MRVL — Marvell Technology
Investor Day finally put a per-share target behind the AI story. Now it has to survive reality.
Marvell sells custom compute silicon plus the connectivity around hyperscale AI systems. The Oct. 6 Investor Day materially improved the underwriting because management finally put a per-share outcome behind the growth story: $70B–$90B of FY2031 revenue and more than $30 of non-GAAP EPS. I still haircut those targets. The important change is that the denominator now has something concrete to earn toward.
The problem I had with Marvell was never whether the AI business was growing.
Why I care about this one
The easy version of the Marvell thesis is “hyperscalers want custom chips.”
True, but Investor Day made the connectivity side harder to treat like supporting cast.
At the roughly $80B FY2031 midpoint, management sees about $30B of custom revenue and roughly $37.5B of interconnect revenue, with another ~$10B from switching and storage. (Barron’s)
That matters because the best version of Marvell is not one company betting everything on one accelerator program.
Custom compute can bring Marvell into the architecture. Then optics, DSPs, switching, PCIe/CXL and scale-up connectivity can attach around the same systems. Celestial AI and XConn make more strategic sense in that version of the business than they do as isolated acquisitions.
The current operating numbers are already moving in that direction. Q2 FY2027 revenue was $2.739B, up 37% year over year, while data-center revenue grew 46% to $2.172B and represented 79% of the company. Q3 guidance is about $3.15B of revenue and $1.10 of non-GAAP EPS. (Q2 FY2027 results)
There is still customer concentration. Four customers represented 72% of gross accounts receivable at the end of Q2. More products per hyperscaler do not make hyperscaler bargaining power disappear. (Q2 FY2027 10-Q)
This is the part I still refuse to wave away.
Celestial cost about $3.5B and included roughly 24.5M Marvell shares. NVIDIA’s $2B preferred can convert into common stock. Google’s expanded agreement came with a warrant for up to roughly 59M shares. Most of that is performance-linked rather than free dilution: after about 1.36M time-based shares, the remaining tranches vest one per $500M of qualifying product revenue, implying roughly $120B of qualifying purchases for full performance vesting. SBC is also large. (Q2 FY2027 10-Q; Google agreement)
The good news is that management is finally talking in EPS, not only revenue.
If Marvell actually earns more than $30 per share in FY2031, the denominator obviously did not ruin the story.
The annoying footnote is that this is non-GAAP EPS. Stock-based compensation is still an economic cost even when management excludes the expense from the metric. And a four-year target can look beautifully precise right up until the world changes.
So the >$30 number gets credit. It does not get blind faith.
Fiscal Q2 2027 + Oct. 6, 2026 Investor Day
Primary sources: Marvell Investor Day 2026 ↗ · Investor Day revenue targets — Reuters ↗ · Investor Day EPS / margin / mix — Barron's ↗ · Fiscal Q2 2027 results ↗ · Fiscal Q2 2027 10-Q ↗ · Google custom-silicon agreement ↗
At its October 6 Investor Day, management outlined a FY2031 revenue target of $70B–$90B and greater than $30 of non-GAAP EPS. That gives the custom-silicon and interconnect thesis a measurable per-share checkpoint, but does not erase the dilution risk. My independent FY2031 Base assumes around $55B revenue and $20 EPS at 26x, or roughly $520.
My position
What I actually bought and sold.
| Date | Type | Account | Dollars | Shares | Price | Return now* |
|---|---|---|---|---|---|---|
| Dec 12, 2025 | Buy | Taxable | $3,000 | 33.963 | $88 | +224.9% |
| Jan 29, 2026 | Reinvest | Taxable | $2.04 | 0.024 | $85 | +237.7% |
| Apr 30, 2026 | Reinvest | Taxable | $2.04 | 0.013 | $157 | +82.9% |
| Jul 30, 2026 | Reinvest | Taxable | $2.04 | 0.012 | $170 | +68.8% |
*Return now uses the latest available quote for buys and reinvestments. Sales and fees show “—”.
Valuation & scenario math
Capital Allocation stays at 6.75/10.
The strategic reinvestment looks better today. A larger connectivity opportunity gives Celestial and XConn more ways to earn their keep, while custom silicon is scaling into a real business rather than a collection of design-win slides.
What has not changed is how much equity and dilution management has been willing to use to get there.
Maybe that ends up being exactly the right trade.
The score moves when the returns show up, not when the Investor Day deck says they will.
FY2031 per-share earnings + custom/interconnect conversion + dilution
No static ladder. The Investor Day materially improved expected return, but I already own roughly a 3% position and a lot of correlated AI infrastructure elsewhere. At about $297, I am holding rather than adding. More capital needs either a better quote or enough FY2028/FY2029 evidence that the $23 base-case EPS bridge starts looking conservative.
Investor Day gave me useful evidence, not a guaranteed outcome. My scenarios start with the business I can independently underwrite, then compare that result with management's $70B–$90B revenue and greater-than-$30 EPS framework for FY2031.
| Normalized cycle EPS | ~$20 |
| Terminal P/E | 26x |
| Checkpoint | Jan 31, 2031 |
| Implied base price | ~$520/share |
This is a normalized-cycle checkpoint, not me compounding peak economics forever.
| Case | What has to be true | Target range | Annualized return from today |
|---|---|---|---|
| Bear | ~$12.00 normalized EPS · 20x P/E. FY2031 revenue reaches only about $40B, custom/interconnect ramps slower than management expects, operating leverage stalls and diluted shares drift toward roughly 1.0B. The AI thesis still works, just nowhere near the Investor Day slope. | $216–$264 center ~$240 | -4.1% |
| Base | ~$20 normalized EPS · 26x P/E. FY2031 revenue reaches roughly $55B, independent of management's $70B–$90B target. Custom and interconnect both scale, margins improve and buybacks offset enough issuance to keep the denominator manageable, but execution is meaningfully below the Investor Day headline. | $468–$572 center ~$520 | +14.8% |
| Bull | ~$30 normalized EPS · 30x P/E. FY2031 revenue reaches roughly $75B, custom and interconnect both hit substantial scale, margins hold despite mix pressure and repurchases contain most dilution. This is the case where management gets reasonably close to its long-range framework. | $810–$990 center ~$900 | +30.4% |
Returns use the current quote and the time remaining to Jan 31, 2031. The scenario targets stay fixed until the underwriting changes.
Capital allocation
Investor Day finally put per-share economics on the board. The financing bill still exists.
The strategic case for custom silicon, optics, switching and the Celestial/XConn assets looks stronger after Investor Day. The score stays put because the shareholder math still carries a large equity component from Celestial, NVIDIA's convertible preferred, Google's performance-linked warrant and heavy SBC. Google's warrant is unusually aligned with revenue milestones, which makes it less ugly than the headline share count suggests, but the >$30 FY2031 EPS target is still evidence to track rather than a realized return on capital.
Stage-relative, not some objective rating-agency grade. The number is there to force the judgment into the open. See the rubric →
What management is actually doing with the cash
The product strategy still makes sense: custom silicon, electro-optics, switching and scale-up connectivity all reinforce the same AI-infrastructure position. Investor Day gave me a much stronger argument that the spending can create value per share, but it did not erase the financing bill. Celestial used meaningful equity, NVIDIA owns convertible preferred, and Google holds a warrant for up to roughly 59M shares. Most of Google's performance-linked warrant only vests alongside enormous qualifying purchases, so it is not free dilution. SBC remains real. The per-share targets still need to become reported per-share economics before management gets extra credit.
What I’m doing with the next dollar
Marvell is already a meaningful position inside the 0%–5% framework.
The position does not need help.
The Investor Day materially improved the expected value of the holding, which makes the old idea of automatically trimming once the taxable lot becomes long-term much less convincing. Taxes can affect the timing of a decision. They are not a reason to keep using a model that no longer reflects the business.
No add today either.
There is already plenty of AI-infrastructure exposure elsewhere in the portfolio, the stock immediately repriced some of the better story, and a four-year management target is still a four-year management target.
So the current stance is simple: hold, no add.
The next dollars have to compete with the index and every other holding. Investor Day earned Marvell more patience. It did not earn unlimited capital.
What can break this
Investor Day supplied a useful long-range per-share checkpoint, but management’s $70B–$90B revenue target and greater-than-$30 EPS ambition are not guaranteed outcomes. My own scenarios allow for a much slower ramp and for the shareholder dilution the programs could bring.
Here is my FY2031 range:
| Case | FY2031 revenue read | FY2031 EPS | Exit P/E | Implied value |
|---|---|---|---|---|
| Bear | ~$40B | $12 | 20x | ~$240 |
| Base | ~$55B | $20 | 26x | ~$520 |
| Bull | ~$75B | $30 | 30x | ~$900 |
Base assumes Marvell can become a roughly $55B revenue business, earn about $20 per share and still deserve 26x in FY2031. That is an exceptional growth outcome and depends on custom silicon, interconnect, margins and dilution all behaving reasonably well.
From the Oct. 6 close, the Base is around 15% annualized. Keep $20 of EPS but use a 23x multiple, and the return drops closer to 12%. More aggressive operating leverage and a richer exit valuation belong in Bull.
The evidence I’m watching
FY2028 lands around or above $20B, FY2029 custom clears $12B without sacrificing the interconnect ramp, the $37.5B midpoint interconnect framework proves real, diluted shares flatten despite warrants/SBC and reported per-share earnings start making the >$30 FY2031 target look conservative.
Hyperscaler capex slows, a major custom program slips, connectivity share disappoints, gross-margin pressure overwhelms operating leverage, Celestial/XConn do not earn their cost, or the Google/NVIDIA/SBC dilution keeps expanding the denominator faster than earnings can outrun it.