Qubic’s Useful Proof of Work - The Future of AI Compute

In the world of AI development, two major hurdles stand out: technological limits and economics. Large Language Models (LLMs), the technology behind ChatGPT, Claude, Gemini, and others, have already hit a point of diminishing returns when it comes to scaling with more traditional compute. That topic deserves its own article.

The second, and in my view far more pressing issue right now, is the cost, availability, and absolutely brutal economics of the datacenters that power today’s AI industry.

If you’ve spent more than five minutes around the Qubic community, you’ve probably heard the term Useful Proof of Work (UPoW). You may also know that Qubic’s miners successfully took over 51 % of Monero’s hashrate as a live demonstration of the concept. Marketing buzz aside, I believe this model is pointing toward the only realistic long-term solution for AI training and a recent article made that crystal clear to me.

The Article That Made Everything Click

Harris Kupperman, CIO of Praetorian Capital, recently published a piece that spells out the math behind the current AI datacenter boom with painful clarity. Key points from his analysis:

Data Center Overcapitalization

• 2025 AI datacenter spending is estimated at ~$400 billion.

• This creates roughly $40 billion in annual depreciation alone (10-year asset life).

• Current total AI industry revenue: only $15–20 billion.

• Depreciation is already 2× revenue.

• At an optimistic 25 % gross margin, the industry would need $160 billion in annual revenue just to cover depreciation — a 10× increase.

• To earn a modest 20 % Return on Invested Capital, it would require $480 billion in revenue per year — a completely unrealistic figure.

In short, the sector is turning asset-light tech companies into capital-intensive utilities with terrible returns. Without a miraculous, sustained revenue explosion, this path leads to massive value destruction.

Qubic Does the Exact Opposite

Qubic uses a Proof-of-Work consensus mechanism like Bitcoin, but with one critical difference: the compute power contributed by miners is not wasted on random hashing. Instead, it is directed toward useful tasks; specifically, training Qubic’s own AI model called AIgarth.

This is what Qubic calls Useful Proof of Work.

A centralized Corporation has to:

• Raise or spend hundreds of billions upfront

• Deal with huge fixed costs

• Wait years for build-out

• Eat massive depreciation every year

Qubic has none of these problems.

Thousands of individual miners and small datacenters all over the world voluntarily contribute their hardware. They are paid in QUBIC tokens for the useful work they perform. No giant capex budget, no multi-year construction delays, no crushing depreciation schedule.

The result? Qubic’s network is already one of the largest supercomputers on the planet, ranked #6 at the last evaluation, built in a completely decentralized, crowdsourced way.

Closing Thoughts

You can wait for the inevitable reckoning when the current datacenter bubble runs into the reality of the numbers.

Or you can start paying attention to the projects that are quietly solving the real bottleneck with sustainable, scalable, economically viable compute. Qubic exists in the corners of blockchain and decentralized protocols that most people still ignore and that is your advantage of being very early to the future narrative.

The future of AI isn’t waiting for the next $50 billion campus to come online.
It’s already training, right now, on Qubic’s Useful Proof of Work network.

—-

Reference: Harris Kupperman — “Global Crossing Is Reborn…” – Praetorian Capital
https://pracap.com/global-crossing-reborn/

🌐 Join the Revolution

Be early. Be active. Be recognized.
Join the discussion on X, Discord, and Telegram.
Visit the Qubic.org network to become a Computor, miner, or innovator.