OCP Global Summit 2025 | AMD | A Fully Open and Collaborative AI Ecosystem
Introduction
At OCP Global Summit 2025, AMD shared its view of AI infrastructure under the theme “A Fully Open and Collaborative AI Ecosystem”. AMD stressed that the AI market is growing at a staggering pace, pushing data centers into a new round of infrastructure upgrades. The core message was clear: only through openness and collaboration can AI's potential truly be unlocked.
Content
The Explosive Growth of the AI Market
AMD noted that the AI market was about $45 billion in 2023 and could exceed $500 billion by 2028, a CAGR of as much as 60%, with inference reaching as high as 80%. Striking as these numbers are, they may still underestimate the total demand AI will create.
Why an Open Ecosystem Matters
AMD argues that every major breakthrough in the IT industry has come from openness: from Linux and TCP/IP to the mobile ecosystem. AI is no exception. AMD is pushing not only open hardware (UCIE, CXL, UALink, UEC) but also open software (ROCm, PyTorch, Triton) — a sharp contrast with NVIDIA's closed CUDA ecosystem.
Hardware and Interconnect Standards
On interconnect, AMD presented a complete strategy:
Scale-out: backing UEC (Ultra Ethernet Consortium) to advance Ethernet's capability for AI workloads.
Scale-up: leading UALink, contributing its own Infinity Fabric as the foundation.
ESON: allowing UALink over Ethernet for cross-tier compatibility.
Helios Rack: based on Open Rack Wide, integrating EPYC CPUs, MI300 GPUs, CXL, UALink and PCIe 6.0 into a modular open architecture.
Software and the ROCm Ecosystem
ROCm has a ten-year history, expanding from HPC to AI. In 2024, MI300X generated $5 billion in revenue. Starting in 2025, AMD will open ROCm further, support more AI frameworks, and plans to bring ROCm to Ryzen + Radeon PC platforms so more developers can get started quickly.
Confidential Computing and the CPU Ecosystem
AMD also emphasized the importance of Confidential Computing, extending security to GPUs and accelerators by combining Caliptra with OCP attestation.
Meanwhile, AMD and Intel have jointly formed the x86 Advisory Group to standardize AI instruction sets and ensure the x86 architecture stays dominant in the AI era.
Conclusion
AMD's message at OCP 2025: Open vs Closed.
On hardware, build an open foundation with UEC, UALink and the Helios Rack.
On software, challenge NVIDIA CUDA's closed model with ROCm and open-source frameworks.
On security and the CPU ecosystem, sustain x86 through confidential computing and collaboration with Intel.
AMD's strategy isn't to seize the lead but to become the “second choice” the industry urgently needs, preventing a single-ecosystem monopoly by NVIDIA and thereby preserving balance and competition in the market.
Further Perspectives
Technical impact
AMD's Helios Rack and UALink strategy shows its ambition in modularity and interconnect standardization. This differs from Broadcom's ESUN and NVIDIA's Spectrum-X, setting up a three-camp showdown.
Supply chain observations
By contributing Infinity Fabric to UALink, AMD is effectively releasing core IP in exchange for ecosystem support. That is a new opportunity for networking chipmakers (Broadcom, Marvell) and optical module makers (InnoLight, Lumentum).
If ROCm adoption succeeds, it could shake CUDA's monopoly and draw more EDA and cloud vendors into collaboration.
Market trends
Hyperscaler customers are looking for options to avoid single-vendor lock-in, and AMD's “open” positioning could make it a strategic procurement choice.
Over the long run, if ROCm + PC platforms can attract developers, AMD has a chance to gradually erode NVIDIA's software advantage and establish a true two-player landscape.
Comments