AMD's CPU Comeback Story: Wall Street's $720 Target Meets the Push for On-Device AI

Bank of America raised its AMD price target to $720, betting agentic AI will lift server CPU demand beyond graphics accelerators.

For years the semiconductor trade had a single, tidy logic: want exposure to artificial intelligence, buy the companies that make graphics accelerators. That consensus is now cracking, and the fault line runs straight through the humble central processing unit.

AMD closed Friday's session in German trading at EUR 554.30, a level that leaves the stock just 2.2 percent shy of its 52-week high. The bigger number, though, is the one measuring the year: a gain of 201 percent since January. What looks at first glance like another momentum chase, in the view of a growing camp on Wall Street, is actually the opening of a second act in AI adoption — one in which general-purpose processors and local compute become indispensable rather than obsolete.

Bank of America Raises the Bar

The clearest signal came Friday from Bank of America Securities, where analyst Vivek Arya lifted his price target on AMD from $620 to $720 while keeping the shares as his top pick among processor makers. Arya's case rests on a structural shift: the emergence of so-called agentic AI systems, which he expects to drive materially higher demand for conventional server CPUs in the years ahead.

The logic is straightforward enough. When software stops merely answering queries and starts autonomously planning and executing multi-step workflows, the strain on server architecture changes character. Running a large model once is one thing; continuously coordinating many moving parts is another, and it leans heavily on CPU capacity. Arya's target implies a valuation of roughly 30 times estimated earnings per share for calendar 2028 — a rich multiple that nonetheless reflects how far into the future the market is willing to look.

That argument has been building for weeks. Reporting from Bloomberg and Semafor earlier in the week highlighted how early traction for Meta's AI assistant Muse electrified the broader chip sector, reinforcing the idea that intelligent agents need fundamentally different infrastructure than plain chatbots. Where model training once dominated, the operational deployment of agents demands versatile control and orchestration logic.

AMD is arriving at that moment with product in hand. The sixth generation of its EPYC processors, developed under the codename "Venice," is already in production. Leading cloud providers are targeting their first installations this year, while major server manufacturers are readying their platforms.

Intelligence Moves to the Desk

The shift is not confined to hyperscale data centers. On Thursday, AMD and AI specialist Perplexity unveiled a partnership that carries the same trend directly to the workplace. The Perplexity Portable Computer will run on supported Windows machines equipped with Ryzen AI Max processors, letting users handle demanding AI tasks and repetitive workflows locally through platforms such as the Ryzen AI Halo developer system — no round trip to the internet required.

The strategic weight of this is hard to overstate. Enterprises and individuals alike will route compute-intensive work to wherever it is cheapest and safest: the device itself. AMD has reinforced the ambition with additional moves, including a mid-September collaboration with F-Secure via Silo AI focused on adaptive routing of data streams. The company is repositioning from pure chip supplier to full-scale architect of hybrid infrastructure.

Whether the modern office desk will eventually require its own always-on thinking machine is an open question, but the direction of travel is clear. Data security and execution speed give companies ample reason to keep operational computing close at hand rather than delegating every task to distant servers. For chipmakers, that opens a second leg of demand alongside the server room — one that could underpin years of steady utilization.

A Lofty Multiple and a November Test

None of this comes without caveats. After crossing the trillion-dollar valuation threshold, AMD trades at a level that leaves little room for disappointment, and the pace of the climb itself argues for caution. CEO Lisa Su sold portions of her own holdings in mid-September under a pre-arranged trading plan — the kind of signal that understandably gives cautious investors pause.

The structural case, though, still carries the day for many. The market is rewarding a portfolio that stretches from personal computers to high-performance data centers, and the debate over margin sustainability is unlikely to be settled before the company reports third-quarter 2026 results on November 2. Look past the current quarter, and the picture that emerges is of a company no longer playing perpetual catch-up in semiconductors, but helping to shape the next wave of computing architecture. The competition, increasingly, is not about how many accelerator cards ship — it is about how seamlessly the entire processor chain, from cloud to desktop, fits together.