Google is diversifying its AI chip supply chain by entering talks with Marvell Technology to develop two new custom inference chips, according to a report from The Information cited by Reuters. This move adds a third design partner to Google’s AI chip ecosystem, complementing its existing relationships with Broadcom for Tensor Processing Units.
Why This Matters
The partnership comes at a critical time. With AI inference demand projected to grow 45% in 2026, Google is strategically building redundancy in its semiconductor supply chain. Custom ASICs (Application-Specific Integrated Circuits) offer significant advantages over general-purpose GPUs in terms of cost efficiency and power consumption for specific AI workloads.
Marvell, known for its expertise in custom chip design and storage infrastructure, has been expanding its AI accelerator business. The company has established partnerships with major cloud providers and developed a custom HBM (High Bandwidth Memory) compute architecture that optimizes AI accelerator performance.
Industry Context
This isn’t Google’s first move toward chip diversification. In March 2026, Meta announced a landmark $100 billion partnership with AMD to develop custom AI chips, signaling a broader industry trend toward vertical integration in AI hardware.
The AI chip market is projected to reach $1.3 trillion by 2030, according to Bank of America forecasts. Both Google and Meta’s strategies reflect a recognition that controlling silicon destiny is becoming a competitive necessity—not just for cost savings, but for guaranteeing supply chain resilience in an era of export controls and shortages.
What Comes Next
If the talks materialize into production agreements, these new Marvell-designed chips would likely target specific inference workloads in Google’s data centers, potentially reducing reliance on third-party GPU suppliers. With inference costs still a major bottleneck for AI deployment, custom silicon could provide meaningful efficiency gains.
The broader implication: the AI hardware landscape is fragmenting into vertically integrated giants, as major players seek to control their own destinies from model architecture to silicon.
Sources: Reuters, The Next Web