NextFin News - Google has handed Marvell Technology an option to buy up to $12.2 billion of the chipmaker's stock, but the fine print reveals the real transaction: the equity only vests if Marvell delivers a wall of custom AI chip revenue over the next seven years. Marvell disclosed in a regulatory filing on Tuesday that it issued Alphabet's Google a warrant for 58,970,907 shares at $206.58 apiece, a package worth roughly $12.18 billion at the strike price, tied to an expanded commercial agreement to develop custom semiconductors for Google's tensor processing unit ecosystem. Shares surged as much as 13 percent in premarket trading, extending a 2026 rally that has more than doubled the stock.
The structure of the deal matters more than the headline number. Google is not writing Marvell a $12 billion check, and the arrangement is not a cash investment in the conventional sense. It is a revenue-contingent equity grant that makes Google a partner in Marvell's upside while binding Marvell's engineering capacity to Google's AI roadmap through fiscal 2033. The question the market is now pricing is whether Marvell can actually earn it.
The Deal: A Warrant That Pays Only If Marvell Delivers
Under the agreement signed on July 29, 2026, Marvell will develop custom semiconductor products that attach to Google's TPU ecosystem, including AI inference accelerators, storage controllers, network interface controllers, memory interface controllers and near-memory compute solutions. The equity component, issued on August 18, 2026, is a warrant exercisable in whole or in part until August 18, 2033.
The vesting schedule is the mechanism that turns the headline into a performance contract. A small tranche of 1,360,867 shares vests in equal quarterly installments during the first year after execution. The remaining roughly 57.6 million shares vest in 240 equal tranches, with one tranche unlocking for every $500 million in custom product revenue generated from Marvell's fiscal third quarter of 2027 through fiscal 2033. The filing, reported under Items 1.01, 3.02 and 9.01, says the warrant shares were issued in reliance on the Section 4(a)(2) private-placement exemption and carry customary anti-dilution adjustments, limited transfer rights and registration rights for Google.
Do the arithmetic and the ambition becomes visible. Fully vesting the revenue-tied portion of the warrant requires 240 tranches of $500 million each, or $120 billion in Google-related custom product revenue over roughly seven and a half fiscal years. That implies an average annual run rate of about $16 billion. For context, Marvell's full-year fiscal 2027 revenue guidance is approximately $11.5 billion across its entire business, and its custom silicon operation generated roughly $1.5 billion in fiscal 2026 after doubling year over year. The company's own long-term target is more than $10 billion in custom silicon revenue by fiscal 2029. The vesting schedule is not just ambitious; it assumes that Google-related custom revenue alone will exceed Marvell's current total company revenue within a few years.
"We are seeing exceptional AI-related bookings, and as a result, we are significantly raising Marvell's revenue outlook for both fiscal 2027 and fiscal 2028 compared with the guidance we provided last quarter," Chairman and CEO Matt Murphy said in May, when the company lifted its fiscal 2027 outlook to nearly $11.5 billion and its fiscal 2028 outlook to approximately $16.5 billion.
If every warrant share is eventually exercised, Google would become Marvell's fifth-largest investor, according to data compiled by LSEG. That is a meaningful ownership stake for a customer, but it arrives with a crucial asymmetry: Google pays for it only through dilution that occurs alongside revenue, not through cash that Marvell must deploy profitably.
Why Google Wants a Second Source, and Why Marvell Wants Google
The partnership lands at a moment when Google's primary custom-chip relationship is already committed elsewhere. In April, Broadcom disclosed a long-term agreement to develop and supply custom Tensor Processing Units for Google's future generations of TPUs, along with a supply-assurance agreement covering networking and other components for Google's next-generation AI racks through 2031. Broadcom has also broadened its footprint beyond Google, expanding an arrangement with Anthropic that gives the AI developer access to about 3.5 gigawatts of computing capacity built on Google's AI processors beginning in 2027, while Meta has extended its own custom-processor work with Broadcom.
That concentration creates a problem for Google and an opportunity for Marvell. When a single supplier designs the accelerators that power a company's most strategically important workloads, switching costs compound with every generation. Adding Marvell as a second design partner for inference-oriented products gives Google negotiating leverage, supply resilience and a hedge against Broadcom's capacity constraints as its order book fills with commitments from Anthropic, Meta and other hyperscalers.
For Marvell, the relationship is a stamp of credibility in a market where design wins are everything. Counterpoint Research projects Broadcom will hold roughly 60 percent of the custom AI accelerator market by 2027, with Marvell at approximately 25 percent. A deep, equity-linked engagement with Google helps Marvell defend and expand that share while converting engineering effort into multi-year revenue visibility.
There is a third actor in this realignment. In March, Nvidia announced a $2 billion investment in Marvell alongside a partnership built around NVLink Fusion, a platform that lets hyperscalers build semi-custom AI infrastructure compatible with Nvidia's software ecosystem. The arrangement covers silicon photonics collaboration, chip-level integration between Marvell's custom silicon and Nvidia infrastructure, and AI radio-access-network technology. The implication is that Marvell is positioning itself as the neutral partner for hyperscalers who want custom silicon without fully abandoning the Nvidia ecosystem.
The Market's Read: Momentum, Dilution and the Earnings Test
Investors reacted to the filing as a validation of the custom-silicon thesis. Marvell closed the prior session at $216 and traded as high as roughly $244 in premarket activity, a gain of about 13 percent, before settling near $242.50 in early trading, up roughly 12 percent. The intraday market capitalization approached $200 billion. The stock has risen about 142 percent so far in 2026, after falling roughly 23 percent in 2025, and its 52-week range extends from $61.44 to $329.88.
But the warrant also introduces a measurable overhang. The maximum 58,970,907 shares represent roughly 6.8 percent of Marvell's diluted share base of about 870 million shares. That is not trivial for existing holders, yet the structure deliberately defers the dilution: shares vest only as revenue lands, so the denominator grows alongside the numerator. In effect, Google is buying a call option on Marvell's execution, and Marvell's shareholders are writing that option in exchange for a locked-in customer.
The near-term test arrives with the fiscal second-quarter earnings report, expected on August 27, 2026. Marvell has guided revenue to $2.7 billion, plus or minus 5 percent, above the consensus estimate of about $2.61 billion, with non-GAAP diluted earnings per share of $0.88 to $0.98. Management has said it expects revenue growth to accelerate every quarter in fiscal 2027, driven by data center demand, and has called for third- and fourth-quarter sequential growth of at least 10 percent each.
Wall Street's posture toward the stock is already bullish. Among 28 analysts tracked over the prior three months, 23 rate the shares a buy or strong buy, five hold, and none rate them a sell, with an average 12-month price target around $273. Raymond James reiterated a Strong Buy rating and raised its target to $235 from $105 in May, citing continued strength in custom compute and optical interconnect. More recently, Stifel reiterated a Buy rating with a $350 target and UBS reiterated Buy at $300, while TD Cowen maintained a Hold rating with a $225 target.
Cyclical Wave Inside a Structural Shift
The cleanest way to frame the deal is to separate two forces that are often conflated. The AI infrastructure spending cycle is, at the margin, cyclical: hyperscaler capital expenditure ebbs and flows with model-training schedules, inference demand and financing conditions, and history shows semiconductor upcycles eventually meet capacity and digest. A cyclical claim requires evidence of mean reversion, and the evidence exists in the form of prior capex pauses, inventory corrections and the fact that today's orders are being placed against a backdrop of unusually cheap capital that will not last forever.
But the force underneath this specific transaction is structural, and it will not revert on its own. Hyperscalers insourcing their silicon is a regime change in the semiconductor industry's architecture, not an inventory fluctuation. Google, Amazon, Meta and Microsoft are all building proprietary accelerators to escape the economics of buying standardized GPUs at premium prices. Once a company has designed a workload-specific chip, trained its software stack around it and amortized the development cost across millions of units, the cost advantage compounds with each generation. The rules of the industry have changed: the customer now owns the architecture, and the merchant silicon vendor must compete on the customer's terms.
The warrant structure is a cyclical bet placed inside that structural shift. Google is betting that its inference workload will grow fast enough to justify $120 billion of Marvell-sourced custom revenue, and Marvell is betting it can scale engineering, supply chain and manufacturing to collect it. The structural thesis says custom accelerators will keep taking share from general-purpose GPUs. TrendForce projects custom chip sales will grow 45 percent in 2026, compared with 16 percent growth in GPU shipments, and the custom ASIC market is projected to reach $118 billion by 2033.
The Second-Order Effect: What the Warrant Says About Nvidia's Pricing Power
The first-order reading of this deal is straightforward: Marvell gains a marquee customer and Google gains a second source. The second-order effect is what the market has not fully priced. When a hyperscaler of Google's scale is willing to take equity exposure to a chip supplier's upside, it is signaling confidence that custom silicon will displace merchant GPU spend at a magnitude large enough to move a balance sheet.
That signal cuts against Nvidia's pricing power. Nvidia's dominance has rested on the combination of performance and a software moat, but custom accelerators need only be good enough for a specific workload at a specific cost per inference. If Google's economics favor Marvell-designed inference chips over Nvidia GPUs for a meaningful slice of its workload, other hyperscalers will draw the same conclusion, and the merchant GPU market faces a slow but steady share shift rather than a sudden cliff. Broadcom's semiconductor revenue from AI rose 143 percent year over year to $10.8 billion in its fiscal second quarter of 2026, a data point that shows how quickly custom silicon can scale once design wins convert to production.
The cross-asset implication is a widening dispersion within the semiconductor complex. Companies that design to a specific customer's workload and attach to that customer's roadmap, such as Marvell and Broadcom, gain revenue visibility and pricing leverage, while pure-play merchant accelerator vendors face a customer base that is increasingly capable of building its own. That does not mean Nvidia's growth ends; it means the growth-rate premium the market has awarded to merchant GPUs is now contestable.
The Counter-Thesis: Marvell Is the Junior Partner in a Broadcom Game
The strongest argument against the bullish read is also the simplest: Broadcom already holds the primary Google relationship through 2031, and Marvell's role may be confined to lower-value attach products rather than the core accelerator. The warrant's product scope supports that concern; it explicitly covers inference accelerators, storage and network controllers, memory interfaces and near-memory compute, a list that reads like the periphery of Google's TPU ecosystem rather than its center. If the highest-margin workloads stay with Broadcom, Marvell's revenue per wafer and per engineering hour could disappoint, and the 240-tranche vesting schedule could stretch well beyond the market's expectations or fail entirely.
There is also execution risk. Marvell's custom silicon business doubled to roughly $1.5 billion in fiscal 2026, but scaling from $1.5 billion to an implied $16 billion annual run rate from a single customer requires a step-change in engineering headcount, design methodology and supply-chain coordination that the company has not yet demonstrated. And the dilution, though deferred, is real: roughly 59 million shares over a base of about 870 million is a 6.8 percent claim on future earnings that accrues to Google, not to existing holders.
These objections are serious but not fatal to the thesis, because the warrant's structure answers them. Google does not collect the equity unless Marvell delivers the revenue, which means dilution arrives only alongside the growth that justifies it. For Marvell's shareholders, the asymmetry is favorable: the company receives a locked-in customer and a credible signal of Google's commitment without an upfront cash obligation, while the downside if the ramp disappoints is a smaller equity grant, not a stranded investment. The bear case would be confirmed only if Google's custom silicon spend remained concentrated with Broadcom and Marvell's Google-related revenue failed to approach the pace the vesting schedule assumes.
What to Watch Next
Three signals will determine whether this deal is a structural inflection or a well-marketed design win. First, the fiscal second-quarter report on August 27, 2026, must confirm the $2.7 billion revenue guide and the expectation of accelerating data center growth. Second, Marvell's commentary on custom silicon revenue and its fiscal 2029 target of more than $10 billion will show whether the Google ramp is tracking the vesting schedule's implied pace. Third, any disclosure of Google-related custom product revenue in future filings will be the direct read on whether the 240 tranches are attainable.
The falsifying signal is quantifiable: if Marvell's Google-related custom product revenue does not approach a roughly $16 billion annual run rate by fiscal 2030, the structural-share-shift thesis for Marvell specifically should be downgraded, even if the broader custom-silicon trend remains intact.
Across time horizons, the picture differs. In the short term, the stock is a momentum trade riding a 142 percent year-to-date gain, and any earnings miss could trigger a sharp reversal. Over the medium term, the question is whether the Google ramp converts guidance into revenue at the pace the vesting schedule demands. Over the long term, the deal is a vote on whether hyperscaler insourcing of silicon is a durable regime shift, and on whether Marvell can hold roughly a quarter of a custom accelerator market projected to reach $118 billion by 2033.
Google is not buying chips with this deal. It is buying optionality, and it has made Marvell's shareholders earn every share of it.
Explore more exclusive insights at nextfin.ai.
