AMD surges 11% on AI inference push as MI350, ROCm gain traction

AMD's 11% surge reflects growing conviction that its MI350 accelerators and ROCm software can capture a meaningful slice of the AI inference market.

Advanced Micro Devices Inc. is emerging as a more credible contender in AI infrastructure as inference workloads shift the competitive dynamics, with its MI350 accelerators and ROCm software stack narrowing the gap against Nvidia's dominant CUDA platform.

"AMD is becoming a more serious contender in AI infrastructure as inference demand, MI350 accelerators, and ROCm improvements reshape the bull case," Rick Orford, an analyst at The Motley Fool, said.

The stock surged 11.01% on June 7, reflecting investor optimism that AMD's AI strategy is gaining traction beyond the training market where Nvidia holds an estimated 80% share. AMD's MI350 series, built on TSMC's advanced process nodes, targets inference workloads — the process of running trained AI models — where cost efficiency and latency matter more than raw training throughput. The company's ROCm software platform, a direct competitor to Nvidia's CUDA environment, has improved developer tooling and library support, reducing the switching cost for AI teams evaluating alternatives.

The rally pushes AMD's valuation into a higher bracket, raising the stakes for the company's ability to deliver earnings growth that matches the optimism. AMD must demonstrate that its MI350 roadmap can convert inference demand into sustained revenue gains, or the stock risks repricing lower. The next catalyst is the company's quarterly earnings report, where investors will scrutinize data center segment revenue and MI350 adoption metrics.

The Inference Opportunity vs. Nvidia's Moat

The AI chip market is bifurcating. Training large language models requires massive parallel compute, a domain where Nvidia's H100 and upcoming Blackwell GPUs dominate with an estimated 80% to 90% market share. But inference — the deployment phase where trained models generate responses — is a different game. Inference workloads prioritize memory bandwidth, latency, and cost per query over raw floating-point operations.

AMD's MI350 accelerators are designed for this shift. The chips leverage high-bandwidth memory (HBM) and AMD's Infinity Architecture to move data efficiently, a critical advantage when serving AI models to millions of users. Meanwhile, ROCm's growing library support — including optimizations for PyTorch and TensorFlow — lowers the barrier for developers who have historically defaulted to CUDA.

Nvidia is not standing still. Its Blackwell architecture, expected to enter mass production in late 2026, includes inference-optimized features such as a dedicated transformer engine and improved sparse computation support. But AMD's pricing strategy — historically undercutting Nvidia by 20% to 30% on comparable hardware — could pressure Nvidia's data center gross margins, which stood at 78% in its most recent fiscal year.

Valuation Risk Hinges on Execution

AMD's stock has more than doubled over the past 12 months, pricing in a significant AI revenue contribution. The company trades at a forward price-to-earnings multiple that assumes data center segment revenue growth of at least 50% annually over the next two years, according to consensus estimates. If MI350 adoption falls short or Nvidia responds with aggressive pricing, those assumptions unravel.

The broader AI infrastructure buildout provides a tailwind. Global spending on AI data centers is projected to exceed $200 billion in 2026, according to industry estimates, with inference accounting for a growing share. AMD's ability to capture even 10% to 15% of the inference chip market would represent billions in annual revenue. But execution risk remains high: software platform lock-in is Nvidia's strongest moat, and developer habits change slowly.

For investors, the AMD thesis rests on a single question: Can earnings grow fast enough to justify the valuation? The MI350 and ROCm roadmap offers a credible path, but the market has already priced in significant success. Nvidia shares, trading at roughly 35 times forward earnings, reflect a similar premium — meaning both stocks carry elevated expectations. If AMD delivers on its AI roadmap, the upside could be substantial. If it stumbles, the downside is equally large.

This article is for informational purposes only and does not constitute investment advice.