FULL STORY

AMD Advancing AI: From Teaser to Rack-Scale Launch

AMD used its Advancing AI event to unveil a major Anthropic partnership and its new MI455X and Helios rack-scale systems, escalating the AI hardware race while facing subsequent production challenges.

2026-07-22 ~ 2026-07-25 · 7 episodes · 88 posts

Episode 1 · AMD Teases Advancing AI 2026 Event (2026-07-22, 4 posts)

AMD is teasing its upcoming Advancing AI 2026 event in San Francisco. CEO Lisa Su will deliver a keynote featuring data center and client product announcements, alongside an open AI panel with Hugging Face's CEO.

Episode 2 · AMD and Anthropic Secure Multi-Billion Dollar AI Partnership (2026-07-22, 63 posts)

AMD and Anthropic have finalized a strategic AI infrastructure partnership worth tens of billions of dollars to bolster competition against Nvidia. Under the agreement, Anthropic plans to deploy up to 2 gigawatts (2GW) of AMD's next-generation Instinct MI450 accelerators starting in the first half of 2027, with the first 1GW expected by the end of 2026. Additionally, AMD will invest up to $5 billion in Anthropic as deployment milestones are met. Meanwhile, AMD CEO Lisa Su disclosed major hardware specs and strategic pivots.

Confirmed

**Scale and Investment**: AMD and Anthropic secured an AI server deal worth tens of billions. Anthropic will deploy up to 2GW of MI450 GPUs on AMD's Helios rack-scale solution. AMD will invest up to $5 billion based on milestones.

**Hardware and Architecture**: AMD CEO Lisa Su detailed new hardware specs. The new AI accelerator features 320 billion transistors, utilizing TSMC's 2nm and 3nm processes with 3D chiplet stacking. The Helios rack-scale solution delivers 15% more compute than competitors and offers 50% higher HBM4 capacity and bandwidth. Additionally, the air-cooled MI350P supports up to 260 billion parameter model inference per card.

**Software Ecosystem**: AMD and Anthropic signed an engineering agreement to optimize chip performance using Claude. Anthropic noted that Claude assisted in the bring-up process for the AMD Instinct MI355X + ROCm platform, a task originally expected to be massive but completed by just one engineer.

Why it matters

**Compute Focus Returning to CPU**: AMD asserts that as AI shifts from chatbots to Agentic AI, the compute focus will return to CPUs. Workload ratios are shifting from 1 CPU to 4 GPUs towards 1 CPU to 2 GPUs. Consequently, AMD introduced the 256-core Venice chip for Agent sandboxes and raised its 2030 server CPU TAM expectation to over $200 billion.

**High-end Inference Competition**: According to Ben Bajarin, AMD partnering with Cerebras and Nvidia acquiring Groq indicate intensifying competition in the high-end AI inference token market.

43 more related posts →

Episode 3 · AMD Advancing AI Conference Packed, Showcasing Microsoft Partnership and Ecosystem (2026-07-23, 6 posts)

AMD's Advancing AI conference attracted a massive crowd, highlighting its hardware systems and ROCm software ecosystem alongside a deep collaboration with Microsoft. This marks a significant acceleration in AMD's data center and AI business initiatives.

Confirmed

Long queues formed an hour before the keynote, with dense crowds prompting the organizers to activate live streams and overflow rooms. AMD CEO Lisa Su was present to thank attending customers, partners, and developers. The exhibition covered CPUs, GPUs, networking, and AI infrastructure, with a strong focus on the ROCm software ecosystem. Additionally, several co-branded displays with Microsoft were visible, including "Microsoft Foundry on AMD" and "AMD on Azure" booths and banners.

Why it matters

The enthusiastic reception and the showcased partnership with a top-tier cloud provider indicate that AMD is actively and effectively building its own AI ecosystem. The company teased further releases of product highlights, technical details, and customer cases in the coming days, which is of great strategic significance for consolidating its competitive position in the AI hardware acceleration and infrastructure market.

Episode 4 · AMD MI455 System Faces Production Scaling Challenges (2026-07-24, 2 posts)

AMD's first rack-scale system, the MI455 Helios, faces significant production and scaling complexities due to its new architecture, while the next-gen Vera Rubin ramps up more smoothly.

Episode 5 · AMD Launches MI455X and Helios, Escalating AI Compute to Rack-Scale (2026-07-24, 9 posts)

AMD has officially launched the Instinct MI455X GPU and the Helios rack-scale system, marking a shift in AI hardware competition from single chips to complete rack-level system design. With major breakthroughs in memory capacity, process node, and bandwidth, alongside an accelerating software ecosystem, AMD is now positioned to compete directly with NVIDIA at the rack level.

Confirmed

According to the announced specs, the MI455X is built on a 2nm process and delivers approximately 40 PFLOPS of MXFP4 compute. @LysandreJik pointed out that its single-card memory of 432GB HBM4 exceeds the combined total of five H100 GPUs. The architecture achieves over a 3x increase in scale-up bandwidth via 36 UALoE. At the system level, @SumitGup noted that the Helios rack integrates 72 MI455X GPUs and 12 Broadcom chips, delivering 260 TB/s of aggregate rack bandwidth. On the software front, AMD has shortened ROCm's release cadence to 6 weeks; @AnushElangovan and @ryanshrout view this as AMD's push to make its GPU software stack more user-friendly and optimization more automated to close the gap with CUDA. Furthermore, the AMD Advancing AI 2026 event focused entirely on rack-scale infrastructure, ROCm, and broader ecosystem development. Both @AccBalanced and @ryanshrout believe the MI455X breaks the traditional single-chip competition mindset, giving AMD a true rack-level response to NVIDIA.

Unconfirmed

There is a dispute regarding the performance comparison between Helios and NVIDIA's Vera Rubin. @karlfreund cautioned that AMD's Helios uses a "double-wide rack" configuration, whereas NVIDIA's Vera Rubin uses a "single-wide rack," meaning the two cannot be directly compared on a strict one-to-one basis.

Episode 6 · AMD MI455X Early Tests Reveal 23.3TB/s HBM4 Bandwidth (2026-07-24, 2 posts)

Early tests of AMD's MI455X show it features 432GB of HBM4 with a peak bandwidth of 23.3TB/s. However, its pin speed of 7.5 Gbps lags behind Nvidia's reported 10.7 Gbps.

Episode 7 · AMD Nears HBM4 Supply Deal with Samsung for Helios (2026-07-25, 2 posts)

AMD is reportedly finalizing an HBM4 supply deal with Samsung for its Helios rack-scale AI server system, which integrates 72 Instinct chips and features 432GB of memory per card.