Peter Thiel Predicts AI Chips Will Become "Dirt Cheap" as Alternatives Emerge

Silicon Valley investor Peter Thiel stated that artificial intelligence (AI) chips will eventually become inexpensive due to increasing competition and the development of alternative technologies. Thiel's remarks suggest that the current high profitability in the AI chip market, largely driven by Nvidia, may not be sustainable.

Nvidia's Market Position Challenged

Nvidia's significant profits in the AI sector have been attributed to its dominant position, particularly its CUDA software ecosystem, which has created a barrier to entry for competitors. This has led to a situation where AI companies have limited alternatives, contributing to higher GPU costs. However, this structure is facing challenges as other solutions mature.

Rise of Alternative Hardware

The market is seeing the emergence of viable alternatives to Nvidia's GPUs. AMD's hardware performance has improved, with its chips now competing closely with Nvidia's in certain inference and training scenarios. As the performance gap narrows, factors such as price, supply, and customization capabilities are expected to become more influential in purchasing decisions.

Additionally, Application-Specific Integrated Circuits (ASICs) are gaining prominence. ASICs are designed for particular tasks, offering higher efficiency, lower cost structures, and greater customization compared to general-purpose GPUs. With the convergence of large model training paradigms, ASICs are poised to become more effective for specific AI workloads.

Major Tech Companies Adopt Alternatives

Several major technology companies are already integrating alternative AI chip solutions. Google's Gemini 3 model operates entirely on its Tensor Processing Units (TPUs), indicating the maturity and viability of TPUs for core AI applications. Similarly, Anthropic, a competitor to OpenAI, utilizes Trainium for its extensive training and inference operations, highlighting its reliance on non-Nvidia hardware for compute-intensive tasks.

Potential Shift in Profit Dynamics

As the performance differences between GPUs, TPUs, ASICs, and Trainium diminish, industry observers anticipate a shift in profitability from the hardware layer to the application layer. This mirrors historical trends where hardware, once a primary profit driver, becomes a tool, with the main value generated in software and services. While Nvidia is not expected to collapse immediately, a change in growth logic and profit margins could lead to a re-evaluation of its market ceiling by capital markets. The transition from a "compute scarcity" to "compute abundance" environment could alter Nvidia's role in the AI ecosystem.

GPU-Backed Debt Concerns

Coinciding with Thiel's comments, reports have emerged regarding a financing model where data centers and cloud service providers use high-end Nvidia GPUs as collateral for loans. This model, exemplified by companies like CoreWeave, has seen the debt market for GPU-backed loans exceed $20 billion by mid-2025. CoreWeave alone has raised over $12 billion, with debt constituting a significant portion of its assets.

Concerns have been raised about the high interest rates, which can reach up to 10%, and the rapid depreciation of GPUs. As older chips are phased out, their collateral value is expected to decrease, posing risks to the private credit capital invested in this AI debt market. Some analysts have drawn parallels between this situation and the "asset bubble + high leverage" conditions observed before past financial crises, noting that the tech industry's current debt levels surpass those seen during the dot-com bubble of the 1990s. A systemic issue with Nvidia could have broader economic implications given its central role in the global computing and capital system.