Marvell Swings 3.9% in a Session, Broadcom Just 1.4%
Both chipmakers are levered to the same AI build-out, but their last sessions looked nothing alike: Marvell's intraday swing was nearly three times Broadcom's.
Marvell Technology (MRVL) closed at 261.94, up 1.15%, after trading a 257.43–267.48 range, while Broadcom (AVGO) closed at 352.81, up 0.70%, in a far tighter 349.43–354.43 band on Sept. 25, 2026, as both chipmakers ride the same AI infrastructure build-out.
Two names dominate the custom-silicon side of the artificial intelligence build-out, and they are not interchangeable. Broadcom (AVGO) and Marvell Technology (MRVL) both design the application-specific chips and networking parts that hyperscale data centers order in volume, and both have been pulled higher by the same wave of AI infrastructure spending. What separates them, for an investor deciding where the next dollar goes, is not the direction of the business but the amount of turbulence attached to it.
The most recent session makes the point cleanly. As of the last trade on Friday, Sept. 25, 2026, Broadcom closed at 352.81, up 0.70% from a prior close of 350.36, with the day spent inside a band of 349.43 to 354.43. Marvell closed at 261.94, up 1.15% from 258.95 — the bigger gain — but it got there the hard way, printing a high of 267.48 and a low of 257.43. On an illustrative basis, that intraday range works out to roughly 3.9% of Marvell's prior close, against about 1.4% for Broadcom. Same theme, same day, nearly three times the churn.
Why the two stocks respond differently to the same news
The AI infrastructure trade is a single story told through very different balance sheets. Broadcom is the larger, more diversified of the pair: custom accelerators and networking silicon sit alongside a substantial infrastructure software business, which means a given piece of AI news moves a smaller share of the whole. Marvell is the more concentrated expression of the same idea — a higher proportion of its story rides on custom silicon programs and the data center, so sentiment shifts land harder.
That is what a wide intraday range is actually measuring. It is not a verdict on the business; it is a record of how far buyers and sellers were willing to move the price before the close settled the argument. Marvell traded through a span of more than ten points on the day and finished near the lower half of it, which is the signature of a stock where the marginal holder is still debating the terms.
The comparison that matters over a holding period is customer concentration. Custom-chip programs are won account by account, and a single hyperscaler's design decision can reset a multi-year revenue line. The more of a company's growth that sits inside a handful of programs, the more binary each renewal becomes. That asymmetry, rather than any difference in end-market demand, is the core of the risk-reward argument that Motley Fool lays out in favor of one of the two names.
Both outran the tape, one by more
Neither stock was simply carried along by the market. The S&P 500, via SPY, closed at $771.35, up 0.54% on the day. The Nasdaq 100 proxy QQQ finished at $744.50, up 0.46%, and the Dow 30 tracker DIA closed at $517.49, up 0.94%. Broadcom's 0.70% gain and Marvell's 1.15% both cleared the Nasdaq 100's move, which is the benchmark that matters most for large-cap semiconductor exposure.
That is a meaningful detail in a market where the AI complex has periodically traded as one instrument. When chip names move only with the index, there is no information in the price. When they separate — and separate by different amounts, with different intraday paths — the market is pricing company-specific outcomes again. Friday looked like the second case.
What an investor is actually choosing between
Framed plainly, the decision comes down to three variables, none of which requires a forecast to understand:
- Breadth of revenue. Broadcom's software and legacy infrastructure businesses dilute the AI cycle in both directions — less upside torque, less downside when orders slip.
- Customer concentration. The fewer the programs, the more each one matters. This is the single largest swing factor in custom-silicon economics.
- Price volatility as a cost. An investor who cannot tolerate a near-4% intraday range will sell it at the wrong moment, which turns volatility into a realized loss rather than a paper one.
Valuation sits on top of all three. Both stocks have re-rated on the strength of the AI capital cycle, and a higher multiple on a more concentrated revenue base compounds risk rather than offsetting it. The investor buying the narrower business is, by construction, buying a larger bet on fewer decisions made inside a small number of purchasing departments.
The tells to watch from here
The investor buying the narrower business is, by construction, buying a larger bet on fewer decisions made inside a small number of purchasing departments.
The next real information comes from order visibility, not from daily prints. Specifically: whether custom accelerator programs are being extended or expanded at named hyperscalers, whether networking attach rates hold as rack architectures change, and whether either company's disclosed backlog stretches further out than it did a quarter ago. Backlog duration is the cleanest available proxy for how durable this cycle is, because it reflects commitments already made rather than demand still being modeled.
Two secondary signals are worth tracking. The first is whether Marvell's intraday ranges compress toward Broadcom's over coming sessions. Persistent wide ranges mean the market has not settled on a valuation, and that unsettled state tends to resolve violently in one direction. The second is relative performance against QQQ. If both names begin tracking the index tick for tick again, the AI trade has gone back to being a single macro position, and stock selection between the two will stop paying.
Nothing in the last session resolves the question of which is the better buy. What it does establish is that the two are not substitutes. Broadcom offered a smaller gain inside a tight band; Marvell offered a larger gain after a session that traveled nearly three times as far. An investor who treats those as the same exposure, sized the same way, is taking on risk they have not consciously priced. The AI build-out is large enough to fund both businesses — but portfolio construction still has to choose.
Key facts
- AVGO last close: 352.81, +0.70%, as of Sept. 25, 2026, 20:00 GMT
- MRVL last close: 261.94, +1.15%, as of Sept. 25, 2026, 20:00 GMT
- Intraday ranges: AVGO 349.43–354.43; MRVL 257.43–267.48
- Benchmark: QQQ closed $744.50, +0.46%; SPY $771.35, +0.54%
Frequently asked questions
How did Broadcom and Marvell close in the most recent session?
As of the last trade on Friday, Sept. 25, 2026, Broadcom closed at 352.81, a gain of 0.70% from its prior close of 350.36. Marvell Technology closed at 261.94, up 1.15% from a prior close of 258.95. Both finished ahead of the Nasdaq 100 proxy QQQ, which rose 0.46% on the day to $744.50.
Which stock was more volatile during the day?
Marvell, by a wide margin. It traded between 257.43 and 267.48, while Broadcom stayed within a band of 349.43 to 354.43. On an illustrative basis those ranges equal roughly 3.9% and 1.4% of each stock's prior close respectively, meaning Marvell's intraday swing was close to three times Broadcom's despite both being AI infrastructure names.
Why does customer concentration matter for custom chip makers?
Custom silicon is sold program by program to a small number of hyperscale data center operators. A single design win or loss can reset a multi-year revenue line. The fewer programs a company depends on, the more binary each renewal decision becomes, which raises the volatility of both earnings and the share price regardless of overall AI demand.
Do Broadcom and Marvell compete directly?
They overlap substantially in custom accelerators and data center networking silicon, the two product categories most directly tied to AI infrastructure spending. Broadcom is the broader company, with an infrastructure software business alongside its chip operations, which dilutes the impact of any single AI development on its overall results in both directions.
What should investors watch next on these two stocks?
Order visibility rather than daily prices: whether custom accelerator programs are being extended at named hyperscale customers, whether networking attach rates hold as rack designs change, and whether disclosed backlog is stretching further out. Also watch whether Marvell's intraday ranges compress toward Broadcom's, which would signal the market settling on a valuation.
How did the broader market trade that session?
All three major benchmarks finished higher. The S&P 500 tracker SPY closed at $771.35, up 0.54%. The Nasdaq 100 proxy QQQ closed at $744.50, up 0.46%. The Dow 30 tracker DIA closed at $517.49, up 0.94%, the strongest of the three. Both chip names outperformed the Nasdaq 100 benchmark.
