AI compute market is around $500B
Constellation's inference plan would turn spare gaming rigs and Macs into AI servers
They'd serve models while their owners aren't using them paid in
$DAG per request
What qualifies
• 16GB+ VRAM on Windows or Linux, or 32GB+ unified memory on a Mac
• Common cards that clear the bar: RTX 5070 Ti, RTX 5080, RX 9070 XT, plus 24GB cards like the 3090 and 4090
• On Mac, the M6 Mac mini can be configured up to 32GB and the M5 Pro Mac mini reaches 64GB
How much is out there
• 16GB is the most common VRAM tier on Steam, at 26.92% in August
• In July, another 5.43% ran 24GB
• In one retail count, none of 15 current cards above $500 ship with 8GB so new mid-to-high-end buyers clear the bar by default
What it could mean for
$DAG
• About 400 node operators run on MainNet, IntegrationNet and TestNet today, with roughly 600 more on the waiting list
• The new validator set is cutting back to 30
• Alex's own server ran $809 in tokens at API rates over 30 days
If just 200 of those 1,000 people have qualifying hardware and fill it with paid requests at that rate, that's about $160K a month in
$DAG bought on the open market
These operators are already invested in the ecosystem