AMD and Supermicro drive open-source AI alternative to hyperscaler lock-in
TensorWave, AMD, Supermicro, and Vast Data are collaborating on open-source, heterogeneous AI infrastructure to provide enterprises with alternatives to legacy hyperscalers. The initiative focuses on building flexible, GPU-agnostic platforms that optimize the integration of compute, storage, and networking for large-scale AI workloads.
Key Takeaways
- TensorWave introduced ScalarLM, a CC0-licensed open-source platform that enables training and inference across both AMD and Nvidia GPUs without code changes.
- AMD's upcoming MI455X GPU scales to 72 interconnected units in a single pod, up from the eight-GPU configuration of the current MI355X.
- Supermicro’s H14 server series uses 5th Gen AMD EPYC processors and is designed to support up to 10 GPUs in a 5U rack height.
- Vast Data’s AI Operating System has been integrated with the AMD EPYC and Supermicro H14 stack to eliminate storage I/O bottlenecks during AI workloads.
Why It Matters
The shift from component-level competition to integrated full-stack platforms marks a critical transition for streaming and enterprise AI. By decoupling software from specific silicon, these collaborations allow operators to move intensive workloads between hardware providers based on cost and watt-efficiency rather than proprietary lock-in. For the streaming industry, this means more competitive pricing for metadata enrichment and real-time content moderation at scale. The emergence of 'neoclouds' focused on dense, interconnected GPU clusters provides a specialized alternative to general-purpose legacy clouds. Watch for whether AMD's 72-GPU coherent domain can match the latency performance of Nvidia’s NVL72 architecture in production environments.
Additional Context
The collaboration arrives as the competitive race for AI data center dominance intensifies. Per Tom's Hardware in July 2026, AMD’s new Instinct MI455X, built on a 2nm process with 432GB of HBM4 memory, directly challenges Nvidia’s 'Rubin' architecture by offering 23.3 TB/s of peak memory bandwidth. This hardware push is supported by the rise of specialized AI cloud providers, or 'neoclouds,' which prioritize high-density compute over the general-purpose services offered by legacy hyperscalers. TensorWave, an AMD-exclusive cloud provider, recently secured a $350 million Series B round to expand this infrastructure, signaling strong investor appetite for hardware diversity. While this initiative promotes an open ecosystem, its participants are also deepening ties with the market leader. Vast Data, despite its role in this AMD-centric stack, announced a fully CUDA-accelerated AI operating system at its VAST Forward 2026 conference, per Tech In Asia in February 2026. This dual-track strategy highlights a growing industry trend: while high-profile partnerships aim to break the 'CUDA moat,' infrastructure players are simultaneously optimizing for Nvidia to capture immediate enterprise demand. Vast Data's $1.17 billion agreement with CoreWeave, reported in November 2025 by The Next Web, underscores the massive scale of current AI storage deployments. The hardware foundation for these efforts is being rapidly refreshed. Supermicro’s H14 lineup, launched in October 2024, incorporates 5th Gen AMD EPYC 'Zen 5' processors capable of up to 192 cores per CPU. According to Supermicro, these systems provide a 2.44x performance leap over earlier generations, enabling a significant reduction in data center footprints. This efficiency is vital as organizations move from experimental AI to production-grade agentic applications, which require the high-speed interconnects and liquid cooling solutions showcased by the TensorWave and AMD partnership.
Read full article at siliconangle.com
Get this in your inbox → Subscribe
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source