SambaNova launches new AI chip with $350M funding, teams up with Intel to challenge

SambaNova’s new “RDU‑2” processor, unveiled alongside the funding announcement, is built on a custom dataflow architecture that merges tensor‑core‑style matrix multiplication with a programmable scalar engine. According to SiliconANGLE, the chip integrates a unified memory subsystem that eliminates the need for separate high‑bandwidth memory (HBM) stacks, reducing latency and board‑level cost. The design targets inference workloads at the 10‑to‑100‑peta‑FLOP scale while maintaining power consumption under 300 W, a figure the company claims is competitive with Nvidia’s Hopper GPUs on a per‑watt basis. By offloading sparsity handling to dedicated logic blocks, the processor can exploit unstructured model pruning without sacrificing throughput, a capability that Intel’s partnership is expected to accelerate through co‑development of optimized compiler stacks.

The $350 million round was led by Vista Equity Partners, with participation from Cambium Capital and other investors, as reported by SiliconANGLE. Vista’s involvement signals a strategic pivot toward enterprise‑grade AI infrastructure, a sector where Intel has been seeking to regain relevance after its failed acquisition talks with SambaNova, which CNBC confirmed were abandoned earlier this year. Intel will contribute its Xeon Scalable CPUs, interconnect technology, and the OpenVINO software ecosystem to create “end‑to‑end” AI systems that combine SambaNova’s accelerator with Intel’s general‑purpose compute fabric. The partnership is also expected to leverage Intel’s 3‑D XPoint memory and upcoming Ponte Vecchio GPU‑like accelerators to further reduce memory bottlenecks in large‑scale model serving.

From a market perspective, the move positions SambaNova as a more direct challenger to Nvidia’s dominance in the data‑center AI accelerator space. Reuters notes that the combined investment from Vista and Intel underscores confidence that a heterogeneous stack—CPU, accelerator, and software—can erode Nvidia’s share of the $30 billion AI chip market projected for 2025. Analysts cited by Reuters have highlighted the importance of cost‑effective alternatives for enterprises that cannot afford Nvidia’s premium pricing, especially as model sizes continue to swell beyond the 100‑billion‑parameter threshold. By offering a solution that promises comparable performance with a lower total cost of ownership, SambaNova hopes to capture a slice of the growing demand from cloud providers and hyperscale operators.

Technical validation of the RDU‑2 architecture is still in its early stages. SiliconANGLE reports that SambaNova has begun internal benchmarks on transformer‑based workloads, showing up to a 1.8× speedup over previous generation chips while maintaining a 30 % reduction in power draw. However, the company has not yet released third‑party performance data, and the lack of a public SDK limits immediate adoption by developers accustomed to CUDA or ROCm ecosystems. Intel’s involvement may mitigate this gap; the partnership is slated to deliver a unified programming model that abstracts hardware differences, allowing existing AI frameworks to target the new accelerator with minimal code changes. If successful, this could lower the barrier to entry for enterprises wary of vendor lock‑in.

The funding and alliance arrive at a time when the AI hardware landscape is fragmenting. Nvidia continues to roll out its Hopper and Blackwell GPUs, while emerging players such as Graphcore, Tenstorrent, and Cerebras are courting niche workloads. SambaNova’s strategy—pairing a bespoke accelerator with Intel’s broad CPU portfolio and deep software stack—aims to create a vertically integrated offering that can compete on both performance and price. As CNBC notes, the failed acquisition underscores Intel’s willingness to pursue collaborative routes rather than outright purchases, a trend that may shape future consolidation in the AI chip sector. Whether SambaNova can translate its technical promise into market share will depend on the speed of product rollout, ecosystem support, and the ability to demonstrate real‑world cost advantages against Nvidia’s entrenched platform.