The Beginning of Scarcity in AI

The AI gold rush is hitting a wall. It has been reported that GPU rental prices for Nvidia’s Blackwell chips spiked to $4.08 per hour this week — up 48% from $2.75 just two months ago — and that CoreWeave raised prices 20% while extending minimum contracts from one year to three. It has been reported that OpenAI CFO Sarah Friar warned, “We’re making some very tough trades at the moment on things we’re not pursuing because we don’t have enough compute.” Short supply. Big consequences.

Access becomes a privilege

It has been reported that Anthropic has limited access to its newest model to roughly forty organizations. That’s a carve‑out of the bleeding edge; relationship selling replaces open platforms. Observers have sketched five hallmarks of this moment: relationship‑based selling; “AI to the highest bidder”; available-but-slow service even for paying customers; inflationary commodity dynamics pushing prices up; and forced diversification toward smaller models or on‑prem solutions. The era of abundant, cheap compute? It has been reported that it’s over — for now.

What it means for startups and the market

Who gets the keys to the kingdom? Bigger, well‑funded firms and strategic partners, mostly. Startups face a squeeze: either raise more money, pay higher prices, or downgrade expectations by using smaller models or slower access. The Wall Street Journal has reported that AI’s energy appetite is straining infrastructure, and buildouts to catch up could take years — so this isn’t a short blip. Procurement and margin management suddenly matter as much as model architecture.

If true, this shift will reshape competition and creativity in AI. Expect more “workarounds”: on‑prem deployments, niche models, and fresh bets on efficiency. It’s a frustrating moment for many — thrilling for a few. And it begs a blunt question: are we entering an age of gated AI, or just a rough patch on the way to new capacity? Either way, the party’s tempo just changed.

Sources: tomtunguz.com, Hacker News