Google Partnership Reveals an Approximately $120 Billion TPU Attach-Chip Opportunity for Marvell Through FY33

Overview

The report interprets the product scope, warrant structure, and potential financial impact of Marvell's expanded long-term custom-chip partnership with Google. J.P. Morgan views it as a TPU-adjacent chip opportunity rather than a core TPU order, but believes its scale is sufficient to support upward revisions to Marvell's medium- to long-term revenue and earnings expectations, and reiterates its Overweight rating.

Core views

First, J.P. Morgan emphasizes that the announcement should not be interpreted as Marvell having won Google's core TPU accelerator program. Although Marvell used the term “AI inference accelerator” in its 8-K, the report concludes that the partnership actually concerns XPU attach or TPU-adjacent products—custom chips deployed alongside and supporting the core TPU. The product portfolio includes an SRAM-based AI inference offload engine resembling an LPU architecture, as well as storage controllers, NICs, SmartNICs, DPU-class controllers, CXL memory-interface controllers, near-memory computing chips, and other AI infrastructure silicon products. This means Google is expanding custom-chip development beyond its core TPU rather than replacing its existing core TPU silicon partner, and multiple partners can participate in these adjacent areas. Second, the report derives the partnership's revenue capacity from the warrant vesting terms. Marvell granted Google warrants to purchase up to approximately 59 million shares, of which approximately 1.36 million will vest ratably during the first year; the remaining approximately 57.6 million shares will vest from Marvell's F3Q27 through FY33 based on Google's purchases of Marvell custom products. This portion is divided into 240 equal tranches, with one tranche vesting for every $500 million of related custom-product revenue generated. Multiplying 240 tranches by $500 million implies approximately $120 billion of cumulative custom-product revenue if all tranches vest; spread over approximately 6.25 years, this equates to potential average annual revenue of approximately $19.2 billion. The report also explicitly states that this arrangement is not a formal purchase commitment from Google and that full vesting of the warrants cannot be regarded as guaranteed revenue. Third, the implied scale is significantly above existing expectations. Potential average annual revenue of approximately $19.2 billion not only exceeds Marvell's previously stated target of approximately $11 billion in total CY28 AI ASIC revenue from its prior analyst day, but also surpasses consensus company revenue expectations of approximately $11.5 billion and $16.8 billion for FY27 and FY28, respectively. J.P. Morgan believes that a substantial portion of the opportunity represented by the agreement could be incremental to current market expectations, thereby providing a clearer path to upside in FY27/FY28 revenue forecasts; however, the degree of realization will still depend on Google's actual purchases and the pace of the program ramp. Fourth, after incorporating the partnership opportunity into its model, J.P. Morgan believes Marvell's CY28 earnings power could reach approximately $11.00, meaningfully above the current consensus EPS estimate of $9.52. If the Google program begins to ramp, revenue and earnings forecasts could be revised upward. The partnership also adds a new driver to Marvell's multiyear data center growth thesis and strengthens its position as an important merchant-silicon partner for hyperscale AI infrastructure customers. Finally, the report views the partnership as validation of the broader custom ASIC addressable market. Hyperscale cloud providers are expanding their custom-chip investments from core TPU/XPU compute into adjacent areas such as networking, storage, memory interfaces, and near-memory computing. J.P. Morgan therefore considers the announcement clearly positive for both Marvell and the custom ASIC ecosystem and reiterates its Overweight rating.

Analysis framework

The report first clarifies the product boundaries covered by the announcement, distinguishing the core TPU from custom chips surrounding the TPU. It then uses the number of warrants, vesting tranches, and revenue associated with each tranche to derive the cumulative and average annual opportunity. Next, it compares the implied scale with Marvell's existing AI ASIC target, consensus FY27/FY28 revenue expectations, and the consensus CY28 EPS estimate. Finally, it incorporates the potential revenue into its earnings model and assesses the implications for the company's data center growth and the addressable market for custom ASICs.

Methodology notes

• Event Games and Behavioral FinanceEvent-driven analysis

  Announcement-Driven Event and Expectations-Revision Analysis

  The report treats the Google partnership announcement as the triggering event, first assessing the product characteristics the market may misinterpret and then analyzing whether an actual purchasing ramp could drive upward revisions to FY27/FY28 revenue and CY28 earnings expectations.

• (Method Outside the Vocabulary)

  Deriving the Revenue Opportunity from Warrant Vesting Conditions

  The report multiplies 240 vesting tranches by $500 million of related revenue per tranche to derive a cumulative opportunity of approximately $120 billion, then converts this into potential average annual revenue of approximately $19.2 billion over approximately 6.25 years. This estimate represents the scale accommodated by the terms, not confirmed orders.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Marvell Technology Inc (MRVL.US)

  As a custom-chip partner serving the periphery of Google's TPU ecosystem, Marvell could benefit from increased purchases of networking, storage, memory-interface, near-memory computing, and inference-offload products.

  Strengths

  It has custom-chip capabilities spanning multiple categories of AI infrastructure silicon and is becoming a more important merchant-silicon partner for hyperscale AI infrastructure customers.

  Weaknesses

  The partnership does not represent an order for the core TPU accelerator, and the scale indicated by the terms does not yet constitute confirmed purchases.

  Comparison

  The implied average annual revenue of approximately $19.2 billion exceeds consensus revenue expectations of approximately $11.5 billion for FY27 and $16.8 billion for FY28; potential CY28 earnings power of approximately $11.00 exceeds the consensus EPS estimate of $9.52.

  Risks

  Google has made no formal purchase commitment, and full vesting of the warrants and the approximately $120 billion revenue opportunity may not materialize.

Key data

• Marvell Share Price$237.27Price on August 19, 2026
• Maximum Google WarrantsApproximately 59 million sharesMaximum number of Marvell shares Google may purchase
• First-Year Vesting AmountApproximately 1.36 million sharesVests ratably during the first year
• Purchase-Linked Warrant AmountApproximately 57.6 million sharesVests from F3Q27 through FY33 based on Google's purchases of custom products
• Purchase-Linked Vesting Tranches240 tranchesEach tranche is equal in size
• Revenue Threshold per Tranche$500mOne tranche vests for every $500 million of related custom-product revenue generated
• Implied Cumulative Revenue at Full VestingApproximately $120bn240 tranches multiplied by $500 million per tranche
• Implied Average Annual RevenueApproximately $19.2bnThe annual average of approximately $120 billion over approximately 6.25 years
• Existing AI ASIC Revenue TargetApproximately $11bnMarvell's previously stated target for total CY28 AI ASIC revenue from its prior analyst day
• FY27 Consensus Revenue EstimateApproximately $11.5bnBelow the approximately $19.2 billion in average annual revenue implied by the partnership terms
• FY28 Consensus Revenue EstimateApproximately $16.8bnBelow the approximately $19.2 billion in average annual revenue implied by the partnership terms
• Potential CY28 Earnings PowerApproximately $11.00J.P. Morgan's estimate after incorporating the partnership opportunity into its model
• CY28 Consensus EPS Estimate$9.52Below J.P. Morgan's estimated earnings power of approximately $11.00

Impact & implications

The report believes Google is expanding its custom-silicon investment from the core TPU into adjacent areas such as networking, storage, memory interfaces, and near-memory computing, adding a new multiyear data center growth source for Marvell. If purchases ramp as expected, the incremental revenue represented by the agreement could drive upward revisions to FY27/FY28 revenue and CY28 earnings expectations while validating the expanding addressable market for custom ASICs.

Risks

• The agreement does not constitute a formal purchase commitment from Google, and the approximately $120 billion of cumulative revenue implied by full vesting of the warrants is not guaranteed.
• Revenue and earnings upside depends on Google's actual purchases and the program beginning to ramp; the potential opportunity may not materialize at the maximum level specified by the terms or within the expected timeframe.

What to watch

• Monitor when Google's actual purchases of Marvell custom products begin to ramp.
• Monitor purchase-linked warrant vesting and related custom-product revenue from F3Q27 through FY33.
• Monitor whether FY27/FY28 revenue and CY28 earnings expectations are revised upward after the Google program ramps.