Investors who once financed powerful graphics processing units (GPUs) for artificial intelligence are now turning to a different kind of chip. A recent $400 million loan backed by inference chips signals a new trend in AI infrastructure funding. GPUs have been essential for training AI models, but they consume a lot of energy and are not ideal for running models after they are trained. Inference chips are designed specifically to execute AI tasks efficiently, making them cheaper and faster for real-world applications. This shift is important as companies deploy AI in everything from chatbots to self-driving cars. The $400 million deal is one of the first major loans tied to inference chips. It suggests that lenders and investors believe inference hardware will be a key part of future AI systems. The move comes as demand for AI services grows, pushing companies to find more cost-effective ways to run their models. Experts say this could open the door for more specialized chips in AI data centers. While GPUs remain crucial for training, inference chips may dominate the next wave of infrastructure spending. The deal also highlights how financial markets are adapting to the evolving needs of the AI industry.