Why Is Marvell Stock Rising Today? Google’s $12.2 Billion AI Chip Deal Explained

Marvell Technology stock surged around 13% before the US market opened on August 19, 2026, and remained up about 12% in early trading. But the market is not celebrating a simple $12.2 billion investment from Google. It is repricing Marvell’s position inside the AI chip industry, from an important Amazon-focused custom-chip supplier to a strategic partner spanning the Google, Amazon and NVIDIA ecosystems.

Let’s break down what Marvell and Google actually signed and why MRVL stock is rising. We will also test the revenue, dilution, accounting and valuation implications hidden behind the headline.

Why Is Marvell Stock Rising Today?

Marvell disclosed that it had entered into a commercial agreement with Google to develop custom semiconductor products connected to Google’s Tensor Processing Unit, or TPU, ecosystem.

The agreement covers several parts of the AI data-centre architecture:

• AI inference accelerators
• Storage controllers
• Network interface controllers
• Memory interface controllers
• Near-memory computing products

Marvell also issued Google a warrant to acquire up to 58.97 million MRVL shares at an exercise price of $206.58 per share. Multiplying the maximum shares by the exercise price produces the widely reported $12.18 billion figure.

MRVL stock rose to around $243.90 in pre-market trading, up approximately 12.9%, before remaining around 12% higher after the opening bell. Broadcom stock fell more than 3% before the market opened, showing that investors also see Marvell’s expansion as a possible competitive threat to Broadcom’s Google business.

Here is what the companies have actually agreed:

The complete terms are available in Marvell’s SEC filing and warrant agreement.

Is Google Investing $12.2 Billion in Marvell Today?

No. This is the most important clarification.

The $12.18 billion figure is not an upfront investment, a guaranteed payment or the value of confirmed chip orders. It represents what Google would pay if all 58.97 million warrant shares vested and Google exercised them for cash at $206.58 each.

A warrant gives Google the right, but not the obligation, to acquire Marvell shares after specific conditions are satisfied.

Think of it as a procurement-linked rewards programme. As Google purchases qualifying products from Marvell, it unlocks the right to acquire more Marvell shares at a fixed price.

Google can also choose a cashless exercise. In that situation, it would receive fewer net shares instead of paying the full exercise amount. Therefore, Marvell is not guaranteed to receive $12.18 billion in cash even if the commercial relationship reaches its maximum scale.

If fully exercised, Reuters estimates that Google could become Marvell’s fifth-largest investor. But that is an end-state scenario, not Marvell’s ownership structure today.

The Hidden $120 Billion Number Inside the Deal

The most revealing figure is not $12.2 billion. It is $120 billion.

Only 1.36 million warrant shares vest based primarily on time during the first year. The other 57.61 million shares vest in 240 tranches.

Each tranche requires $500 million of qualifying product revenue.

240 tranches × $500 million = $120 billion

Google would, therefore, need to generate $120 billion of qualifying purchases for Marvell during the measurement period for all performance-based shares to vest.

The measurement period runs from August 1, 2026, to January 29, 2033. Spread evenly, full vesting would require average qualifying revenue of roughly:

$120 billion ÷ 6.5 years = $18.5 billion per year

For perspective, Marvell generated total revenue of $8.20 billion in fiscal 2026. Its fiscal second-quarter 2027 guidance of $2.70 billion translates into an annualised revenue level of about $10.8 billion.

In other words, the average Google-linked revenue needed for full vesting would be:

• More than twice Marvell’s entire fiscal 2026 revenue
• Around 70% above Marvell’s annualised Q2 guidance
• Large enough to transform the size of the whole company

That tells us the $120 billion figure should be treated as a maximum ceiling, not a base-case forecast.

However, companies do not usually create a 240-step incentive structure without expecting meaningful commercial activity. The scale of the warrant suggests that Marvell and Google want enough room for multiple products and several generations of AI infrastructure.

What Exactly Will Marvell Develop for Google?

The agreement is much broader than a single chip.

Google’s TPU is the main processor designed for AI workloads. But an accelerator cannot operate alone. It needs memory, storage, networking and controllers to move data through the system.

Imagine Google’s TPU infrastructure as a fast-growing city. The accelerator is the central power plant, but the city still needs roads, traffic controls, warehouses and connections to surrounding cities.

Marvell can provide several of those supporting systems.

The phrase used in Marvell’s filing is important. The programs “attach to the TPU ecosystem.”

That does not automatically mean Marvell has taken control of Google’s main TPU design from Broadcom. It means Marvell could supply multiple chips around the TPU, while also participating in disclosed inference-accelerator programs.

This distinction matters because a single-chip contract can disappear when a product generation changes. A portfolio spread across compute, networking, storage and memory creates more opportunities to remain inside the customer’s architecture.

The “Kestrel” Clue Hidden in the Warrant

The full warrant agreement contains another useful detail that received little attention in early coverage.

It defines a “Kestrel Product Launch” as the date when Google completes the qualification of the final production version of a custom product known as Kestrel.

The filing does not explain whether Kestrel is an accelerator, controller or networking product. However, the agreement states that transfer restrictions on the time-based warrant shares can end after the Kestrel launch.

This suggests that at least one specific product is already important enough to receive its own launch condition.

It is not evidence that commercial volume has started. Chip qualification can still take months, and qualification does not guarantee large orders. But it tells us this agreement is tied to a real product-development roadmap, not only a broad statement of future cooperation.

Why Does Google Need Marvell?

Google’s own financial results explain the urgency.

In Q2 2026:

Alphabet said demand for AI infrastructure continued to exceed available capacity. It also began recognising revenue from TPU systems delivered directly to customer data centres, with most revenue from existing TPU system agreements expected to arrive during 2027.

This creates three reasons for Google to deepen its Marvell relationship.

• First, Google needs more chip-design and supply capacity. Depending on a single partner for every component creates execution risk when AI demand is rising faster than infrastructure can be deployed.
• Second, custom silicon can reduce the cost of processing AI workloads. NVIDIA GPUs remain highly flexible, but specially designed chips can offer better performance per dollar for repeatable, high-volume workloads such as AI inference.
• Third, Google is beginning to supply TPU systems beyond its own data centres. That expands the number of chips, controllers and networking components required around each TPU deployment.

The warrant aligns Marvell with this expansion. If Google’s TPU ecosystem grows, Google earns more rights to acquire Marvell equity. If product purchases remain limited, most of the warrant never vests.

Is Marvell Replacing Broadcom?

The available evidence does not support that conclusion.

Broadcom signed a separate long-term agreement with Google in April 2026 covering future generations of custom AI chips and components for next-generation AI racks through 2031.

The more realistic interpretation is that Google is building a broader supplier network.

Broadcom can remain a major TPU partner while Marvell provides additional accelerators and surrounding infrastructure. Large cloud companies commonly divide work across several suppliers to secure capacity, improve negotiating power and reduce dependency on any one chip designer.

Still, Broadcom’s initial stock decline is not meaningless. Investors appear to be pricing in the possibility that Marvell could capture some incremental custom-silicon opportunities that might otherwise have gone to Broadcom.

The key question is not whether Broadcom loses the entire Google relationship. It is whether Marvell’s share of Google’s expanding AI chip budget becomes large enough to change both companies’ growth rates.

How Could the Google Deal Affect Marvell’s Revenue?

Marvell already has strong AI exposure.

Its Q1 fiscal 2027 revenue increased 28% to $2.42 billion, while data-centre revenue grew 27% to $1.83 billion. Data centres contributed 76% of total revenue. Marvell also guided Q2 revenue to $2.70 billion, representing 35% year-on-year growth at the midpoint.

The Google agreement can affect the company in four ways.

1. A larger custom-chip pipeline: Marvell can now participate in several Google programs instead of depending primarily on existing hyperscale opportunities.
2. Longer product cycles: Custom silicon is usually developed years before volume production. Once qualified, changing suppliers can be expensive and technically risky.
3. More components per AI system: Marvell is targeting compute, storage, networking and memory. This allows revenue to increase with the size and complexity of Google’s AI clusters.
4. Greater hyperscaler exposure: Marvell becomes less dependent on any one custom AI relationship, but more dependent on the capital-spending decisions of a small group of cloud companies.

Marvell’s fiscal 2026 filing showed that its ten largest customers generated 82% of revenue. Its biggest direct customer contributed 14% of annual revenue and 16% in Q1 fiscal 2027. The Google deal could diversify Marvell across more hyperscalers, but an exceptionally large Google ramp would eventually create another concentration risk.

How Much Dilution Could the Google Warrant Create?

Using approximately 875.8 million current Marvell shares, the maximum 58.97 million warrant shares could create roughly 6.3% post-issuance dilution.

But that maximum is only reached alongside $120 billion of qualifying revenue.

The table includes the 1.36 million time-based shares and assumes all eligible warrants are exercised for shares. A cashless exercise would reduce the number of shares issued.

Dilution is real, but it is tied to revenue creation. If Marvell reaches the maximum 6.3% dilution, it would mean the company generated an extraordinary $120 billion from qualifying products.

The more important question is whether the additional revenue produces enough gross profit and cash flow to outweigh the equity cost.

The Accounting Cost Investors Shouldn’t Miss

Customer warrants are not economically free for Marvell.

Under US accounting rules, equity granted to encourage customer purchases is generally treated as a reduction of the transaction price and, therefore, a reduction of reported revenue unless the customer provides a separate, identifiable service.

Marvell already uses this accounting treatment for earlier customer warrants.

Its Q1 filing disclosed:

The 4.2 million-share warrant was connected to Marvell’s multiyear AWS collaboration. Marvell said its fair value would be recognised as a reduction of revenue as qualifying purchases occurred.

The new Google warrant is almost 14 times larger by share count. Its final accounting impact will depend on the grant-date fair value and the number of performance conditions considered probable.

Marvell has not yet quantified that impact.

The new warrant agreement even defines qualifying revenue before the contra-revenue effect created by the warrant. That wording strongly suggests investors should watch for a meaningful difference between gross product activity and reported revenue.

Marvell’s next disclosures should clarify:

• The warrant’s grant-date fair value
• How much of the award is considered probable to vest
• The quarterly reduction to reported revenue
• The impact on gross margin
• Whether the cost is included in Marvell’s existing guidance

This could become the deal’s biggest near-term financial surprise.

What Does the Stock-Market Reaction Really Mean?

MRVL traded near $242.50 at the time of writing, compared with its previous close of roughly $216.

A 12% increase adds approximately $23 billion to Marvell’s market value.

At $242.50, the simple intrinsic value of all 58.97 million warrants would be:

($242.50 − $206.58) × 58.97 million = approximately $2.12 billion

That calculation assumes all warrants had already vested, which they have not. It also excludes the warrants’ time value.

Still, the comparison is useful. The market added roughly $23 billion to Marvell’s valuation, while the full warrant’s immediate theoretical intrinsic value is only around $2.1 billion.

Therefore, the stock movement is not a mechanical reaction to Google’s equity option. Investors are assigning more value to three things:

1. Future Google-linked revenue
2. Validation of Marvell’s custom-chip technology
3. A stronger competitive position against Broadcom

That is the real reason Marvell stock is rising.

Marvell Is Becoming an Infrastructure Bridge Across AI Ecosystems

Marvell’s broader strategic position may be the most durable part of this announcement. The company already has:

• A five-year, multi-generation collaboration with AWS
• A $2 billion NVIDIA investment and NVLink-related partnership
• A newly expanded custom-chip relationship with Google

AWS is developing its own AI accelerators. Google is scaling TPUs. NVIDIA dominates general-purpose AI acceleration. These companies compete, but Marvell can provide custom silicon, optical connectivity, switching and data movement technology across their ecosystems.

This turns Marvell into something more valuable than a single-product chip supplier. It is becoming an infrastructure bridge between competing AI computing architectures.

The opportunity is significant because every AI platform faces the same problem: processors are becoming faster, but data still needs to move between memory, storage, accelerators and separate data centres. Marvell’s portfolio sits directly inside that bottleneck.

Is the Current Marvell Valuation Ignoring the Risks?

The Google agreement improves Marvell’s long-term revenue credibility. It does not remove valuation risk. At around $242.50, Marvell’s equity value is approximately $212 billion. That equals:

These are demanding multiples for a semiconductor company, especially one exposed to customer concentration, changing product cycles and margin pressure.

The Google agreement makes Marvell’s growth story more believable. But at the current valuation, execution must also be exceptional.

A product delay, slower Google purchasing, lower margins or a decline in hyperscaler capital expenditure could produce a sharp valuation reset even if Marvell’s revenue continues growing.

What Should Investors Track Next?

The most useful way to follow this agreement is through a three-part framework.

Marvell is expected to report fiscal second-quarter results on August 27. The company may not provide detailed Google forecasts immediately, but investors should listen for comments on revenue timing, product qualification, warrant accounting and whether the agreement changes fiscal 2027 or fiscal 2028 expectations.

The Bottom Line

Marvell stock is rising because Google has validated the company as a major supplier across several layers of its TPU ecosystem. The agreement potentially places Marvell inside AI compute, storage, networking and memory programs through fiscal 2033.

However, the headline needs to be read correctly.

Google has not invested $12.2 billion today. Marvell has not disclosed a guaranteed $120 billion order. Nearly all warrant shares vest only when Google-linked purchases reach specific milestones, and the full $120 billion threshold is far beyond Marvell’s current annual revenue.

Our view is that the agreement is strategically stronger than the headline but financially less certain.

It is stronger because the scope reaches across several products and can deepen Marvell’s role in Google’s AI architecture for years. It is less certain because the revenue is discretionary, the accounting cost is unknown, and MRVL’s valuation already assumes rapid execution.

The next stage of the story will not be decided by the warrant’s maximum value. It will be decided by how quickly Google’s qualifying purchases ramp, how profitable those products are and whether Marvell can convert a major design relationship into sustained cash flow.