🤖💲 AI’s compute crunch turns Silicon Valley rivals into desperate deal-seekers

The AI boom has hit a critical bottleneck: computing power.

➡️ chips (GPUs) do the heavy mathematical work needed to train AI models

➡️ data centers house chips, along with the electricity, cooling and networking needed to keep them running

➡️ computing power is the processing capacity all this hardware provides to advance the AI boom

With chip capacity scarce, costs soaring and not enough data centers, rivals are scrambling for resources, reports the Wall St. Journal.

🟥 Anthropic co-founder Tom Brown reportedly visited xAI to negotiate access to computing power, before the Claude maker struck a deal to lease over 300 MW from Musk’s SpaceX as part of a deal worth up to $45B

🟥 Anthropic CEO Dario Amodei is said to have approached Meta’s chief AI officer for more capacity

🟥 Google is also battling over compute, with co-founder Sergey Brin reportedly overriding chip allocations

🟥 startups are squeezed hardest, with deals collapsing, and venture-capital firms dangling access to lure them in

Challenges pile up:

🔴 hourly rental costs for Nvidia H100 chips on one-year contracts have jumped 60% in a year (SemiAnalysis)

🔴 cloud providers demand multiyear commitments and hefty upfront payments

🔴 roughly 50% of US AI data centers planned for 2026 deployment may be delayed or canceled (Sightline Climate)

🔴 US data-center developers could face a 32 GW net power shortfall by 2028 - equivalent to about 34% of their projected electricity requirements (Morgan Stanley estimates)

👍 Substack | Chat | @geopolitics_prime