Chip giant AMD (AMD) has officially launched its strongest challenge yet to Nvidia’s (Nvidia) dominance in AI computing infrastructure. On July 20, U.S. Eastern Time, AMD announced it will begin delivering its first rack-scale AI system, Helios, to customers including Microsoft (Microsoft). This marks the first time in years that Nvidia’s popular Grace Blackwell and Vera Rubin systems face a genuine competitor, escalating the arms race for hyperscale data center computing power to a new level.

Microsoft CEO Satya Nadella confirmed the partnership in a statement: “We are expanding our Azure infrastructure portfolio with AMD Helios, giving customers the performance, scale, and choice they need to build and run the next generation of AI applications.” According to CNBC, the Helios system will be deployed to support Microsoft’s frontier model inference workloads and Azure AI services. Microsoft will also add two compute instances on AMD’s latest “Venice” central processors, serving agentic AI and data pipelines, as well as semiconductor design, respectively.

AMD’s announcement extends beyond landing Microsoft as a heavyweight customer. Multiple sources indicate that AMD’s AI computing footprint is expanding rapidly. According to Wall Street CN, chip industry analysis firm SemiAnalysis concluded that top-tier AI lab Anthropic will purchase AMD chips, based on interpreting code published by AMD’s Senior Director of AI Business on public code-hosting platform GitHub. If confirmed, this signals a new dynamic in the high-end AI training market, which has been predominantly controlled by Nvidia and Google’s TPUs.

A Formidable Compute Coalition Takes Shape: Challenging Nvidia’s Dominance

With Microsoft’s official announcement, AMD’s Helios system has assembled what can only be described as an “all-star” roster of customers. Beyond Microsoft, Meta has announced plans to deploy up to 6 gigawatts of AMD Instinct GPUs, with the first 1-gigawatt deployment based on custom MI450 chips set to launch on Helios racks later this year. OpenAI and Oracle (Oracle) have also made substantial deployment commitments, with Oracle building a Helios supercluster containing 50,000 GPUs. India’s largest IT company, Tata Consultancy Services, has also joined the fold.

AMD CEO Lisa Su previously stated in a CNBC interview that compared to Nvidia’s comparable systems, Helios offers “significant advantages in inference capability, memory bandwidth, and memory capacity.” Forrest Norrod, head of AMD’s data center business, emphasized cost-effectiveness: “We are very focused on delivering the best total cost of ownership, meaning the lowest cost per token, and our customers tell us we are achieving that.”

While AMD declined to comment on specific pricing, market research firm Futurum Group provided a reference range. Based on its estimates, a Helios system costs between $5 million and $5.5 million. By comparison, Nvidia’s second-generation rack-scale system, Vera Rubin, is expected to cost between $3.5 million and $4 million. Physically, the 7,000-pound Helios is also wider and heavier than Vera Rubin.

Currently, Nvidia maintains an overwhelming advantage in the data center GPU market. Futurum Group data shows Nvidia controls over 95% of the market, while AMD holds approximately 4.5%. However, this landscape could be redrawn by Helios. Daniel Newman, CEO and Chief Analyst at Futurum Group, predicts: “I think AMD has a significant opportunity to reach 20% to 25% market share. That translates to hundreds of billions of dollars in revenue.”

Financial Engine Shift: Data Center Becomes AMD’s Growth Core

Helios’s launch comes amid a surge in AMD’s data center business. In the first quarter of 2026, the data center segment already accounted for the majority of AMD’s revenue, with a year-over-year growth rate of 57%, reaching $5.78 billion. AMD told CNBC it plans to achieve tens of billions of dollars in data center AI revenue starting in 2027, with the majority coming from Helios.

This growth trajectory underpins Wall Street’s bullish sentiment. According to 24/7 Wall St., the stock carries a “Buy” rating with a price target of $562.88, representing approximately 13.5% upside from current levels. In a bull-case scenario, if new customers like Anthropic materialize and execution remains smooth, the stock could reach $629.03 within 12 months. AMD has risen 131.49% year-to-date but has recently pulled back from its 52-week high of $584.73.

However, elevated valuations also imply risk. AMD currently trades at a trailing P/E ratio of 186x and a forward P/E of 69x, far exceeding Nvidia’s 41x multiple. Any negative surprises at the “Advancing AI 2026” event on July 22-23 — such as Anthropic partnership details falling short of expectations or MI500 roadmap delays — could trigger a valuation correction. Additionally, export controls remain a sword of Damocles; previous restrictions on MI308 chip sales to China resulted in approximately $440 million in net charges for fiscal year 2025.

From the Brink to the Peak: AMD’s Decade-Long Turnaround

Helios’s origins are rooted in AMD’s more than decade-long comeback. Forrest Norrod recalled to CNBC that when Lisa Su took the helm in 2014, AMD was on the verge of massive layoffs due to a series of product delays and missteps that had severely eroded its market share. The turning point began with the launch of the first-generation EPYC server CPU in 2017. Norrod stated: “We did something very unusual at the time — we published a detailed roadmap for the next three generations of products and fully delivered on those promises.”

It was EPYC’s success in the data center CPU market that gave AMD the capital to stage its return to prominence. Today, Helios integrates four core capabilities developed in-house by AMD: GPU, CPU, networking, and software. Each Helios rack contains 18 compute trays, with each tray housing four Instinct GPUs driven by a single EPYC CPU. Its networking chip technology originates from the 2022 acquisition of Pensando. Additionally, the nearly $50 billion acquisition of Xilinx in 2022 and the 2025 acquisition of server manufacturer ZT Systems both completed the puzzle for Helios’s development.

Nevertheless, significant challenges remain. Counterpoint Research analyst Neil Shah noted that while AMD’s Helios chips have reached “parity” with Nvidia at the hardware level, “the secret sauce lies in software and optimization.” Nvidia, with its widely deployed CUDA software ecosystem, still maintains a deep moat. AMD’s open-source alternative, ROCm, continues to improve but still needs time to shift developer habits.

Regarding AMD’s rise, analyst Newman posed a fundamental question: “The question is, is AMD winning because they are technologically ahead? Or simply because the market is so starved for compute that as long as they can build it, someone will buy it?” As Helios begins shipping at scale, the answer will soon emerge from the hum of data centers.