Google in Talks With Marvell for Custom AI Inference Chips, Diversifying Beyond Broadcom

Google is in talks with Marvell Technology to develop two new chips for running AI models, according to a report by The Information summarised by The Next Web. One is a memory processing unit designed to work alongside Google's existing Tensor Processing Units, and the other is a TPU built specifically for inference, the phase of AI where models serve users rather than learn from data. The Next Web reports that Marvell would act in a design-services role similar to MediaTek's involvement on Google's latest Ironwood TPU, and that the discussions have not yet produced a signed contract.

The talks came days after Broadcom, Google's primary custom chip partner, announced a long-term agreement to design and supply TPUs and networking components through 2031. The Next Web writes that the timing suggests Google is not replacing Broadcom but adding a third design partner to a supply chain that already includes Broadcom for high-performance chip variants, MediaTek for cost-optimised "e" variants at a reported 20 to 30% lower cost, and TSMC for fabrication.

Why inference matters now

Google's seventh-generation TPU, Ironwood, debuted this month. Google-reported performance figures cited by The Next Web put Ironwood at ten times the peak performance of the TPU v5p, scaling to 9,216 liquid-cooled chips in a superpod spanning roughly 10 megawatts and producing a vendor-reported 42.5 FP8 exaflops. The Next Web reports that Google plans to build millions of Ironwood units this year.

The Next Web frames the shift from training to inference as reshaping the chip market, noting that training a frontier model is a one-time event requiring enormous compute for weeks or months, while inference runs continuously and its costs scale with demand. The outlet argues that as AI products reach hundreds of millions of users, inference becomes the dominant expense, and that purpose-built inference silicon offers cost and efficiency advantages over general-purpose GPUs. A similar market reading underpinned Nvidia's Groq 3 LPU launch at GTC, which Nvidia positioned as an inference-optimised design.

The strategy is diversification, not substitution.

The Granite Redux backstory

The Google-Marvell relationship has a longer history than this week's report suggests. The Information reported in 2023 that Google had been working since 2022 on a chip codenamed "Granite Redux" that would use Marvell instead of Broadcom, with Google then expecting to save billions of dollars annually. At the time, Google's spokesperson called Broadcom "an excellent partner" and said the company was "productively engaged with Broadcom and multiple other suppliers for the long term."

The Next Web writes that what changed between 2023 and now is that Google appears to have abandoned the idea of dropping Broadcom entirely, with the through-2031 agreement locking in that relationship. Instead, according to the outlet, Google is building a multi-supplier architecture in which Broadcom, MediaTek, and potentially Marvell each handle different parts of the TPU programme. The outlet compares the approach to how automotive companies manage component suppliers, where no single vendor gets enough leverage to dictate terms.

What Marvell brings

Marvell reports data centre revenue of $6.1 billion in its fiscal year ending February 2026, with total revenue of $8.2 billion, up 42% year over year, according to figures cited by The Next Web. The company says it runs a custom silicon business with a $1.5 billion annual run rate across 18 cloud-provider design wins, building chips for Amazon (Trainium processors), Microsoft (Maia AI accelerator), and Meta (a new data processing unit), in addition to its existing work with Google on the Axion ARM CPU. Meta's broader silicon plans were laid out in its own custom chips and open models roadmap.

The Next Web reports that Nvidia invested $2 billion in Marvell at the end of March, partnering through NVLink Fusion to integrate Marvell's custom chips and networking with Nvidia's interconnect fabric. In December 2025, Marvell acquired Celestial AI for up to $5.5 billion, gaining photonic interconnect technology that CEO Matt Murphy said would deliver "the industry's most complete connectivity platform for AI and cloud customers." According to The Next Web, Murphy is targeting 20% market share in custom AI chips and expects roughly 30% year-over-year revenue growth in fiscal 2027.

The outlet reports that Marvell's stock has rallied approximately 50% year to date, with a 30% gain in April alone following the Nvidia partnership and the Google talks. Barclays analyst Tom O'Malley upgraded the stock to overweight and raised his price target from $105 to $150, per The Next Web.

Broadcom's position

The Next Web writes that the Marvell talks do not appear to have weakened Broadcom's position. The outlet reports that Broadcom commands more than 70% market share in custom AI accelerators, and that its AI revenue hit $8.4 billion in its most recent quarter, up 106% year over year, with guidance of $10.7 billion for the following quarter. According to figures cited in the report, Broadcom is targeting $100 billion in AI chip revenue by 2027.

Broadcom's shares rose more than 6% on the day it announced the Google extension, The Next Web reports. Mizuho analysts estimated the company would record $21 billion in AI revenue attributable to its Google and Anthropic relationships in 2026, rising to $42 billion in 2027, with Anthropic accessing approximately 3.5 gigawatts of next-generation TPU-based compute starting in 2027.

TrendForce projects custom chip sales will increase 45% in 2026, compared with 16% growth in GPU shipments, according to figures cited by The Next Web. Counterpoint Research projects Broadcom will hold roughly 60% of the custom AI accelerator market by 2027, with Marvell at approximately 25%, and the overall market reaching $118 billion by 2033.

A four-partner chip strategy

Google's chip strategy now involves four external partners — Broadcom, MediaTek, Marvell, and TSMC — alongside its own in-house design team, according to The Next Web. The outlet writes that the product line spans training, inference, and general-purpose cloud compute, and argues that every hyperscaler depending on a single chip supplier faces pricing risk, supply risk, and the strategic vulnerability of building a business on another company's silicon. The scale of spend on new AI capacity has also driven interest in faster-to-deploy infrastructure, including truck-deployable AI data centers that compress build timelines from years to months.

The Next Web reports that the inference focus of the Marvell discussions reflects a shift in where the money goes, with Google serving billions of AI-augmented search queries, Gemini conversations, and Cloud AI API calls every day, and that reducing the cost per inference compounds at that volume.