AI computing power has been back in the conversation these past couple of days.
Not because of Nvidia, but because someone wants to provide computing power through mining.
The logic sounds pretty straightforward.
Mining used to make GPUs calculate meaningless hashes, burning electricity on idle work.
Now the idea is to use them for AI matrix operations and inference.
Miners get rewards, AI gets computing power—it seems like both sides win.
I’m on board with this direction.
Computing power is expensive, and there really are a lot of idle GPUs.
Putting mining hardware to work on something useful is better than simply burning electricity.
But there’s one question I keep coming back to:
Being useful and being used are two different things.
A GPU can run inference, but if nobody places an order or pays for it,
then essentially, that computing power is still just mining coins—it’s only been given a new name.
Some research has already raised doubts, suggesting that much of the computing power on a number of these early networks went toward tasks nobody needed.
So when I look at this sector, I don’t focus on how much computing power it has.
I look at demand.
Are there real AI teams willing to pay for this computing power?
Can it be cheaper than cloud providers, while still offering comparable reliability?
What share of miners’ income comes from paid usage, and how much still relies on newly issued token subsidies?
There’s another point that’s easy to overlook.
These projects often have very little circulating supply early on, with tokens gradually released through mining later.
Miners have electricity bills and hardware costs, so they have a reason to sell every day.
Without real revenue to absorb that supply, the token price can only be propped up by new stories.
My take is this:
Mining for AI computing power is the right direction, but for now, most projects are still at the storytelling stage.
The sector will only be truly up and running once revenue from paid orders exceeds token issuance subsidies.
Not because of Nvidia, but because someone wants to provide computing power through mining.
The logic sounds pretty straightforward.
Mining used to make GPUs calculate meaningless hashes, burning electricity on idle work.
Now the idea is to use them for AI matrix operations and inference.
Miners get rewards, AI gets computing power—it seems like both sides win.
I’m on board with this direction.
Computing power is expensive, and there really are a lot of idle GPUs.
Putting mining hardware to work on something useful is better than simply burning electricity.
But there’s one question I keep coming back to:
Being useful and being used are two different things.
A GPU can run inference, but if nobody places an order or pays for it,
then essentially, that computing power is still just mining coins—it’s only been given a new name.
Some research has already raised doubts, suggesting that much of the computing power on a number of these early networks went toward tasks nobody needed.
So when I look at this sector, I don’t focus on how much computing power it has.
I look at demand.
Are there real AI teams willing to pay for this computing power?
Can it be cheaper than cloud providers, while still offering comparable reliability?
What share of miners’ income comes from paid usage, and how much still relies on newly issued token subsidies?
There’s another point that’s easy to overlook.
These projects often have very little circulating supply early on, with tokens gradually released through mining later.
Miners have electricity bills and hardware costs, so they have a reason to sell every day.
Without real revenue to absorb that supply, the token price can only be propped up by new stories.
My take is this:
Mining for AI computing power is the right direction, but for now, most projects are still at the storytelling stage.
The sector will only be truly up and running once revenue from paid orders exceeds token issuance subsidies.