Gavin Baker - AI Market Jitters - [Invest Like the Best, EP.485]

The market fears rising credit costs will stall the AI buildout. However, existing GPU compute is contracted at prices far below current spot rates. As these contracts expire, repricing will accelerate hyperscaler operating cash flow, allowing them to self-fund expansion without needing as much debt.

The market frets that cheaper open-source models cannibalize expensive frontier models. This is a misconception. Open source drives token elasticity, increasing total compute demand. It merely shifts high margins away from model providers to the underlying AI infrastructure players who provide the compute.

Investors are increasingly using AI models like Claude to interpret news, creating a "Walter Cronkite effect" where a single interpretation dominates. This breakdown in cognitive diversity leads to correlated trading behavior, causing entire market cycles to compress from years into weeks.

In the high-stakes AI chip race, breaking a Long-Term Supply Agreement (LTA) is corporate suicide. The risk of being cut off from future supply by a partner—who can simply allocate to your competitor—creates an unbreakable commitment, stabilizing the supply chain and pricing.

A new generation of AI-native companies is fundamentally restructuring its cost base. Instead of hiring more knowledge workers, they are allocating significant portions of their budget—up to 30% of what would be spent on compensation—directly to AI token consumption, driving massive productivity gains.

To bridge the financing gap for the massive AI buildout, NVIDIA is offering a clever new business model. They are essentially acting as a credit facilitator for their customers, and in exchange, they receive a share of the ongoing revenue generated by the GPUs, shifting their model towards recurring revenue.

The optimal strategy for enterprise AI is not to rely solely on expensive frontier models. Instead, companies use a powerful model like Claude or GPT-4 to plan tasks and then delegate the execution to cheaper, fine-tuned open-source models. This massively reduces cost while maintaining high performance.

Instead of running an entire inference task on a single GPU, the next efficiency leap will come from breaking it down. Tasks like prefill, attention, and feed-forward networks will be routed to specialized chips, such as SRAM-based accelerators, that are best suited for each job, dramatically improving performance and ROI.

The AI industry's failure to communicate its benefits (blue-collar job creation, community investment) has created a political narrative vacuum filled by fear. This has led to significant and misguided regulatory risks like data center moratoriums, which are entirely self-inflicted.