Just In: Marvell Secures Major Order from Google

Just In: Marvell Secures Major Order from Google

According to recent reports, Marvell Technology will help Google develop its much‑needed custom chips, and will grant the tech giant an option to become one of its largest investors through an equity investment of up to $12.2 billion – the latest move by Google to capitalise on the artificial intelligence boom.

The chipmaker's shares surged more than 11% in pre‑market trading, while shares of its larger rival Broadcom – a key custom‑chip partner for Google's parent company Alphabet – fell more than 2%.

Demand for in‑house chips such as Google's Tensor Processing Units (TPUs) has soared as businesses seek cheaper alternatives to Nvidia's graphics processors, as well as technologies better suited for inference – the process of running trained AI models.

The new collaboration covers a broad range of chips and related technologies designed to work in tandem with Google's TPU ecosystem, which underpins the majority of Google's AI infrastructure.

Under the agreement, Google received a warrant to purchase up to 58.97 million Marvell shares at $206.58 per share.

Specifically, in the agreement, Marvell stated that on July 29, 2026, Marvell Technology, Inc. entered into a commercial agreement with Google LLC covering the development of custom semiconductor products for Google. The expanded partnership encompasses a range of custom chip projects that interface with the TPU ecosystem, including AI inference accelerators, storage controllers, network interface controllers, memory interface controllers, and near‑memory computing.

In connection with this collaboration, on August 18, 2026, the company issued a warrant to Google (the “Warrant”), allowing Google to purchase up to 58,970,907 shares of the company's common stock (the “Warrant Shares”) at an exercise price of $206.58 per share.

According to Reuters calculations, if fully exercised, the warrant would be worth approximately $12.18 billion. Based on data from the London Stock Exchange Group (LSEG), a stake of that size would make Google the fifth‑largest shareholder of Marvell. Most of the warrant will only become exercisable after Google achieves agreed‑upon procurement targets before fiscal year 2033, meaning that the scale of Google's potential future holdings in Marvell is tied to the volume of its purchases from the chipmaker.

Just weeks ago, major tech companies reaffirmed their expectations to spend more than $700 billion on AI infrastructure this year – an unprecedented amount, representing a significant increase from last year's $400 billion.

Companies like Google, which are pouring massive investments into artificial intelligence infrastructure, are seeking ways to reduce chip costs – including those of Nvidia Corporation's AI accelerators, widely regarded as the best in the industry. Other hyperscale data centre operators, such as Amazon.com Inc., are also developing their own semiconductors for inference, i.e., running AI models.

Google's TPUs have become a key factor in attracting customers, as they allow Google Cloud clients to use these chips in their own data centres.

Citizens analyst Andrew Boone said earlier this year that Alphabet is on track to generate approximately $3 billion in revenue from TPU‑related infrastructure this year, and $25 billion by 2027.

As the business grows, Google has been seeking more custom chips. In April this year, Google signed an agreement with Broadcom Inc. for Broadcom to develop and supply custom TPUs. Shares of Broadcom fell 2.4% at the open following news that another major chipmaker would collaborate with Google on TPUs.

Google and Broadcom Join Forces

On April 6, Broadcom filed a Form 8‑K with the U.S. Securities and Exchange Commission. In the one‑page regulatory filing, the company confirmed two major developments that are set to reshape the AI infrastructure landscape: a long‑term agreement with Google to jointly develop and supply custom AI chips for future generations of Tensor Processing Units (TPUs), and an expanded collaboration with Anthropic, under which Anthropic will receive approximately 3.5 gigawatts of TPU‑based compute capacity starting in 2027.

For context, Broadcom CEO Hock Tan noted during the March earnings call that Anthropic was expected to consume about 1 gigawatt of compute capacity by 2026. This new commitment triples that figure before the end of the year.

Since 2016, Broadcom has been Google's behind‑the‑scenes chip partner, quietly designing TPUs while Google handled the commercial side. The new agreement formalises and significantly expands the partnership, covering not only future TPU products but also networking and other components for Google's next‑generation AI racks, with the collaboration running through 2031.

One final detail is crucial. In this partnership, Broadcom is not merely a chip designer but has become a full‑stack infrastructure partner, providing interconnects and components to connect these systems at scale. A team led by Mizuho Securities analyst Vijay Rakesh estimates that Broadcom's collaboration with Anthropic alone could generate $21 billion in AI revenue by 2026, rising to $42 billion by 2027. Whether these figures materialise will depend on Anthropic's growth trajectory.

The broader implication lies in how this triangular relationship – with Broadcom handling R&D, Google providing support, and Anthropic consuming the output – will affect the competitive landscape for AI chips. For a long time, custom AI chips designed for specific workloads have been seen as a more cost‑effective alternative to general‑purpose GPUs for large‑scale deployments.

Broadcom has now become the strongest example of this theory being validated in practice. It has signed a multi‑year agreement with OpenAI, struck a deal with a fifth undisclosed XPU customer, and now expanded its collaborations with Google and Anthropic to a position approaching dominance in the infrastructure space.

The custom AI chip space represents the next phase in how AI computing is built and delivered. Anthropic has committed to deploying the vast majority of its new compute capacity within the United States – and views this transaction as an extension of its November 2025 pledge to invest $50 billion in U.S. computing infrastructure.

For hyperscalers and cloud‑dependent enterprises in the Asia‑Pacific region, this U.S.-centric buildout is a noteworthy data point: as the most powerful AI models increasingly run on infrastructure based in a single geographic location, issues of latency and data sovereignty will only become more acute.

It is also worth noting that just days ago, reports emerged suggesting that AMD may also participate in Google's TPU design efforts. All in all, the ASIC market is heating up significantly.