Record GPU Shortage, Large Model Labs Monopolize Supply
Yuchen Jin pointed out that the GPU shortage is more severe than ever, with the price of the H100 currently higher than it was three years ago and not available on demand.
Large AI labs have locked in most of the supply for several years, making it difficult for university researchers and independent developers to obtain GPUs.
AI developers and academic institutions in the market face computational power barriers, while large labs maintain their lead through long-term contracts. Cloud service providers and labs with GPU resources benefit, while small developers and universities face short-term pressure, accelerating the concentration of funds towards GPU infrastructure and alternative computing solutions.
Source: Public Information
ABAB AI Insight
Yuchen Jin, as a researcher with an academic background, has previously discussed the accessibility of AI infrastructure. This warning continues the trend of supply-demand imbalance for H100/H200 since 2023, with large labs like OpenAI, xAI, and Anthropic locking in supplies of NVIDIA's high-end cards through long-term pre-purchase contracts and self-built data centers.
In terms of capital pathways, large companies are converting hundreds of billions of dollars in financing directly into long-term GPU procurement and data center construction, creating a de facto supply chain monopoly. The motivation is to use computational power as a core competitive barrier, preventing newcomers from catching up through scale locking, while turning GPU scarcity into pricing power for model training and inference.
Similar to how early cloud computing was monopolized by AWS, leading small businesses to multi-cloud solutions, and how graphics card supply was locked by mining farms during the Bitcoin mining era, the current AI industry is at a critical stage of transitioning from a model parameter competition to a monopoly of physical computational resources, systematically pushing universities and independent developers out of cutting-edge research.
Structural judgment: This essentially belongs to capital concentration. GPUs, as the "new oil" of the AI era, are being locked in early by a few large labs through massive capital. The mechanism is that NVIDIA's production expansion speed is far lower than the growth in demand, leading to a high concentration of scarce resources among top players who already have significant funds and contractual advantages, forcing innovation and research capabilities to shift from dispersed academic/individual efforts to a few closed large model platforms.
ABAB News · Cognitive Law
The scarcer the resource, the earlier it is locked.
Large companies lock GPUs, while small ones can only dream.
Computational power is the barrier, monopoly is the future.