Recently, Google and Marvell have reached a chip customization cooperation agreement. This agreement not only covers a wide range of AI chip customization services, but also achieves an unprecedented deep integration of commercial interests between the two parties through a warrant worth up to 12.2 billion US dollars.
The intensification of competition in AI infrastructure is driving tech giants and chip manufacturers to explore closer cooperation models. Recently, Google and Marvell have reached a chip customization cooperation agreement. This agreement not only covers a wide range of AI chip customization services, but also achieves an unprecedented deep integration of commercial interests between the two parties through a warrant worth up to 12.2 billion US dollars.
The "dual lock" of capital and business
According to the disclosure, Marvell issued to Google a warrant allowing it to purchase up to approximately 58.97 million ordinary shares at a strike price of $206.58 per share (approximately RMB 1,390.3), with a potential total value of approximately $12.2 billion. If fully exercised, Google is expected to become the fifth-largest shareholder of Marvell.
However, this is not a simple financial investment or a cash order. The vesting of the warrant is designed as a highly customized "performance bet" mechanism. Apart from a small number of shares being unlocked on a quarterly basis at the beginning of the agreement, the vesting of the vast majority of shares will be strictly linked to the actual purchasing scale in the future: from Marvell's fiscal year 2027 third quarter to fiscal year 2033, whenever Marvell confirms $500 million in revenue from Google's custom chip business, Google can unlock a batch of shares (a total of 240 batches).
In simple terms, the higher the purchasing scale of Google, the higher the potential equity incentive.
This means that the $100 billion equity commitment will only be fully fulfilled when Marvell has cumulatively recognized revenue from Google's custom chip business reaching the potential $120 billion. This mechanism ingeniously ties Google's purchasing intention with Marvell's R&D delivery capability, ensuring long-term strategic synergy between the two parties over the next seven years.
"Side-Step Breakthrough in the TPU Ecosystem"
At the technical cooperation level, Marvell targeted the core peripheral links of the Google TPU (Tensor Processing Unit) ecosystem. The covered product lines include AI inference accelerators, storage controllers, memory interface controllers, network interface controllers, and near-memory computing chips.
This layout is highly strategic. In the operation of large AI models, in addition to the core computing chips, the storage, transportation of data, and network interconnection are also bottlenecks that determine the overall efficiency of data centers. Marvell's profound accumulation in these "supporting chips" fields can effectively help Google increase the data throughput rate of the TPU cluster and reduce energy consumption.
It is worth noting that Marvell did not directly touch the core computing chip design of TPU, but chose to complete the ecosystem around the core computing power. This "side-step breakthrough" not only demonstrates Marvell's technical strength but also showcases its flexibility in supply chain cooperation.
Supply chain diversification and industry pattern reshaping
After the announcement, Marvell's stock price soared significantly, rising by up to 13.54%. By the end of the trading day, the stock price was $237.27, up by 9.85%, and the total market value reached $207.8 billion. Meanwhile, the stock price of its long-term competitor Broadcom dropped significantly. Currently, the company's stock price is $362.48, down by 4.61%, and the total market value is approximately $1.72 trillion.
For a long time, Broadcom has been a core partner of Google in the field of custom chips. Google's high-profile introduction of Marvell and the provision of equity incentives have been widely interpreted by the market as a key measure taken by large cloud providers to actively promote supply chain diversification in the context of the surging demand for AI computing power.
Analysis indicates that Broadcom previously held a majority share in Google's TPU orders. However, after Marvell's entry, this pattern is likely to evolve into a more balanced "duopoly" situation, with Broadcom's market share shifting from "dominance" to "relative leadership".
Despite the risk of a diluted share, this does not mean that Broadcom's business with Google will experience a cliff-like decline. Because in the core field of custom AI chip design, Broadcom is still regarded by the industry as "best-of-breed".
The current macro background is that global AI computing power demand is experiencing explosive growth, and the overall market is in a state of "shortage of supply". Google's introduction of Marvell is more to meet its own increasingly large computing capacity needs rather than simply to "exclude" Broadcom. As industry experts say, this reflects the overall expansion of the TPU sales pie. Broadcom's fundamentals remain extremely strong, its AI semiconductor revenue maintains high growth, and the management has reaffirmed the ambitious goal of exceeding $100 billion in AI revenue in the fiscal year 2027. Therefore, even if the share ratio decreases, Broadcom's absolute revenue scale in Google's business is still expected to maintain growth.
For Google, distributing the supply rights of core chips to multiple suppliers can not only effectively reduce the risk of dependence on a single manufacturer but also gain stronger bargaining power in the fierce competition for production capacity. For Marvell, obtaining the capital endorsement of a tech giant and the expectation of long-term orders is undoubtedly an important milestone in establishing its leading position in the AI custom chip field.
Overall, this "marriage" between Google and Marvell goes beyond the traditional buying and selling relationship, marking that the upstream and downstream of the AI industry chain is moving from simple supply-demand cooperation to a new stage of capital and business integration. In the era of computing power being power, who can build a more flexible and efficient customized chip supply chain will have the true initiative in the next round of the artificial intelligence wave.
