CScale has raised $145 million to fund development of AI networking technology, with Nvidia and Intel among its backers, Reuters reports.

The round lands at a moment when networking has moved from an afterthought to a bottleneck in AI infrastructure. Compute gets most of the capital, but the interconnects linking thousands of accelerators now determine how much of that compute a data center can actually use. That shift is drawing strategic money from the two chipmakers whose GPUs and CPUs sit on either side of those links.

Nvidia's participation is notable for what it implies about its own networking stack. The company already sells InfiniBand and Ethernet switching through its networking segment and treats interconnect bandwidth as a competitive moat for its accelerated-computing platforms. When Nvidia invests in an external networking startup, it signals either a gap it cannot yet fill internally or a hedge against rival interconnect architectures gaining ground in large AI clusters.

Intel's stake points the same direction from the other side. The company is rebuilding its position in AI compute with its GPU roadmap and has positioned itself as the merchant-silicon alternative to Nvidia's vertically integrated stack. Backing CScale gives Intel a stake in networking technology that could pair with its accelerators in clusters where customers want options beyond a single vendor's fabric.

Reuters did not report the round's valuation, the lead investor, or a breakdown between strategic and financial backers. The $145 million figure is confirmed as the total raise.

The funding arrives amid surging demand for high-bandwidth interconnects across AI data centers. As training clusters scale past tens of thousands of accelerators, the share of cluster cost and power consumed by networking keeps rising, and the performance of the fabric increasingly sets the ceiling on how large a single training run can grow. That dynamic has already reshaped the supplier landscape: Nvidia's acquisition of Mellanox for $6.9 billion in 2019, completed as the AI buildout accelerated, now looks like one of the decade's decisive infrastructure deals, and Ethernet vendors are racing to adapt their portfolios to AI traffic patterns.

For CScale, the strategic backing matters as much as the cash. Nvidia and Intel bring not only capital but distribution paths into the largest AI deployments, where networking choices are made alongside accelerator selection rather than after it. A startup selling fabric technology into that market without an anchor relationship with a major silicon vendor faces a steep climb; CScale now has two.

Whether CScale's technology ships into hyperscale AI clusters in volume, and whether it complements or competes with Nvidia's in-house fabric, will define whether this round becomes a footnote or a marker on the industry's networking map.