⚔️ AMD goes after Nvidia’s crown.
Until now, a simple rule applied: if you were a major cloud provider (Microsoft, Google, Meta) wanting to build the most powerful data center for training artificial intelligence, you had to knock on Nvidia’s door. But that’s now changing.
Společnost AMD (AMD) uzavřela masivní, průlomovou dohodu s Microsoftem, který do své cloudové platformy Azure ve velkém nasadí nový vlajkový systém od AMD nazvaný Helios. Tento krok způsobil na Wall Street pořádný poprach – akcie AMD ihned reagovaly růstem a analytici napříč trhem zvedají cílové ceny.
What exactly is it, why does this monstrous rack cost a fortune, and how much does it threaten Nvidia?
🖥️ What is the AI Rack Helios and why does it cost up to $5.5 million?
Helios is neither an ordinary chip nor a standard server. It’s a so-called rack‑scale solution – a complete, nearly 2.3‑ton supercomputing rack filled with liquid cooling, connected with top‑tier optics, and ready to be plugged straight into a data center.
Analyst firm Futurum Group estimates the cost of a single Helios rack at an astronomical $5.0 to $5.5 million.
Why is it so expensive?
Inside a single Helios rack lies brutal computing power:
👉 72 cutting‑edge Instinct MI455X AI accelerators, delivering an incredible 31 TB of high‑bandwidth HBM4 memory.
👉 18 6th‑generation EPYC server processors (“Venice”) that manage all operations.
👉 Extreme power – one such system consumes 225 to 245 kW of electricity and delivers exaFLOPS‑class computing power. 🛡️ Does Nvidia have an answer? (Comparison with GB200 and Vera Rubin)
Nvidia, of course, isn’t standing on the sidelines. It was the company that started this whole‑rack trend with its GB200 NVL72 system (Blackwell architecture), priced at around $3 million. Moreover, Nvidia is currently rolling out the next‑generation Vera Rubin, with an estimated cost of $3.5 to $4 million per rack.
Why is Microsoft paying about 40% more for AMD’s Helios than for a system from Nvidia? The answer lies in memory capacity and inference (generating AI responses). With the Helios system, AMD offers a massive advantage in throughput and HBM4 memory size. For the largest AI models with hundreds of billions of parameters, this means developers can run more complex workloads on fewer interconnected racks.
⚠️ Why is this a real threat to Nvidia?
Until now, Nvidia has held a virtual monopoly in the full rack‑scale system segment. It sold them with gross margins exceeding 80%.
Loss of pricing dominance: By choosing AMD as a full‑fledged rack supplier, Microsoft has shown other hyperscalers (Amazon, Google, Meta) that a real alternative exists. Nvidia will no longer be able to dictate terms as easily as before.
AMD market share growth: Analysts expect that AMD’s share of the data‑center accelerator market could rise from today’s modest 4.5% to 20–25% thanks to Helios. For AMD, that means the potential for tens of billions of dollars in new revenue. 📈 Conclusion.
The Microsoft‑AMD deal is clear proof that the battle for AI infrastructure is entering its second phase. It’s no longer just about buying individual graphics cards, but about massive, whole‑datacenter architecture.
Nvidia, of course, remains the dominant market leader with a huge software moat (CUDA). Yet AMD’s entry into the top‑tier rack‑scale league shows that the competition isn’t sleeping. For investors, this is great news – diversification in the AI sector is increasing, and AMD is proving it can sell products with huge margins and high added value.
❓ How do you see this clash of the giants? Do you believe AMD can carve out a significant slice of the market pie from Nvidia, or is Nvidia’s lead still unshakeable?
🗣️ Personally, I see it as healthy competition. I don’t think it will threaten Nvidia all that much. Nvidia has a very broad portfolio and is eyeing other sectors where it could be put to use. For now, it remains a cheaper company to me than AMD.
If there are any other eToro investors here and we aren’t following each other, drop me a line! 🙋♂️
