From early February to early August, Marvell spent $400 million on buybacks, yet its share count rose 3.5%. Revenue grew more than 30% year over year in the quarter ended August 1. As the AI pie grows, how much reaches each share?
This video looks at:
- Four hands reaching for a slice: Google's warrant covering up to nearly 59 million shares, with vesting driven mainly by its own purchases; NVIDIA's $2 billion in convertible preferred stock; employee stock-based compensation that more than doubled; and more than 26 million shares issued for acquisitions;
- Flexible orders versus firm commitments: customers can cancel or defer purchase orders at short notice, while Marvell has more than $8.5 billion in unconditional purchase commitments to wafer and packaging and testing suppliers;
- Management's rising targets: the fiscal 2028 revenue target has increased to $20 billion, but the verified financial results only run through August 1, 2026;
- Two scorecards: half of executives' cash bonuses for the previous fiscal year were tied to revenue. Growing the company and increasing the value attributable to each share are different measures of success;
- Evidence on the other side: in the same report, second-quarter earnings per share rose from $0.22 to $0.33, while first-half earnings per share fell from $0.43 to $0.38;
- What to watch in the third-quarter report: how much of the additional gross profit remains for each share after R&D, capacity payments, stock-based compensation, and customer warrants.
Produced by MILO Investments, based on eight Marvell research reports from Milo Alpha, with research current through October 26, 2026. Financial data cover the second quarter ended August 1, 2026. The cash figures in this video are diagnostic measures calculated in the research, not company-reported free cash flow. This video does not assess whether the stock is expensive or cheap and offers no trading recommendation.
Research commentary only, not investment advice; the author may hold the stocks mentioned.
0:00From early February to early August, Marvell spent $400 million buying back its own stock. Six months later, its share count hadn't shrunk; it had grown by 3.5%.
0:13This isn't because business is bad. Quite the opposite: in the quarter ending August 1, revenue grew more than 30% year over year, and over 90% of the new revenue came from the data center.
0:27The eight Milo Alpha studies behind this episode chase the same question: as the AI cake keeps growing and more hands reach in to cut it, how much ends up with each share?
0:42The first hand belongs to a customer: Google. In August, Marvell signed a custom-product partnership with Google and issued Google a warrant, a right to buy shares at an agreed price, for up to 58,970,907 shares.
1:01That's a ceiling, not shares already issued; once vested, Google still has to pay the exercise price, or exchange on a net basis. How do they vest? Apart from a small time-based portion, mostly by how much Google itself decides to buy; the contract's cumulative purchase scale of $120 billion is a ruler for unlocking shares, not an order and not revenue.
1:32So the more Google buys, the more Marvell sells, and the more shares Google can get; and for some products in the deal, Google decides whether they can be sold to other customers. Is that a win-win, or a slice taken away? The research's view: both are possible, and it can't be settled yet.
1:57The second hand is NVIDIA: in March it invested $2 billion for preferred stock convertible into roughly 21.8 million common shares, sharing dividends and liquidation on an as-converted basis.
2:14The third hand is employees: stock-based compensation in the second quarter more than doubled year over year.
2:23The fourth hand is acquisitions: in the half year, more than 26 million shares were issued for acquisitions while buybacks retired only 2.5 million, the main reason the share count rose despite the buybacks.
2:39To be fair: last fiscal year, buybacks did cut the share count by 2.2%; this half year it went the other way. And this new equity bought assets, talent and capital; it wasn't free.
2:57There's also an asymmetry: the company says a good share of its sales run on purchase orders that can be cancelled or deferred at short notice; yet Marvell's unconditional purchase commitments to wafer and packaging partners, orders already signed that can't simply be returned, come to more than $8,500 million. Customers can change their minds; this money can't just be pulled back. This isn't borrowed debt, and it isn't due all at once; the research hasn't seen any loss yet; the point is that if demand slows, the company would find it hard to scale back.
3:40Management's targets, meanwhile, climb one step higher at a time. A quick note: Marvell's fiscal year ends around late January, so fiscal 2028 is roughly the year through late January 2028.
3:56At its October 6 investor day, the fiscal 2028 revenue target was raised to $20 billion, two-thirds above this fiscal year's target; and this fiscal year's target itself hasn't been met yet.
4:12Further out, the fiscal 2031 target is $70 billion to $90 billion. But the verified financials only reach the step of August 1 this year. Winning more AI business doesn't mean each share earns faster.
4:31One more detail: half of executives' cash bonus last fiscal year was tied to revenue, and the profit metrics excluded stock-based compensation.
4:42In this June's advisory vote on executive pay, the plan passed, but support fell from over 80% last year to just over 50%. These incentive arrangements deserve questions, but the research accuses no one, and governance hasn't been fully re-examined. To be fair, by value 70% of the CEO's annual equity award is performance-based, meaning the payout depends on how well the metrics are met; one batch of older performance shares actually paid out at just over 80% of target. In the end, a bigger company and a thicker slice per share are two different exams.
5:29So the strongest bear case in the research isn't that AI demand collapses, but that Marvell may become an important supplier while leaving only a limited gain for each share.
5:43But the evidence on the other side isn't weak either. Start with cash: free cash flow is the cash from operations minus spending on equipment; the research then also deducts stock-based compensation and technology license payments. In the second quarter alone, that figure was down nearly half from a year earlier: a number for the bears; across the whole half year, the same measure was up more than 30%: a number for the bulls.
6:13Earnings per share cut both ways too: in the second quarter, they rose from $0.22 to $0.33.
6:21But for the half year combined, they fell from $0.43 to $0.38. The same report gives bears and bulls each their own numbers.
6:34R&D, capacity and acquisitions may cost first and pay back later: of the two acquisitions, the smaller one, XConn, is planned to start contributing revenue in the quarter ending in late October; Celestial, paid for mostly in stock, was originally planned to start contributing revenue only in the second half of fiscal 2028; its contribution is still small, which isn't grounds to call it a failure.
7:03The research concludes: the bear case has evidence, but the hypothesis of compounding cash per share is neither proven nor disproven; nor does the research judge whether the stock is expensive or cheap today, and it gives no trading conclusion.
7:20The next checkpoint is the third-quarter report for the period ending October 31; as of October 26, the company had not confirmed a release date. The company guided revenue of $3.15 billion, plus or minus five percent; at the midpoint of guidance, adjusted gross margin would be 0.9 percentage points below the second quarter; the company said so in advance, so a decline within guidance is no surprise; it should be read together with gross profit dollars.
7:55But don't just check whether it hit the target; look at how much of the new gross profit is left for each share after R&D, capacity payments, stock compensation and customer warrants; and whether the period-end share count went up or down.
8:13The AI wave is real, but it won't flow into the share in your hand on its own. Next time you hear about orders, targets or partnerships, ask: how much of this belongs to each share? The author may hold the stocks mentioned in this video; this is research commentary only, not investment advice. See you next time.
Anthropic, the company behind the AI assistant Claude, filed its IPO prospectus on October 26. In Q2 2026, it earned more than $970 million in operating profit, its first in the six quarters listed, yet booked a net loss of more than $66 billion. Is it actually making money?
Ahead of Tesla's Q3 earnings, a closer look at Q2: revenue rose by a quarter year over year, while operating profit fell by more than half. More vehicle sales did not translate into higher earnings.
Before Alphabet reports Q3, a closer look at Q2: of the more than $9 in earnings per share, over $6 came from gains on equity investments. The money is real, but customers did not pay it.
