AMD Triples Server TAM to $220 Billion, Points to Database Business Potentially Exceeding $100 Billion
Goldman Sachs Communacopia + Technology Conference, September 11, 2026
Advanced Micro Devices delivered a striking upward revision to its total addressable market at Goldman Sachs' Communacopia and Technology Conference, with Corporate Vice President of Financial Strategy and Investor Relations Matt Ramsay disclosing that the company's server TAM projection for 2030 has moved from $60 billion at its November Analyst Day to $120 billion and now to $220 billion. "We talked about more than 60% growth of the data center franchise, more than 80% of growth of the AI business. And at that time, we thought we were well above where the market was and talking about a $60 billion server TAM, and we've now more than tripled that," Ramsay said, adding that AMD's own growth outlook has been raised alongside the market figures.
The magnitude of the revision matters because Ramsay went further, suggesting that SVP and General Manager of Compute and Enterprise AI Dan McNamara's server unit could ultimately capture "over 50% of that TAM," implying a potential $100 billion-plus standalone server business over time. "I've been following and now part of AMD's server business for a very, very long time, and to talk about building a $100 billion server business is pretty exciting. No pressure, Dan," Ramsay quipped. Earnings power guidance moved in step: the prior "more than $20" per-share target over the strategic window has been updated to "significantly more than $20," though management stopped short of a new specific figure.
Agentic AI Is Rewriting Server Architecture, Not Just Adding GPUs
The most technically substantive portion of the session centered on why McNamara believes agentic AI workloads are structurally different from the prompt-response era that defined ChatGPT-style inference from late 2022 through last year. He described the prior model as a "very linear" pipeline of web servers, application servers, database storage and GPU servers with easily calculable attach ratios. Agentic workloads break that model. "With Agentic, as you all know, it's an entirely continuous flow. It's a completely different compute paradigm. It's 24/7 churning, whether within the sandbox spawning numbers of different agents," McNamara said, explaining that a new class of pure CPU-based compute — control plane operations, API calls, database queries, tool execution — sits alongside and uplifts both the GPU servers and the traditional general-purpose server fleet simultaneously.
This is the underlying logic behind AMD's Venice CPU, which McNamara said is built around three distinct workload "swim lanes": GPU-attached head-node servers requiring high IPC and frequency to keep accelerators fed; agentic sandbox servers requiring high thread count per watt, addressed by Venice's 256-core configuration; and the broad general-purpose server market. Management was candid that precise CPU-to-GPU ratios per gigawatt of AI capacity cannot be reduced to a simple model input. "It depends... it's very highly dependent on what the end customer is trying to accomplish," McNamara said, noting only that today's roughly 1:1 ratio is trending higher as agentic deployments scale.
Server Momentum Already Visible in Enterprise Numbers
Beyond the multi-year framing, management pointed to near-term evidence that the CPU thesis is already playing out. Ramsay flagged that AMD's enterprise server segment grew more than 70% in the second quarter, a pace he called essentially unprecedented for the category: "The industry has not seen those type of growth rates in enterprise server basically ever." McNamara also cited Amazon's decision to run its RDS database service — a first-party workload — on AMD's Turin CPUs rather than in-house silicon, framing it as validation that hyperscalers' custom ARM chips do not fit every use case. "It's really that perf-per-core sweet spot and that optimization point on the V-F curve that we pay close attention to," he said, adding that Venice is AMD's broadest day-zero launch yet across OEMs, ODMs, cloud vendors and ISVs.
On the durability of x86 more broadly, both executives pushed back on the narrative that hyperscaler custom silicon threatens the ecosystem. McNamara argued the debate is not an instruction-set issue at all: "There's no fundamental differences in the ISA between x86 and ARM. It's really about delivering to different optimization points... perf per watt per dollar ultimately." Ramsay added that the economics of scaling to a market this size are about "building the best server parts, period," rather than architecture allegiance, and said customer demand signals for AMD's specific optimization points are increasingly multi-generational in nature.
AI Accelerator Business Still Concentrated, With Anthropic as Anchor
On the GPU side, Ramsay reiterated the company's framework with Anthropic, under which the AI lab can purchase up to 2 gigawatts of MI450 capacity, and said AMD is in the "throes of ramping and launching" Helios and MI455. Management reaffirmed prior guidance of more than doubling the data center business in 2027, while acknowledging the AI accelerator business remains concentrated among top customers for now — a deliberate strategy McNamara compared to the early playbook in servers, which started with cloud and national labs before broadening into enterprise. He expects a similar diversification curve in AI, though on a faster timeline, noting enterprises are increasingly evaluating sub-rack deployments — PCIe card configurations or 8-way UBB servers — as they weigh distributed, hybrid infrastructure against cloud-only approaches.
AMD's Own AI Adoption as a Software-Company Pivot
Away from the numbers, McNamara used the session to argue AMD's internal AI transformation — using its own open-sourced data tooling called Optima, plus heavy use of Anthropic's Claude tools across chip engineering — is accelerating time-to-market for both hardware and software. Ramsay said the company has been benchmarked by CFO Jean Hu against peer semiconductor and technology companies and believes AMD is "on the bleeding edge of AI adoption internally," acknowledging higher token costs are being offset by materially larger productivity gains. McNamara went further, characterizing the company's own evolution as a shift in identity: "We are becoming more of a software company and a systems company today than we were even six months ago," pointing to the transition from selling individual chips to delivering fully integrated, rack-scale systems as the change he expects investors to appreciate most over the next 12 to 24 months.