Marvell beat the quarter, raised both years, and signed a Google runway worth up to $120 billion, and the tape still took 10% off, because after a 5x run acceleration was already the consensus.
Revenue of $2.74 billion, data center up 46%, a $3.15 billion third-quarter guide implying 15% sequential growth, and both out-years raised again: roughly $12 billion for fiscal 2027 and $18 billion for fiscal 2028. The stock closed 10% lower the next day anyway, because the July 29 Google agreement, up to $120 billion of custom silicon through fiscal 2033, was supposed to make those numbers bigger, and management says the real payoff lands in fiscal 2029. Three months after Jensen Huang called Marvell the next trillion-dollar company, the market is no longer arguing about the destination, only the arrival time.
Q2 FY2027 earnings, decoded
The August 27 print beat every line it guides, raised both fiscal years, and was sold 10% the next morning: the quarter was fine, the clock was the complaint. Figures are non-GAAP unless marked. Click a row for the read.
Who Marvell buys from & sells to
A fabless designer at the center of the custom-AI buildout: TSMC makes everything, the memory makers wrap it in HBM, and the customer list now includes its own referee. Click any node.
Flow weight is a subjective 1-10 read of how much each relationship moves Marvell's P&L, not a revenue rank. Broadcom is deliberately not a node: neither supplier nor customer, it is the incumbent every Marvell custom win is taken from, so it leads the ripple list instead. And notice the geometry: Marvell's own risk factors flag dependence on a handful of customers, so the concentration powering the bull case is the same map the bear case prices.
Bull, base & bear case
Both sides own the same facts: growth accelerating, custom inflecting in the second half, Google signed for the long haul. The bull reads a contracted decade; the bear reads a crowded trade whose referee already got paid. Tap a case.
- Guided, not hoped Q3 is guided to $3.15B, up 15% sequentially, fiscal 2027 was raised to ~$12B and fiscal 2028 to ~$18B, with management flagging a significant custom acceleration from the second half.
- The Google decade The July 29 agreement spans inference accelerators, storage, NICs and near-memory compute in the TPU ecosystem, with warrant tranches vesting per $500M of purchases, up to $120B through fiscal 2033.
- Optics is the moat 800G and 1.6T optics, 51.2T switching and DCI are why data center grew 46% before custom even inflects; the street models roughly 69% and 85% data-center growth over the next two quarters.
- Blessed by the referee Nvidia paid $2B for preferred stock, opened NVLink Fusion to Marvell custom silicon, and its CEO publicly called Marvell the next trillion-dollar company; Melius sees $20 of EPS power by decade-end.
Price-target calculator
Pick a fiscal 2028 EPS and the multiple you would pay for it. The whole Marvell debate is two sliders and a calendar.
Implied value = EPS × P/E multiple
Price target = forward EPS × P/E. The defaults ($6.70 of FY2028 EPS at 31.5x) roughly reproduce today's $210.98. The StockAnalysis survey models $12.04B of revenue and $4.20 of EPS for fiscal 2027, then $18.22B and $6.70 for fiscal 2028. The bull case pairs ~$7.50 with ~50x, the bear pairs ~$5.50 with ~30x; all three cases above live inside these ranges.
What Wall Street says
Forty-four analysts, thirty-nine at buy or better, none at sell, and an average target 35% above the tape: the street stayed constructive straight through a 10% markdown.
Technical picture
A 5x run, a trillion-dollar blessing, a 48% drawdown and a sold-off record quarter, all inside one fiscal year: the whole Marvell argument is drawn on this chart.
This time last September the stock was $64 and the debate was about losing custom sockets. Three catalysts rebuilt it: an 18% jump the session after the early-March Q4 print, Nvidia's $2 billion convertible stake plus NVLink Fusion on March 31, and an April re-rate as the first Google reports surfaced. Then June went vertical. Jensen Huang's 'next trillion-dollar company' line at Computex was worth 32.5% in a day, the biggest gain in the stock's history, and the AWS-Trainium headlines plus S&P 500 inclusion carried it to an intraday $329.88 on June 18. July clawed 48% of it back, peak-to-trough, on hyperscaler capex fears and ASIC-competition nerves; the Google warrant bought back nearly 10% on August 19; the record quarter was then sold for 10.3%. At $211 the tape is up 3.3x in twelve months and still 36% off the high: a market that believes the story and refuses to prepay the timing twice.
Past performance does not guarantee future results. Prices are daily closes from Yahoo Finance.
Final assessment
The quarter was never the question; the calendar was. Same order book, 10% cheaper.
The 36% reset bought the same acceleration at a much smaller multiple; own it sized for beta-2.2 swings and let October 6 and December 1 grade the thesis.
The print made the growth case boring, in the best sense: a record $2.74 billion, data center up 46%, a $3.15 billion Q3 guide, and both out-years raised again to roughly $12 billion and $18 billion. What Friday repriced was the clock. The Google agreement is real and enormous, up to $120 billion of custom revenue through fiscal 2033, but Murphy himself says it contributes 'much more significantly' in fiscal 2029, Morgan Stanley notes it was largely in the prior guide, and at 50x this year's survey EPS the market had already front-run the press release in June.
That is what makes the reset interesting. At $211 the stock trades at 31.5x next fiscal year's $6.70 survey EPS for a business guided to grow 50%, and the acceleration is contractually incentivized: Google's warrant only vests as purchase dollars land. The bear case needs a confirmed socket loss or a capex winter; the base case needs the estimates to simply hold. Buy into the reset, size it for a stock that swung 48% inside a single quarter, and treat the October 6 Investor Day and the December 1 third-quarter print, the first with the custom acceleration inside it, as the real referendum.
- October 6 puts numbers on the decade An Investor Day long-term model that quantifies the Google ramp and the fiscal 2029 inflection would convert warrant math into street estimates.
- Q3 shows the inflection The December 1 print is the first quarter of the guided custom acceleration; revenue at or above the $3.15B midpoint with gross margin holding 57.5%-plus validates the model.
- A second anchor signs Melius' 'prospects with Microsoft' becoming an announced program would give Marvell two hyperscaler decades and re-rate the multiple on the spot.
- A capex wobble July proved this tape trades at beta 2.2 to hyperscaler headlines; any guided-down AI capex hits a 50x current-year multiple twice.
- Trainium confirmation If Amazon's next-generation sockets are confirmed at Alchip, the concentration thesis gets its first proof and the custom story loses its second name.
- Margin mix A Q3 gross margin below the 57.5% low end would say custom growth is diluting the P&L faster than volume can repay it.
- The overhang becomes supply Roughly 80 million warrant and conversion shares sit at strikes of $206.58 and $91.84; every vesting milestone converts good news into potential sell pressure.
Sources · 23
- Marvell IR: Q2 FY2027 results press release, full P&L, end-market split, guidance and balance sheet
- Zacks via Yahoo Finance: Q2 beat detail, non-GAAP opex and operating income, cash and debt, FY27/FY28 raises
- CNBC: Marvell tumbles as the outlook underwhelms despite the Google deal; Goldman on elevated expectations (Aug 28, 2026)
- Motley Fool: Aug 28 close $216.62, down 10.28% on softer FY2028 guidance and Google-deal timing
- BNN Bloomberg: Morgan Stanley and Melius on Google-deal timing; Murphy on the fiscal 2029 contribution
- Yahoo Finance: Barron's read-through on the Coherent/Lumentum sympathy selloff; data-center growth modeled ~69% and ~85% the next two quarters
- SEC 8-K: the Marvell-Google commercial agreement (Jul 29, 2026) and the 58,970,907-share warrant at $206.58 (Aug 18, 2026)
- Reuters: Google's $12.2B stake option; up to $120B of custom chip sales through fiscal 2033 (Aug 19, 2026)
- CNBC: Marvell pops on the Google warrant; TPU-ecosystem product scope; Broadcom incumbency (Aug 19, 2026)
- Marvell IR: Q1 FY2027 results press release; the Q2 guide and the optics/custom growth drivers (May 27, 2026)
- Marvell 8-K: Nvidia's $2.0B Series A convertible preferred, convertible at ~$91.84 into up to 21.78M shares (Mar 31, 2026)
- Nvidia IR: strategic partnership connecting Marvell to NVLink Fusion, plus silicon photonics collaboration (Mar 31, 2026)
- CNBC: Jensen Huang's trillion-dollar forecast sends Marvell up 32.52%, its biggest one-day gain ever (Jun 2, 2026)
- Bloomberg: Marvell soars the most since 2000 after the $1 trillion call at Computex (Jun 2, 2026)
- Globe & Mail / Motley Fool: the Jun 18 intraday record near $330 on the AWS Trainium report; S&P 500 add Jun 22; the Alchip risk
- Yahoo Finance: the July de-risking, down 31% in a month on hyperscaler capex revision fears and ASIC competition; beta 2.2 (Jul 16, 2026)
- StockAnalysis: MRVL forecast survey, 44 analysts, $284.80 avg / $400 high / $143 low; FY27 and FY28 estimates (Aug 31, 2026)
- MarketBeat: target actions around the warrant and the print, Oppenheimer $325 (Aug 28), Needham $300 (Aug 28), RBC $360 (Aug 20), Citi $275 (Aug 21), Jefferies $325 (Aug 20)
- GuruFocus: Morgan Stanley raises its target to $246, stays equal-weight on valuation (Aug 28, 2026)
- Markets Daily: Oppenheimer to $325; the $0.93 Q2 consensus and $1.03 Q3 EPS consensus; MarketBeat-panel Moderate Buy mix (Aug 31, 2026)
- Marvell newsroom: 2nm platform silicon produced on TSMC's process; the longstanding TSMC collaboration (Mar 2025)
- Marvell newsroom: custom HBM compute architecture developed with Micron, Samsung and SK hynix (Dec 2024)
- Yahoo Finance via yfinance: price, one-year daily closes, market cap, analyst ratings distribution (Aug 31, 2026)