Microsoft Deepens AMD Bet With Next-Gen Instinct and EPYC

AMD will supply Helios AI racks and next-generation EPYC chips to Microsoft Azure, extending a partnership central to the contest for AI infrastructure against Nvidia.

Key Takeaways

• Microsoft will deploy AMD’s Helios rack-scale system, pairing Instinct MI455X GPUs with sixth-generation EPYC “Venice” CPUs, with shipments beginning in the second half of 2026.
• AMD shares jumped more than 5 percent in early trading on July 20, two days ahead of the company’s Advancing AI 2026 conference in San Francisco.
• The deal builds on Azure’s 2023 debut as an early AMD Instinct customer and casts Helios as AMD’s sharpest rack-scale challenge yet to Nvidia’s Vera Rubin platform.

A Widening Compute Partnership

AMD has spent the better part of two years arguing that it belongs in the conversation about who builds the world’s artificial intelligence infrastructure, not merely who supplies parts to it. On July 20, Microsoft gave that argument its strongest evidence yet. The two companies announced an expanded strategic partnership under which Microsoft will deploy AMD’s Helios rack-scale system, combining Instinct MI455X accelerators with sixth-generation EPYC processors codenamed Venice, to run frontier AI model inference across Azure. AMD said shipments to customers, Microsoft included, begin in the second half of 2026.

“AMD and Microsoft have spent years building high-performance infrastructure together, and today we’re extending that partnership across the full stack of AMD AI solutions on Azure,” AMD chair and chief executive Lisa Su said in the companies’ joint release. Satya Nadella, Microsoft’s chairman and chief executive, framed the deal around customer choice, noting that Azure customers need infrastructure suited to training, inference, data preparation, and reinforcement learning alike, and that enterprise customers will access the new capacity through Azure Foundry Managed Compute.

Neither company disclosed the financial value or power capacity of the commitment, a gap investors will likely probe when AMD next reports earnings. The announcement also arrived with pointed timing, two days ahead of AMD’s Advancing AI 2026 conference in San Francisco, where the company is expected to lay out its data center roadmap into 2027. Securing a marquee hyperscaler endorsement ahead of that stage has become a familiar sequence for AMD, which entered prior showcase events with fresh commitments already banked from Meta, Oracle, and OpenAI.

Inside the Helios Architecture

Helios is AMD’s direct answer to Nvidia’s Vera Rubin NVL72, the rack-scale system due to ship later this year. Each Helios rack links 72 Instinct MI455X GPUs with an aggregate 31.1 terabytes of HBM4 memory, yielding up to 1.4 exaflops of FP8 compute and 2.9 exaflops of FP4 performance. AMD is targeting 260 terabytes per second of bandwidth within the rack, comparable to Vera Rubin, and 43 terabytes per second scaling out over UALink and Ethernet, roughly double Nvidia’s claimed figure, though how that Ethernet-based interconnect performs under real training loads at scale remains largely untested.

The EPYC Venice CPUs inside Helios mark AMD’s first products on TSMC’s 2-nanometer node, built on the new Zen 6 core with as many as 256 cores per socket. Azure is pairing that silicon with three new offerings: HDv2 virtual machines for agentic AI and data pipeline workloads, carrying nearly 500 physical EPYC cores and 32 terabytes of local storage per instance; HXv2 machines for chip design work, built around 176 cores clocked above 5 gigahertz with expanded 3D V-Cache; and ND MI455X v7 instances for production-scale inference. Microsoft is also extending its use of AMD’s Pensando networking silicon into Azure Boost, its cloud networking acceleration layer, a detail that signals the partnership now runs well beyond accelerators into the plumbing that connects them.

The Financial Backdrop

The announcement reflects two companies whose results increasingly hinge on this category of spending. AMD’s Data Center segment generated $5.8 billion in the first quarter of 2026, up 57 percent year over year, and now accounts for more than half of total revenue of $10.3 billion, according to results AMD filed with the Securities and Exchange Commission in May. Su told investors that customer engagement around the MI450 series and Helios was strengthening, with customer forecasts exceeding the company’s initial expectations. AMD has guided second-quarter revenue to roughly $11.2 billion, implying 46 percent annual growth, with results expected in early August.

AMD and Microsoft have spent years building high-performance infrastructure together, and today we’re extending that partnership across the full stack of AMD AI solutions on Azure.

Microsoft’s own numbers explain its appetite for a second major silicon supplier. Azure and other cloud services revenue grew 40 percent in the quarter ended March 31, its fastest pace in several quarters, while capital expenditures and finance leases reached $31.9 billion, up 49 percent year over year. Chief financial officer Amy Hood told analysts that full calendar-year 2026 capital spending would total roughly $190 billion, with fiscal fourth-quarter capex alone expected to top $40 billion as new capacity comes online. Management has repeatedly described Azure as capacity-constrained through the year, a backdrop against which broadening accelerator supply beyond any single vendor carries obvious commercial logic. Microsoft reports fiscal fourth-quarter results on July 29.

How Markets Reacted

Investors treated the news as chiefly an AMD story. Shares opened more than 5 percent higher on Monday and were still up roughly 3.5 percent by mid-morning, trading near $513 against a 52-week range of $149.22 to $584.73. Microsoft shares slipped about 0.5 percent over the same stretch, a muted response consistent with a deal that broadens supply options rather than altering near-term guidance. The move came amid an already volatile stretch for chip stocks, several of which had swung more than 5 percent in either direction over the prior two weeks as sentiment shifted around AI capital spending.

The reaction also sat inside a busy run of analyst repositioning ahead of Advancing AI. KeyBanc raised its AMD price target to $725 with an Overweight rating on July 14, Bank of America lifted its target to $620, and TD Cowen moved to $675, all citing strengthening data center momentum. The stock carries a consensus Buy rating. Microsoft shares, by contrast, have traded well below their 52-week high of $555.45 for most of 2026, a stretch shaped less by any single announcement than by investor scrutiny of whether AI infrastructure spending at this scale will convert into proportionate returns.

The Contest Ahead

Nvidia still commands an estimated 70 to 80 percent of the AI accelerator market, and Vera Rubin, which Nvidia says will offer up to five times the inference performance of Blackwell at a tenth of the cost per token, arrives later this year. AMD’s task now is converting marquee wins into durable share. The company has already committed to supplying Meta with up to 6 gigawatts of Instinct GPUs, the first gigawatt on custom MI450-based silicon, and has separately struck a multiyear compute agreement with OpenAI, giving it visibility into hyperscaler demand that extends well beyond this product cycle.

Daniel Newman of the Futurum Group, who tracks AMD’s data center push closely, has argued the company has a credible path to 20 to 25 percent of the market, a share that would represent hundreds of billions of dollars in revenue over time. AMD has said it expects tens of billions of dollars in annual data center AI revenue starting in 2027, with Helios as the primary driver. For Microsoft, the arrangement fits a broader pattern of hedging its AI infrastructure across AMD, Nvidia, and its own Maia silicon, insurance against the risk that no single supplier can deliver capacity fast enough to meet demand.

Whether the hedge pays off will become clearer soon enough. AMD’s Advancing AI conference opens Wednesday, Microsoft reports fiscal fourth-quarter earnings the following week, and the first Helios racks are due to start humming in production before the year is out.