What is the difference between AI training and inference that makes AMD competitive?

Training builds and refines AI models and requires massive raw computing power, where Nvidia's CUDA software gives it a durable advantage. Inference is where a trained AI produces answers—it is much less technically demanding and often depends more on fast memory access than raw compute. AMD's chiplet design allows for larger Key-Value cache and more memory per GPU, and its acquisition of MEXT adds AI-driven software to optimize memory costs, making it well-suited for inference at lower cost.

Why does AMD's smaller market cap matter if it is more expensive per share?

AMD has a market cap of less than $900 billion（約140兆円） compared to Nvidia's $5 trillion（約800兆円）. Even though AMD trades at a higher forward price-to-earnings ratio (39.5x versus under 16x), its smaller absolute size means it has room to grow before hitting the limit of revenue expansion faced by much larger companies. The analysis suggests this size difference could translate to more upside over the next decade.

What is agentic AI and why does it favor AMD?

Agentic AI refers to AI systems that make decisions and manage tasks sequentially on their own, requiring CPUs to provide the logic needed to coordinate multiple steps. AMD is a leader in data center CPUs. For agentic AI, the GPU-to-CPU ratio is expected to shift from 8:1 (as in training) to 1:1, creating what AMD sees as a $120 billion（約19兆円） addressable market where CPUs become as important as GPUs.

Back to articlesAI Stocks & Markets

AI Stocks & Markets

Nvidia and AMD are positioned differently in AI's next phase: Nvidia dominates training, but AMD may have more growth runway in inference and agent AI despite trading at a premium valuation.

Yahoo Finance AI1h ago5 min read

Key takeaway

Nvidia and AMD have emerged as the largest AI stock winners of the past decade, up over 16,320% and 9,770% respectively, but they are now positioned in different parts of the AI value chain. Nvidia dominates AI model training through its entrenched CUDA software platform and has built an end-to-end AI infrastructure business, while AMD is gaining ground in inference (where AI produces outputs) and agentic AI. Although AMD trades at a much higher valuation (39.5x forward P/E versus Nvidia's under 16x), its smaller size and positioning in two rapidly growing segments suggest it may have more upside over the next decade.

Summaries like this, in your inbox every morning.

3 Key Points

What happened
A strategic analysis compares Nvidia and AMD's AI chip positions. Nvidia built dominance in AI model training through its CUDA software platform and has expanded into a complete AI infrastructure company through acquisitions including Mellanox and Groq. AMD, historically behind in training, has strengthened its position for inference (where AI produces answers) and announced an acquisition of MEXT, a memory optimization company, to compete in cost-effective AI server solutions.
Why it matters
The two chips address different parts of AI's value chain. Nvidia's forward price-to-earnings ratio stands at under 16 times fiscal 2028 estimates, while AMD trades at a 39.5x one-year forward P/E, reflecting different growth expectations. AMD's smaller market cap—less than $900 billion（約140兆円） compared to Nvidia's $5 trillion（約800兆円）—and positioning in two emerging areas (inference and agentic AI, which requires CPUs) suggest it may face less headwind from sheer scale, though at a higher current valuation.
What to watch
AMD has two $100 billion（約16兆円） GPU inference deals in place and sees a $120 billion（約19兆円） addressable market in agentic AI as the GPU-to-CPU ratio shifts from 8:1 for training to 1:1 for agentic AI. The inference market is expected to eventually grow larger than the training market. Both companies remain well-positioned for long-term AI infrastructure demand, but their near-term growth drivers differ significantly.

FAQ

What is the difference between AI training and inference that makes AMD competitive?: Training builds and refines AI models and requires massive raw computing power, where Nvidia's CUDA software gives it a durable advantage. Inference is where a trained AI produces answers—it is much less technically demanding and often depends more on fast memory access than raw compute. AMD's chiplet design allows for larger Key-Value cache and more memory per GPU, and its acquisition of MEXT adds AI-driven software to optimize memory costs, making it well-suited for inference at lower cost.
Why does AMD's smaller market cap matter if it is more expensive per share?: AMD has a market cap of less than $900 billion（約140兆円） compared to Nvidia's $5 trillion（約800兆円）. Even though AMD trades at a higher forward price-to-earnings ratio (39.5x versus under 16x), its smaller absolute size means it has room to grow before hitting the limit of revenue expansion faced by much larger companies. The analysis suggests this size difference could translate to more upside over the next decade.
What is agentic AI and why does it favor AMD?: Agentic AI refers to AI systems that make decisions and manage tasks sequentially on their own, requiring CPUs to provide the logic needed to coordinate multiple steps. AMD is a leader in data center CPUs. For agentic AI, the GPU-to-CPU ratio is expected to shift from 8:1 (as in training) to 1:1, creating what AMD sees as a $120 billion（約19兆円） addressable market where CPUs become as important as GPUs.

Discussion

No comments yet. Be the first to share your thoughts!

Nvidia dominates AI chip revenue with $75.2 billion（約12兆円） in data center sales, far outpacing AMD, Broadcom, and Marvell, while trading at a cheaper valuation than its peers.

Yahoo Finance AI1h ago

Micron's blockbuster earnings and $22 billion（約3.5兆円） in locked-in customer contracts signal a structural shift in memory chip economics, potentially ending decades of boom-bust volatility in the semiconductor industry.

Fortune AI3h ago

NVIDIA and Micron both posted blockbuster AI infrastructure quarters, but the market rewarded them opposite ways—NVIDIA's stock fell 10.38% while Micron surged 127.9%—reflecting different bets on where AI hardware spending lands in 2026.

Yahoo Finance AI6h ago

At VivaTech, Europe confronts AI's cybersecurity risks and sovereignty gaps while businesses demand proof that AI spending delivers real returns.

Fortune AI6h ago

Wheels Up, a major private aviation company, will become the first customer to deploy Surf Air Mobility's Enterprise BrokerOS software, which uses Palantir's AI platform to streamline charter booking and decision-making.

Yahoo Finance AI9h ago

AMD and Rackspace sign a deal to deploy 30 MW of AMD-based computing capacity across Rackspace's data centers starting in late 2026, positioning AMD as the silicon partner for enterprise AI workloads.

Yahoo Finance AI12h ago

Stay ahead with AI news

Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.

Get Started Free

Free · takes 30 seconds · unsubscribe anytime

5 minutes a day. The AI essentials.

200+ sources · Email / LINE / Slack

Get it free →