 |
| By Jurica Dujmovic |
Over the past few weeks, we’ve explored how the AI boom is quietly deflating. Not through a single crash, but through mounting economic pressure.
Costs are rising. Funding is drying up. And only a few giants can keep their lights on.
That’s why I clarified the red and green flags AI investors should look for in their potential plays.
But that’s still not the full picture.
Because the same forces squeezing AI today — scarcity of compute, energy and capital — are creating entirely new markets.
And some of those markets are being built on crypto rails.
The Real Limit to AI Growth
AI’s limiting factor is no longer data or algorithms. It’s infrastructure.
Training large models today demands tens of thousands of GPUs, dedicated networking and enormous power consumption. Which has made datacenter expansion the defining arms race of this decade.
Nvidia, Microsoft, Amazon and Google are spending like utilities, not software companies.
As I explained last week, this is not a sustainable business model.
Each new generation of hardware ages out within two or three years. Electricity costs climb faster than efficiency gains. And even if demand for AI services doubles, the margins don’t automatically follow.
Compute is a real, finite commodity.
The “cloud” isn’t abstract: It’s racks, chips and megawatts.
That physical constraint changes the nature of the AI industry. Instead of open innovation, we’re seeing consolidation. Smaller players can’t raise the billions needed to compete. And once access to compute becomes pay-to-play, the “AI revolution” starts to look like any other capital-intensive utility sector: a few producers, many renters.
Scarcity Always Invites Parallel Markets
But we’ve seen this setup before. And whenever an essential resource becomes expensive and concentrated, markets adapt.
History gives us plenty of examples …
- Electricity co-ops in the early 20th century,
- Independent oil producers during OPEC’s reign,
- Even broadband resellers in the early internet.
In AI, the scarce resource is compute. The opportunity is to make that resource tradeable, liquid and globally accessible.
That’s exactly what a new wave of crypto-based infrastructure projects is trying to do.
Render Network (RENDER, “B-”), for instance, aggregates idle GPU capacity from individuals and studios and resells it to the highest bidder.
It doesn’t need to run an expensive datacenter. Instead, it pays a fraction of that cost to harvest idle data from thousands of computers. Then, it can earn money on the sale.
Bittensor (TAO, “E-”) operates on a similar structure, rewarding nodes for contributing useful model output.
Others — Gensyn AI, Akash (AKT, “D+”), io.net (IO, “C-”) — extend the idea to generic compute and storage.
The common thread is simple: Tokenize the resource, turn it into a marketplace and let price discovery handle allocation.
These systems aren’t magic. They still depend on hardware and energy. But they approach the supply problem in a uniquely crypto way …
Instead of financing datacenters with debt, they finance participation through tokens.
The cost of scaling shifts from capex to incentives. Contributors earn as they provide capacity and users pay only for what they consume.
The network grows organically, node by node, without billion-dollar commitments up front.
Decentralization as a Solution
Centralized AI has two balance-sheet problems:
- Enormous, fixed costs that depreciate fast.
- A dependency on power and chip supply chains it doesn’t control.
Decentralized networks, if they work, sidestep both.
They convert spare capacity — whether consumer GPUs, academic clusters or enterprise overstock — into productive assets. In essence, these projects have tokenized hardware and bandwidth.
That’s a step toward real productivity, not speculative hype.
Of course, there are still very real challenges to this approach — performance variance, security, quality control and regulatory uncertainty still loom large.
But the direction is economically sound. If compute is the new oil, these networks are building the spot markets and futures exchanges for it.
The deeper idea is that compute itself is becoming money.
Tokens in these systems are not abstract claims. They’re payment for processing power, storage and inference. As demand for AI surges, these tokens reflect real utility demand, not just speculation.
It’s the digital equivalent of kilowatt-hours or bandwidth minutes being tokenized and tradable.
That blurs the line between two industries that until recently seemed unrelated: AI creates the demand, crypto provides the exchange mechanism.
Together, they may form a new kind of infrastructure economy.
3 Filters for Crypto AI Investors
For investors, this shift reframes the “AI opportunity.”
Instead of frantically wondering which model will win, the question becomes, “Who supplies the scarce inputs that every model needs?”
Three filters help separate signal from noise:
- Does the network trade a tangible resource?
Tokens backed by measurable compute, storage or data are inherently easier to value than “AI + crypto” branding exercises.
- Is there verifiable demand?
Usage metrics matter more than token price. Look for workloads, partnerships or real rendering and training activity.
- Does it complement — not copy — the centralized incumbents?
The viable projects will serve the long tail: smaller labs, indie studios, emerging markets. Competing head-on with Amazon’s GPU farms is futile — offering an alternative to those priced out of them is not.
This framework turns what looks like a speculative niche into a pragmatic hedge. If AI keeps consolidating, decentralized compute provides an outlet.
If AI stalls under its own costs, those same networks capture residual demand at lower price points.
The Broader Implication
What began as an AI “bubble” narrative is maturing into an industrial one.
We’re watching the sector move from experimentation to extraction. From idea generation to resource management.
The next phase won’t be defined by the cleverness of algorithms but by the efficiency of infrastructure.
That’s why crypto’s role here is worth tracking closely.
It’s not a replacement for AI. It’s the financial layer for distributing the hardware and energy that AI consumes.
The same way oil futures once stabilized a volatile commodity market, tokenized compute markets could stabilize — or at least price — AI’s runaway costs.
If that happens, the winners of the next decade won’t just be those who own the biggest models. They’ll be the ones who turn the essential inputs of intelligence — compute, data, bandwidth — into tradeable assets.
That’s where value will migrate as the bubble deflates. And it’s where a real, measurable economy of AI infrastructure will emerge.
As crypto investors, you have an edge. I encourage you not to waste it.
Best,
Jurica Dujmovic
