Marvell Technology saw its stock rise in early trading after reports surfaced that Alphabet's Google is in talks to develop two new chips intended to boost efficiency for AI workloads. By 04:38 ET (08:38 GMT) in premarket trading, Marvell's shares had increased 6.3% as the discussions became public.
According to the report, the collaboration under consideration would produce a memory processing unit designed to operate alongside Google's existing tensor processing unit - or TPU - plus a separate TPU built specifically for inference tasks. The memory-focused chip is the nearer-term project in the talks, with the parties aiming to complete its design as soon as next year and proceed to test production thereafter.
The negotiations are part of a broader effort by Google to expand the role of its TPUs beyond internal use and position them as an alternative to the market's dominant graphics processing units. Google has increased the commercial footprint of its TPU business in recent years, and TPU sales have grown into a more meaningful contributor to Google Cloud revenue as the company seeks to demonstrate returns on its AI investments.
Historically, Google retained most TPUs for its own operations. That changed in 2022, when the cloud division assumed responsibility for the organization that sells TPUs externally, a change that materially increased allocations for customers. As demand for AI compute has risen, Google has scaled both production and sales efforts. Last year the company began offering TPUs directly into customers' on-premise data centers in addition to making them available through Google Cloud.
Earlier this month, Google announced TorchTPU, an initiative to enable native compatibility between its chips and PyTorch, the widely used AI framework. The project is intended to allow developers to move existing PyTorch workloads onto TPUs with minimal code modification - addressing a key friction point for customers whose infrastructure is built around that software and potentially increasing TPU adoption.
The prospective Marvell partnership and the TorchTPU announcement together reflect Google's strategy to broaden TPU accessibility and reduce barriers that slow migration from other hardware solutions. Market reaction to the reported talks was immediate for Marvell's stock, while the longer-term commercial impact of any eventual collaboration will depend on the pace of design, testing, and customer uptake.
Key points
- Marvell stock rose 6.3% in premarket trading after reports of talks with Google to co-develop two AI chips.
- The planned chips include a memory processing unit to complement Google's TPUs and a new TPU focused on inference workloads.
- Sectors affected include semiconductor manufacturing, cloud computing, and data center hardware, as Google pushes to broaden TPU sales and compatibility.
Risks and uncertainties
- There is uncertainty around finalizing the memory chip design and the timing of progression to test production.
- Competition in AI hardware remains intense, with TPUs positioned as an alternative to dominant GPU offerings.
- Broader adoption depends on software compatibility and customer willingness to migrate existing PyTorch-based workloads to TPUs.