Intel said to in $1.6b talks to buy US AI chip startup SambaNova

Intel is in advanced talks to acquire SambaNova Systems, an AI chip startup based in Palo Alto, for about US$1.6 billion including debt, according to sources familiar with the matter.

The deal could be finalized as soon as next month, though the terms and timing may still change.

SambaNova, founded in 2017 by Stanford professors, designs custom AI chips and was valued at US$5 billion in a 2021 funding round.

Intel CEO Lip-Bu Tan is also chairman of SambaNova, and his venture firm Walden International was an early investor.

SambaNova has also signed term sheets with other potential investors, and could pursue alternatives to an acquisition.

Intel and SambaNova declined to comment on the talks.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

SambaNova’s 68% valuation drop signals pressure as Intel rolls out Gaudi accelerators

A proposed US$1.6 billion price, debt included, is 68% below SambaNova’s 2021 US$5 billion mark. The company can run full, unquantized 16-bit (higher-precision) models on one system 1.
Its three-tier memory design and Reconfigurable Dataflow Unit (a chip design that dynamically routes data through the processor rather than relying on fixed cores) suit large mixture-of-experts models, neural networks that selectively route tasks to specialized sub-models 1. Reports in October 2025 said the company explored a sale after it struggled to raise new funding 2 in a market where Nvidia and AMD plus cloud application-specific integrated circuits may take 80% to 95% share 1.
Gaudi 3 rollout has moved slowly, with channel availability defined as broad access through distributors and resellers planned for 2025 3.

Service firms can win work in Intel’s growing AI lineup with Gaudi and SambaNova migration

If Intel brings SambaNova enterprise deployments onto Gaudi AI accelerators, demand for migration and tuning work could follow 4 3.
Managed service providers can bundle Intel Trust Domain Extensions (TDX), a confidential computing technology that isolates workloads at the hardware level, with performance tuning 4. Databricks, a data and AI platform provider, found Gaudi 2 delivered a 1.33x to 1.69x training performance per dollar edge versus select NVIDIA options 5.