Summary of Key Points
AMD and Microsoft have deepened their partnership, with Microsoft's Azure data centers deploying AMD's first rack-scale AI system, Helios, for use in AI inference, training, and enterprise AI workloads. Helios outperforms NVIDIA's similar products but is more expensive. Its clients include leading companies such as Meta, OpenAI, and Oracle. Currently, AMD holds only 4.5% of the data center GPU market (NVIDIA accounts for 95%), but it plans to turn the tide with Helios, aiming to generate several billion dollars in revenue from AI in data centers by 2027. On that day, both AMD and Microsoft's stock prices rose in the US market.
1. Why Does Microsoft Azure Choose AMD? – The Details of the Partnership
This collaboration is no small matter; it involves directly integrating AMD's Helios system into its own Azure data centers. There has already been cooperation between the two companies: AMD chips are used in Surface computers and Xbox gaming consoles, and Azure has previously utilized AMD's MI300X GPUs. With this upgrade, Azure will be able to offer customers more powerful AI services, such as enabling AI models to answer questions quickly (inference) and train large models (teaching AI). Two new specialized computing instances have been added: one for “proxy AI” (e.g., intelligent customer service) and another for semiconductor design (helping chip companies develop chips more efficiently).
Microsoft CEO Satya Nadella stated that this is about providing customers with more options to build more advanced AI applications. AMD CEO Lisa Su believes this partnership marks a “milestone” in expanding the next generation of AI infrastructure. In short, Microsoft needs AMD's hardware to enhance Azure’s AI capabilities, while AMD needs Microsoft’s platform to expand its market.
2. What is Helios? How Powerful is AMD’s “AI Supercomputer”?
Helios can be described as a “supercomputer cluster dedicated to AI, filled with AMD chips,” representing AMD’s first rack-scale system optimized for AI use. What does it include?
- AI GPUs: AMD Instinct MI455X (the “brain” that handles AI tasks)
- CPUs: EPYC Venice (the “manager” that processes data)
- Network Equipment: Pensando (for fast data transfer between servers)
- Software System: ROCm (the “operating system” that manages these hardware components)
Compared to NVIDIA’s similar products, Helios has two notable features:
- Better Performance: AMD claims that its inference capabilities, memory bandwidth (data transfer speed), and memory capacity (for storing more AI models) are superior to NVIDIA’s rack-scale systems.
- Higher Cost: Market research firm Futurum estimates that a set of Helios will cost $5-5.5 million, which is significantly more expensive than NVIDIA’s second-generation systems ($3.5-4 million).
However, Helios is not yet available; it is still in the “preliminary” stage and will be shipped in the second half of 2026.
3. More Than Just Microsoft! Which Big Players Are Competing for Helios?
In addition to Microsoft, companies like Meta, OpenAI, and Oracle have announced their plans to deploy Helios. These are all leaders in the AI field: Meta is developing its own large models, OpenAI is the parent company of ChatGPT, and Oracle is a leading cloud service provider.
There’s also an impressive statistic: eight of the top ten global AI companies are already using AMD GPUs for their AI tasks. For example, Meta stated in February that it plans to use 6 gigawatts of AMD GPU power (the more power, the stronger the performance), with 1 gigawatt of that coming from Helios. The support of these customers indicates that AMD has gained recognition from industry leaders in the AI hardware sector.
4. Challenging NVIDIA’s Monopoly: Can AMD Compete for the Market Share?
NVIDIA currently dominates the data center GPU market, holding 95% of the share, while AMD only has 4.5%. However, the launch of Helios could change this situation:
- Market research firm Futurum predicts that if AMD can increase its market share to 20-25%, it could generate several billion dollars in revenue.
- The performance advantage of Helios is crucial; although it’s more expensive, customers may choose it if they believe the higher performance is worth the extra cost.
- AMD is rapidly expanding its customer base, with the addition of giants like Meta and OpenAI, which could attract more companies to use its products.
However, NVIDIA will not sit idly by and is likely to release even more powerful products. This “AI hardware battle” has just begun.
5. AMD’s Ambition: Generating Several Billion Dollars from AI by 2027?
AMD has big ambitions for its AI business: CEO Lisa Su aims to make the annual revenue from its data center AI business several billion dollars by 2027, with growth of over 80% each year.
Is this goal achievable? Based on current progress, it seems promising: more leading companies are adopting AMD’s products, Helios’ performance is being recognized, and giants like Microsoft are supporting it. On that day, AMD’s stock rose by 1.58%, and Microsoft’s by 2.15%, indicating positive market sentiment towards this partnership and AMD’s prospects.
Of course, there are challenges to overcome, such as the high cost of Helios and potential increased competition from NVIDIA. Nevertheless, AMD has already established a strong presence in the AI hardware sector.
In One Sentence
AMD is leveraging its partnership with Microsoft and attracting customers like Meta and OpenAI to compete with NVIDIA for the AI hardware market, aiming to generate several billion dollars from AI by 2027. This “counterattack” strategy is certainly worth watching.