The artificial intelligence (AI) data center chip market is projected to reach $286 billion by 2030, up from $123 billion in 2024. Yet while investors chase Nvidia (NVDA +2.59%) to astronomical highs, they're overlooking a critical question: What happens when the AI market shifts from training models to running them at scale?
Advanced Micro Devices (AMD +3.50%) might not dominate today's AI training landscape, but the company's aggressive roadmap and cost advantages position it perfectly for the coming inference revolution. The recent 13% post-earnings dip could be a gift for contrarian investors.
Image source: Getty Images.
The $286 billion opportunity nobody's talking about
The real money lies in inference -- running AI models billions of times each day across enterprise applications. That shift directly benefits AMD's data center business, which already produced about $12.6 billion in revenue in 2024, up 94% year over year. Management now sees inference as the catalyst that could expand that segment into "tens of billions" in annual AI revenue.
Nvidia may control roughly 94% of the discrete GPU market versus AMD's 6% and Intel's (INTC +1.05%) sliver, according to Jon Peddie Research, but AMD has shown it can upend entrenched positions. In server CPUs, its EPYC line has grown to about 27% unit share and 41% revenue share, up from single digits just five years ago, per Mercury Research. That track record proves AMD can carve out meaningful share when it delivers superior price-performance.
NASDAQ: AMD
Key Data Points
The China card and valuation story
Export restrictions once hammered AMD shares, but media reports now suggest both AMD and Nvidia have received approval to resume China AI chip shipments under a 15% revenue-sharing agreement with the U.S. government. While unprecedented, the policy reopens access to a multibillion-dollar market that had effectively been shut off.
Meanwhile, Intel continues to fade -- holding virtually no discrete GPU market share and steadily losing ground in server CPUs. That weakness leaves a wider opening for AMD to expand its footprint in critical data center markets.
As of Sept. 8, AMD's stock closed at $151.41, equating to roughly 25 times forward earnings. While that's a slight premium to the broader market, it's well below the multiples carried by AI leaders growing at similar or even slower rates. This gap suggests investors remain skeptical of AMD's ability to sustain its data center momentum, which creates an opportunity if the company executes.
Risks worth understanding
Nvidia still holds the advantage in software, with CUDA serving as the gravitational center for developers. AMD's ROCm 7 update brings performance gains and new data types, but shifting developer habits is a long-term battle. At the same time, AMD's gaming division continues to lag, with only a roughly 6% discrete GPU market share, leaving the company highly dependent on data center growth.
Geopolitics and competition add further uncertainty. The 15% revenue-sharing agreement tied to China shipments highlights how regulatory changes can alter AMD's economics overnight. Hyperscalers are also testing custom silicon and deals with alternatives like Broadcom, while memory supply limits and power constraints could slow deployment even if demand remains strong.
The contrarian opportunity
Training may dominate today's headlines, but inference is where the real scale and profitability lie. If inference grows to 70% of a $286 billion market by 2030 and AMD captures just 25% of it, that would mean roughly $50 billion in high-margin revenue -- more than double AMD's entire 2024 top line from a single product category.
The market, however, is still pricing AMD as a follower rather than a contender. At around $150, the stock trades at only a modest premium to the broader market, despite nearly doubling data center sales last year.
That disconnect reflects skepticism about AMD's ability to break Nvidia's grip -- skepticism that creates opportunity. For investors willing to look past quarterly noise, AMD represents one of the most asymmetric risk-reward setups in technology today.
