, having traded as high as $246.70 and as low as $235.87, while Broadcom Inc. (NASDAQ: AVGO) eased 0.58% to $356.68 against a previous close of $358.76. Alphabet (NASDAQ: GOOG) was quiet, off 0.25% at $343.73. All prices are as of the last trade at 18:03 GMT on 25 August 2026, with the market still open.
The move was not about a quarter or a guidance revision. It was about who gets to design the silicon inside Google's data centers — and whether that work, long treated as effectively Broadcom's to lose, is becoming a contest with more than one bidder.
What Marvell's expansion actually threatens
Custom silicon — an application-specific integrated circuit, or ASIC, built to one customer's specification rather than sold off the shelf — has been the most defensible corner of the AI buildout. Broadcom's value in that business has never rested purely on transistor design. It rests on serialiser-deserialiser blocks, packaging, networking fabric and, above all, years of co-engineering with a single customer's software stack. That accumulated integration is the thing a rival cannot buy in a quarter.
Marvell's expansion into the same category, as GuruFocus framed it, weakens the exclusivity story without erasing the engineering. Those are two different claims and investors should keep them apart. Exclusivity is a pricing argument. Integration is a switching-cost argument. The first can break long before the second does.
Practically, a hyperscaler that qualifies a second design partner gains leverage on price and schedule even if it never moves a single production program. The mere existence of a credible alternate bidder changes the shape of the next negotiation. That is what Tuesday's tape appears to be discounting: not lost revenue at Broadcom, but a thinner premium on the revenue that stays.
Reading the spread between the two stocks
The relative move is more informative than either absolute price. Marvell gained 5.23% on the session while Broadcom lost 0.58% — a spread of 5.81 percentage points between the two on a day when the Nasdaq 100 proxy (QQQ) added just 0.40% to $709.16 and the S&P 500 tracker (SPY) rose 0.20% to $764.99. Neither name was following the index. This was a sector-internal reallocation.
The asymmetry of the two reactions is telling. A percentage point of incremental custom-ASIC share is worth far more to the smaller franchise than the same point is costly to the larger one. That mechanically produces the pattern seen on Tuesday: a sharp bid for the challenger, a shrug for the incumbent. Broadcom finishing the day inside its own range — a low of $356.09 against a high of $362.99 — looks less like a repudiation than a modest de-rating of the exclusivity premium.
Alphabet's near-flat print is consistent with that reading. If a customer secures a second qualified supplier for a critical component, the customer is the structural winner over time, but the benefit arrives slowly, through cost of goods rather than through any single announcement. A 0.25% dip on the day says the market sees nothing to reprice today.
The part nobody can measure from the outside
Here is the honest limit of the analysis. Neither Broadcom nor Alphabet discloses the customer-level split of custom-silicon revenue with enough granularity for an outsider to say what fraction of Broadcom's ASIC business runs through a single hyperscaler. Estimates circulate; they are estimates. Any figure that gets quoted confidently on this question is inference dressed as disclosure.
What can be said without inventing anything:
- Custom accelerator programs run on multi-year design cycles, so a second supplier qualified today shows up in shipped volume well after it shows up in a share price.
- Dual-sourcing typically compresses gross margin on the contested product before it takes any unit volume away, because the incumbent defends the socket on price first.
- Networking content — the switching and interconnect silicon that surrounds an accelerator — is a separate revenue pool and does not automatically travel with the accelerator award.
- The engineering integration Broadcom has accumulated is real, non-transferable and the reason a share shift would be gradual rather than abrupt.
What would confirm or kill the thesis
Any figure that gets quoted confidently on this question is inference dressed as disclosure.
Three things are worth watching, none of which will be visible in a daily quote. First, gross margin commentary from Broadcom on custom silicon specifically: a deliberate distinction between semiconductor solutions margin and blended margin would signal price pressure arriving. Second, the language Marvell uses about design wins — whether programs are described as production-qualified or still in development, because the gap between the two is often years. Third, capital-expenditure disclosure from Alphabet, since a customer expanding its supplier bench while also expanding its total spend can enrich both vendors at once. Dual-sourcing in a growing market is not a zero-sum outcome.
The framing that lost value on Tuesday was the simplest one: that a single vendor owns a single customer's silicon roadmap outright. That framing was always more useful to storytelling than to underwriting. What replaces it is messier — a franchise with deep, sticky engineering advantages and a slightly less generous pricing environment, trading at $356.68 with the stock down less than a percent on the news.
For Marvell holders, the 5.23% move prices optionality rather than earnings. Optionality is legitimate; it is also the first thing to be repriced if the design wins turn out to be narrower than the tape assumed. The custom-chip business has become a contest. Contests have losers on both sides of the trade, depending on which quarter you check.
Frequently asked questions
Why did Marvell stock rise while Broadcom fell?
Marvell gained 5.23% to $241.29 on its expansion into custom chip design work, while Broadcom slipped 0.58% to $356.68. The market read a second credible supplier for hyperscaler custom silicon as adding revenue optionality for Marvell and removing an exclusivity premium from Broadcom, even though Broadcom's existing engineering integration remains intact.
What is a custom ASIC and why does it matter here?
An application-specific integrated circuit is a chip designed to one customer's specification rather than sold as a standard product. For AI data centers, hyperscalers commission ASICs as accelerators tuned to their own software. The business is defensible because it requires years of co-engineering with the customer, making suppliers hard to displace quickly.
How much of Broadcom's custom chip revenue comes from Google?
Neither company discloses customer-level revenue splits for custom silicon at a granularity that allows an outside figure to be stated as fact. Estimates circulate publicly but are inferences, not disclosures. Any confident number on this question should be treated as an assumption rather than reported financial information.
Does a second supplier immediately cut Broadcom's revenue?
Not immediately. Custom accelerator programs run on multi-year design cycles, so a newly qualified supplier appears in shipped volume long after the announcement. The typical first effect is margin pressure, as the incumbent defends its position on price, rather than an outright loss of unit volume.
Why did Alphabet stock barely move?
Alphabet closed the session down 0.25% at $343.73. A customer that secures a second qualified supplier for a critical component gains bargaining leverage, but the benefit arrives gradually through cost of goods sold rather than in a single announcement. The market saw nothing requiring immediate repricing.
What should investors watch next in this story?
Three signals matter: Broadcom's commentary on custom silicon gross margin specifically, Marvell's language on whether design wins are production-qualified or still in development, and Alphabet's capital expenditure disclosure. A growing total market means dual-sourcing can enrich both vendors rather than being zero-sum.
Sources
Photo: Sergey Sergeev · Pexels Licence — source
