Marvell’s Google AI Chip Deal Is Huge. The Dilution Is the Price of Admission.

The Kapital · US Stock Analysis · 20 August 2026

Marvell just received the kind of contract that can change how a semiconductor company is valued. Google agreed to a custom-AI-chip arrangement that includes a warrant for up to 58.97 million Marvell shares at an exercise price of $206.58 and could be tied to as much as $120 billion of revenue through fiscal 2033 if performance thresholds are met. The market immediately understood the strategic significance. What is less obvious is how much of that headline value belongs to current Marvell shareholders after dilution, execution risk and the economics of custom silicon are accounted for.

I find Marvell one of the more interesting second-order AI investments because it is not trying to beat Nvidia at general-purpose GPU computing. It is positioned where hyperscalers increasingly want to spend: custom accelerators, connectivity, data movement and the silicon infrastructure around enormous AI clusters.

The Google agreement validates that positioning in a way that an investor presentation never could. A hyperscaler does not offer a semiconductor supplier a potentially multiyear economic relationship of this scale because the supplier has a good story. It does so because the technology, execution and roadmap matter to a critical part of its infrastructure.

But I also think the immediate bullish interpretation is too simple. A large customer can create large value while simultaneously gaining enormous bargaining power. The stock warrant makes that trade-off unusually visible.

What the agreement actually says

Reuters reported on August 19 that Marvell granted Google a warrant to buy up to 58.97 million shares at $206.58 each. At full exercise, the purchase price would be roughly $12.2 billion. The number of shares that ultimately vest depends on revenue and performance milestones connected with the custom-chip relationship.

The headline potential is enormous: revenue linked to the arrangement could reach as much as $120 billion through Marvell’s fiscal 2033 if the required thresholds are achieved. That figure should not be treated as committed revenue. It is an upper-bound framework tied to execution and customer demand.

This distinction is essential. Investors often capitalize the maximum value of long-term technology contracts as if it were an order backlog. It is not. The correct way to think about the deal is as a distribution of possible outcomes, ranging from meaningful but smaller volumes to a very large multi-generation custom silicon platform.

Why Google wants custom silicon

AI infrastructure economics are becoming too important for hyperscalers to rely on a single compute architecture. Nvidia remains extraordinarily powerful because its GPUs offer performance, software compatibility and ecosystem depth. But Google, Amazon, Microsoft and Meta all have incentives to develop more workload-specific silicon.

A custom accelerator can reduce inference cost, optimize power consumption and give the cloud provider greater control over its hardware roadmap. At enormous scale, even a modest improvement in cost per token can translate into billions of dollars.

That is where Marvell fits. Its value is not simply designing a chip. The company can combine custom compute silicon with high-speed interconnect, optical connectivity, switching and data-center infrastructure. AI systems are increasingly limited not only by raw compute but by how efficiently data moves between processors, memory and racks.

This is why I view the deal as strategically more important than the single revenue headline. It places Marvell inside the architecture decisions of one of the largest AI infrastructure buyers in the world.

The current business was already accelerating before Google

Marvell reported first-quarter fiscal 2027 revenue of $2.418 billion, up 28% year over year. Non-GAAP gross margin was 58.9%, non-GAAP net income reached $718 million, and operating cash flow was a record $638.8 million. Management guided to roughly $2.7 billion of second-quarter revenue.

For fiscal 2026, revenue reached $8.195 billion, up 42%. That growth means Google is not rescuing a stagnant semiconductor company. It is adding scale to a business that was already being pulled by AI data-center spending.

The quality of that growth matters. Semiconductor companies can grow rapidly at the top line while destroying value if gross margin collapses or working-capital requirements absorb the cash. So far, Marvell’s cash generation suggests the current expansion has real economic substance.

The dilution is large enough to matter

Marvell had roughly 876 million shares in the recent period. A fully vested 58.97 million-share warrant would therefore represent dilution in the neighborhood of 6% to 7%, before considering future changes in the share count.

That sounds painful, and in isolation it is. Existing shareholders would own a smaller percentage of the company. But dilution is not automatically value destruction. If issuing 6% more shares helps create far more than 6% additional intrinsic value, the transaction can be strongly accretive on a per-share basis.

The right question is therefore not “Is there dilution?” The right question is “What return does Marvell receive for that dilution?”

If Google’s participation anchors tens of billions of dollars in high-quality revenue, creates manufacturing scale, improves Marvell’s credibility with other hyperscalers and funds deeper R&D, then the warrant could be an excellent trade. If the relationship produces lower-margin volume and Google captures most of the economics, current shareholders may discover that the largest headline in Marvell’s history was less valuable than it appeared.

Custom silicon has a different margin structure than merchant chips

This is the most important issue for my valuation. A merchant chip company designs a product, sells it to many customers and keeps substantial pricing power when demand is strong. Custom silicon is different. The customer participates deeply in the design process and can negotiate from a position of enormous scale.

That can lower gross margin relative to premium merchant silicon. But it can also reduce demand risk because a successful custom platform may ship for years across multiple generations. Revenue can be more predictable, engineering resources can be reused and the relationship can expand into networking or connectivity products.

I would rather own a 45% gross-margin stream with durable volume than a 70% gross-margin stream that disappears after one product cycle. But investors need to model the right economics. Applying Nvidia-like multiples to custom silicon revenue would be a serious mistake.

Google concentration becomes a new risk

A contract this large inevitably changes customer concentration. The more successful the program becomes, the more bargaining power Google gains over Marvell’s future results. A design delay, architecture change or decision to bring more engineering in-house could create a major earnings shock.

This is a familiar pattern in semiconductor supply chains. Winning a hyperscaler can accelerate growth dramatically, but it can also make quarterly numbers dependent on a handful of procurement decisions.

For that reason, one of the things I will watch most closely is whether Marvell uses Google as a reference win to expand with other customers. The best outcome is not simply a giant Google business. It is Google proving the custom-silicon platform and making Marvell the default partner for several hyperscalers.

The deal also says something about Nvidia

The immediate market reaction included pressure on other AI semiconductor names because the deal reinforced the idea that hyperscalers want alternatives to Nvidia. I think that interpretation is directionally right but often exaggerated.

Custom accelerators do not need to replace Nvidia to create enormous value. The total AI compute market can grow quickly enough for GPUs and ASICs to expand simultaneously. Nvidia may continue dominating frontier training while custom chips gain share in inference or highly specific internal workloads.

That is important for Marvell. Its bull case does not require Nvidia to lose. It requires custom silicon to become a larger percentage of a much larger AI infrastructure market.

I explored the broader problem of paying for AI growth before it becomes cash flow in The AI Trade Cracks. Marvell is becoming a particularly useful test because the revenue opportunity is unusually concrete.

My valuation problem: the stock has rerated faster than the earnings base

After the Google announcement, Marvell traded around the mid-$230s. With roughly 876 million shares, that implies an equity value above $200 billion. That is a very large valuation against fiscal 2026 revenue of just over $8 billion.

The market is clearly capitalizing future growth, not current earnings. That can be rational. If Marvell compounds revenue at 25% or 30% for years while converting a healthy portion into free cash flow, today’s revenue multiple will compress quickly.

But the required execution is formidable. A $200 billion-plus valuation already assumes that custom silicon becomes a core infrastructure category, that Marvell remains strategically relevant through multiple architecture cycles and that profitability scales with revenue.

My three valuation worlds

In my bear case, the Google program ramps more slowly than the headline suggests, gross margin declines as custom silicon becomes a larger mix, and customer concentration limits the multiple. I would value the equity in a broad $120–150 billion range.

In my base case, revenue compounds strongly, Google becomes a durable multi-generation customer, Marvell wins additional hyperscaler programs and free cash flow rises toward the high-single-digit billions over time. I can justify roughly $180–220 billion.

In the bull case, Marvell becomes one of the two or three essential custom AI silicon partners globally, captures a large share of the $120 billion potential, and networking plus optical products expand alongside compute. Then $250–300 billion or more becomes plausible.

At a market value around or above $200 billion, the stock sits near the upper half of my base framework. That means I like the business development more than I like the margin of safety.

The next earnings report matters more than usual

Marvell is scheduled to report second-quarter fiscal 2027 results on August 27. Normally one quarter would not change a long-term thesis. This time, investors will be looking for signs that the data-center ramp is translating into the mix, margins and cash generation needed to support the Google narrative.

I will watch revenue against the roughly $2.7 billion guide, gross margin, operating cash flow and any commentary around custom silicon program timing. I will also watch what management does not say. A deal can be economically transformative while customer confidentiality prevents detailed disclosure, so investors may have to infer progress from aggregate data-center growth.

What would make me buy more aggressively

I would become more constructive if three things happen together. First, revenue from custom compute accelerates without a severe gross-margin penalty. Second, operating cash flow scales roughly with earnings rather than being consumed by working capital. Third, Marvell announces meaningful design wins outside Google that reduce concentration risk.

At that point, the company would deserve a platform premium rather than a simple semiconductor-cycle multiple.

What would make me step away

I would worry if management starts emphasizing maximum contract values while underlying cash flow stalls. I would also worry if gross margin falls faster than expected, if the share count continues expanding materially beyond the Google warrant, or if Google becomes so dominant that Marvell’s negotiating position weakens.

The biggest hidden risk is that investors confuse revenue scale with economic value. One hundred billion dollars of cumulative revenue can create less shareholder value than thirty billion dollars of higher-margin, lower-capital-intensity revenue. The quality of the dollars matters.

My conclusion

The Google agreement is one of the strongest strategic validations Marvell could have received. It confirms that custom AI silicon is moving from an interesting niche to a central part of hyperscaler architecture. It also gives Marvell the chance to grow far faster than its historical revenue base would suggest.

But the warrant makes the transaction unusually honest. Google is not merely buying chips. It is negotiating for a share of the value it helps create. Current shareholders should therefore celebrate the demand signal while carefully measuring the per-share economics.

I would not chase the shares purely because the headline says “up to $120 billion.” I would wait for evidence that revenue, margin and cash flow are scaling together. If they are, the current valuation can grow into itself. If they are not, the market may discover that the most exciting number in the deal was also the least useful one.

Sources and data date

Data date: 20 August 2026. Market values are approximate and change continuously. Scenario values are my own framework.

• Reuters: Marvell grants Google stock warrant
• Marvell Q1 FY2027 results
• Marvell FY2026 results

This article reflects my personal analysis and is not investment advice.