GROKIPAEDIA
β‘ THE AI COMPUTE WARS β‘
π₯ $200B+ Race for GPU Supremacy
The most concentrated burst of computing power in human history. Watch companies stack GPUs by the hundreds of thousands in their race to build superintelligence.
2.5M+
Total H100-Equivalent GPUs
Deployed globally by end of 2025
$200B
Total Investment
Hardware + infrastructure
15 GW
Power Consumption
Enough for 15 nuclear plants
555K
Largest Single Cluster
xAI's Colossus (Jan 2026)
π₯οΈ GPU Cluster Visualizer
See the Scale: Pick a Company
xAI
Meta
Microsoft
Google
OpenAI
xAI Colossus
Total GPUs
555,000
Investment
$16.7B
Power Draw
2.0 GW
Training Capacity
50 ExaFLOPS
β‘ THE POWER CONSUMPTION CRISIS
xAI Colossus
2.0 GW
= Powering the entire city of Memphis, Tennessee
= 35 Hoover Dams running at full capacity
= 35 Hoover Dams running at full capacity
Future: 50 ExaFLOPS
35 GW
= 35 nuclear power plants
= All of New York State's electricity
= All of New York State's electricity
Single H100 GPU
700W
= 3.7 MWh per year
= Enough to power 3 average US homes
= Enough to power 3 average US homes
β οΈ The Unsustainable Reality
Elon Musk's 50 million H100-equivalent target by 2030 would consume 35 gigawatts. That's more power than entire countries use. Even with next-gen efficiency improvements, we're talking 4.6+ GW minimumβtriple what Colossus 2 will require. The grid can't handle this. Nuclear plants take a decade to build. Data centers are turning to jet engines and diesel generators just to turn on the GPUs they already own.
π THE ARMS RACE: WHO'S BUYING WHAT
GPU Count by Company (2026)
π° WHAT DOES IT ACTUALLY COST?
Training Cost Calculator