AMD's AI Earnings Signal a Shift: Inference Is Where the Real Money Will Be Made

For two years, the AI investment thesis has been built on one idea: companies need enormous computing power to train their models, and the companies supplying that power — chiefly Nvidia — stand to reap the rewards. Advanced Micro Devices' blowout earnings report on Tuesday suggests that thesis needs an update.

AMD reported first-quarter revenue of $10.25 billion, up 38% year-over-year, with adjusted earnings per share of $1.37 beating analyst estimates near $1.29. But the headline number was buried in the data center segment: revenue surged 57% to $5.8 billion, driven not by training workloads but by inference — the everyday computing that happens every time an AI model generates a response, processes a query, or makes a decision.

CEO Lisa Su highlighted the shift directly, saying demand for "inferencing and agentic AI" is accelerating. The distinction matters. Training is a one-time capital expenditure: you spend heavily to build a model, then you're done. Inference is recurring revenue in hardware form: every AI interaction requires compute power, every single day, forever. An AI model that serves a billion queries a day needs data center infrastructure running around the clock.

AMD also raised its long-term outlook for the server CPU market dramatically. The company now expects the market to grow more than 35% annually through 2030, versus its previous expectation of roughly 18% annual growth. In practical terms, AMD believes the server CPU opportunity could exceed $120 billion by 2030. That is not incremental optimism — it is a structural revision of the market's trajectory.

The inference thesis has implications that extend well beyond AMD. If the AI industry is truly shifting from a training-heavy buildout to an inference-heavy operational phase, the economics of the entire sector change. Training favors the company with the fastest, most specialized chips — which has been Nvidia's competitive moat. Inference favors the company that can deliver the most cost-effective compute at scale, which plays to AMD's strengths in server CPUs and custom accelerators.

There are early signs that the shift is already happening. Mercury Research data shows AMD's server CPU market share climbed to nearly 29% by the end of 2025 — a number that represents real scale, not scrappy-challenger territory. Multi-gigawatt AI infrastructure projects from Meta and partnerships tied to OpenAI suggest that spending on inference capacity will continue to accelerate as more AI applications reach production.

The competitive dynamics are also shifting. Nvidia remains the dominant player in AI training hardware, but its pricing power faces pressure as customers seek alternatives for inference workloads where absolute performance matters less than cost-efficiency. AMD's growing market share in server CPUs gives it a foothold in the data centers where inference happens, creating opportunities to bundle CPU and accelerator solutions that undercut Nvidia's pricing.

Intel, meanwhile, continues to struggle with execution. Its server CPU market share has been declining steadily, and its AI accelerator efforts have yet to gain meaningful traction. The inference shift could either represent a second chance — if Intel can deliver competitive products for cost-sensitive workloads — or an existential threat, if the market moves faster than Intel's engineering pipeline.

For investors, the key question is valuation. AMD trades around 104 times trailing earnings and roughly 16 times sales, numbers that leave little room for execution mistakes. If enterprise AI spending slows, or if Nvidia widens its lead in a critical segment, AMD shares could see sharp volatility. Semiconductor cycles rarely move in straight lines. But the inference thesis provides a structural growth narrative that extends well beyond the current quarter.

Wall Street remains firmly bullish: 31 of 45 analysts rate AMD as a Strong Buy, with a mean price target of $309.32. The wide range between the low target ($240) and high target ($450) reflects both the opportunity and the uncertainty surrounding AI demand.

What This Means For You: If you're invested in AI stocks, understand that the trade is shifting beneath your feet. The first wave — training hardware — favored Nvidia almost exclusively. The second wave — inference infrastructure — opens the door to AMD, Broadcom, and other providers of cost-effective compute. For tech workers, inference is where the hiring will be: companies need engineers who can deploy, optimize, and scale AI systems at production volume, not just researchers training new models. For businesses buying AI services, the inference shift should eventually drive costs down as more providers compete for the operational workloads that represent the bulk of AI spending. If you're considering AI infrastructure investments, think about who runs the model after it's built, not just who trains it.