The AI hardware boom is running on financial faith. Cerebras raised its IPO price target this week, seeking a valuation as high as $48.8 billion. That’s a 70x multiple on its run-rate revenue, a bet that hinges on massive, possibly optional, deals with OpenAI. Jason Calacanis labeled it one of tech’s highest-stakes games on This Week in Startups. If the promised revenue fails, the capital expenditure could bankrupt the business.
“The price-to-sales ratio suggests a total disconnect from traditional market reality.”
- Alex Wilhelm, This Week in Startups
The compute shortage is deepening beyond GPUs. On State of Agentic Coding, Ben Vinegar noted that RAM and NVMe drive prices are climbing as data centers prioritize prompt caching for agentic workflows. He detailed a fivefold cost increase for his code review tools in a single month as providers switched to usage-based pricing. The era of subsidized AI is closing, forcing SaaS companies toward strict per-use billing.
Meanwhile, the physical supply chain for this expansion is half a century behind China’s. On The a16z Show, Turner Caldwell, CEO of Mariana Minerals, argued the US trails by five decades in critical mineral supply chains. His company is treating mining as a software problem, using reinforcement learning to manage refinery feedstocks, aiming to deliver ten major projects in a decade. It’s part of a vertical integration playbook borrowed from Tesla.
Heron Power CEO Drew Baglino applied the same logic to the grid, advocating for solid-state transformers to replace pre-WWII mechanical systems. He argued the US must create a federal highway trust fund equivalent for the grid to support AI’s energy demand. The goal is a co-located supply chain, mimicking China’s clustered industrial bases.
“Automation solves the labor cost differential. In modern factories, labor accounts for less than 10% of the cost of goods sold.”
- Drew Baglino, The a16z Show
The valuation surge rests on a fragile premise: that exponentially growing demand will outpace these physical and financial constraints. The industry is betting it can automate its way out of a fifty-year infrastructure gap before the bill for cheap tokens comes due.
