AMD CEO Lisa Su Says Companies Must Manage AI Costs Like Labor Costs
AMD CEO Lisa Su stated in an interview with CNBC that AI is not just a showcase of technology, but a tool to enhance corporate efficiency. Companies are exploring models that best fit their business, but as AI costs rise rapidly, they must manage AI costs like they manage labor costs.
She believes that in the future, there will not be just one model dominating the market, but rather a coexistence of large models, open-source models, and small to medium-sized models. Companies need to choose the appropriate AI tools based on different scenarios to drive the next phase of growth.
This reflects the public views of semiconductor executives on the commercialization of AI, focusing on cost control and model diversification. Corporate users benefit from practical selection advice, as high-cost single model paths are pressured by efficiency and ROI requirements.
Source: Public Information
ABAB AI Insight
Lisa Su positions AI as a manageable efficiency tool rather than an endless investment in a technology race. As the scale of inference and application expands, costs have moved from the experimental phase to a stage that requires meticulous control, similar to traditional labor budgets.
On the capital path, AMD emphasizes a diverse model ecosystem, which benefits its hardware penetration in various scale deployments. The motivation is to shift customers from 'usable' to 'affordable and effective', extending the AI investment cycle.
Similar to other chip manufacturers' emphasis on inference efficiency and TCO, we are currently in a stage where 'cost becomes a key variable in selection'.
Essentially, this represents a shift in pricing power: the value of AI is transitioning from model scale to scenario matching and cost efficiency, introducing budget discipline into technology adoption decisions.
ABAB News · Law of Cognition
- Once AI costs spiral out of control, efficiency tools will become a burden.
- No single model fits all scenarios.
- Managing AI budgets must be as serious as managing labor budgets.