Lisa Su does not do hype. The AMD CEO has spent years methodically building the company’s AI hardware business while allowing the results to speak for themselves, so when she says something is a turning point, it draws attention.

In a Yahoo Finance interview on July 24, 2026, Su said AI has reached an inflection point at which more people are doing genuinely useful, meaningful work with it. That framing matters because it shifts the conversation from model benchmarks and training runs toward real-world usage, where demand is tied to how often AI systems are actually deployed.

The Anthropic deal is the main headline

Su’s comments came two days after AMD announced a major partnership with Anthropic, the AI safety company behind the Claude model family. Under the deal, AMD plans to deploy up to 2 gigawatts of its latest Instinct MI450 GPUs, integrated into Helios rack systems. The first gigawatt of that deployment is expected to be operational in the first half of 2027.

AMD is also making an equity investment of up to $5 billion in Anthropic, underscoring how closely compute supply, infrastructure planning and model development are now linked in the AI industry.

Inference is becoming the bigger workload

The deeper signal in Su’s interview is a structural change in how the AI industry uses compute. She pointed to 2026 as a historical milestone, saying it is the first year global inference compute is projected to surpass training compute.

Training is the process of building an AI model and requires massive, concentrated bursts of GPU power running for months. Inference is what happens when that model is actually used — every time someone asks ChatGPT a question, every time an AI agent processes a document, and every time a customer service bot handles a ticket.

That shift helps explain why the ecosystem is looking less like a simple hardware-vs-software race and more like a set of long-term capacity decisions across chipmakers, cloud operators and model developers. Su described the broader AI ecosystem as increasingly intertwined, with hardware providers, model developers and cloud operators working more collaboratively than competitively.

AMD has been building toward this position

The Anthropic agreement did not come out of nowhere. AMD announced a partnership with OpenAI in October 2025, involving 6 gigawatts of GPU capacity. It then followed that with a Meta partnership in February 2026. The Anthropic announcement in July 2026 completes a trio of relationships with three of the most consequential AI model developers in the market.

AMD has reportedly committed around 20% of its equity across these AI partnerships.

What to watch next

First, if Su’s projection is correct and 2026 marks the crossover point where inference exceeds training compute globally, the industry’s most important buying decisions may continue shifting toward systems built for ongoing deployment rather than one-time model training.

Second, the equity investment model is becoming a pattern. AMD’s commitment of up to $5 billion to Anthropic resembles the strategic capital commitments that Microsoft made to OpenAI and that Google has made to Anthropic previously.

Third, the first-half-2027 timeline for the initial Anthropic gigawatt is the next concrete milestone to watch.