This startup is creating a tool to break Nvidia's chip lock-in. Read the pitch deck Spectral Compute used to raise $6 million. (www.businessinsider.com)

London startup Spectral Compute has raised $6 million to build SCALE, a software framework that lets applications written for Nvidia’s CUDA platform run on non‑Nvidia GPUs. The team says SCALE is already compatible with certain AMD chip architectures and plans to extend support to Intel and emerging accelerator vendors. Spectral’s “source‑by‑code” approach translates or adapts original CUDA code so it runs on other hardware without relying on Nvidia’s platform in unauthorized ways; the tool is free for academic and non‑commercial use, while commercial deployments are licensed to cloud and enterprise customers. For the AI/ML community this is significant because CUDA has been a major source of vendor lock‑in as Nvidia dominance grew. Tools that lower the friction of moving CUDA workloads to alternative GPUs could reduce cost and procurement risk, enable multi‑vendor deployments, and make it easier for startups and cloud providers to adopt non‑Nvidia accelerators. Spectral faces technical and legal challenges—its compatibility is currently limited to some AMD architectures and follows a different path than past efforts like ZLUDA—but the funding (seed round led by Costanoa) will accelerate product development, go‑to‑market, and hiring for a company aiming to broaden hardware choice across AI infrastructure.