As Nvidia prepares for its fiscal third-quarter earnings report on November 17—the first since Jensen Huang’s departure on November 12—a curious market dynamic is emerging. While the broader crypto-AI token market has faced significant downward pressure, decentralized AI networks are quietly capturing real-world demand from customers unable to secure GPU capacity from hyperscale cloud providers. With Nvidia guiding toward $108 billion in revenue, investors are left to weigh whether these decentralized alternatives are building sustainable businesses or merely riding a wave of speculative token incentives.

The Disconnect Between Tokens and Utility

The divergence between the plummeting value of AI-linked tokens and the rising revenue of the networks behind them is stark. Bittensor’s TAO token, for instance, has struggled, trading roughly 70% below its all-time high. Yet, the underlying utility is expanding. Data from SubConnect indicates that Bittensor’s subnets have scaled to approximately $70 million in annual revenue, doubling their pace from just two months prior. Similarly, Lium, a specialized GPU rental subnet, saw a massive 87% surge in monthly customer spending, reaching $2.7 million.

Customers don’t buy tokens, they buy GPU hours, and those are priced in dollars. That’s why demand and token prices can move in opposite directions. — Greg Osuri, Akash Network

Why Startups Are Flocking to Decentralized Compute

The primary driver for this shift is accessibility. Major cloud providers often prioritize long-term, high-volume contracts for their largest enterprise clients, leaving smaller startups in a perpetual queue. Decentralized networks offer a nimble alternative, providing immediate access to hardware like H100 chips without the bureaucratic friction of traditional cloud giants. For researchers and developers working on open-weight models, these networks serve as a vital, cost-effective spot market for compute.

The Competitive Landscape

  • Inference vs. Training: Decentralized networks are not replacing hyperscalers for massive model training; they are positioning themselves as the go-to infrastructure for inference.
  • Cost Efficiency: By leveraging underutilized hardware in regional data centers and private homes, these networks often undercut traditional cloud pricing.
  • The Trust Factor: Beyond price, some users are turning to decentralized protocols to avoid the data-privacy risks associated with centralized, proprietary clouds.

The Sustainability Challenge: Revenue vs. Rewards

A critical question for investors remains: what happens when the token subsidies dry up? Industry experts warn that many projects are currently operating at a deficit, where the value of distributed tokens far outweighs the actual revenue generated from paying customers. Pieter van Poecke of NextBlock VC emphasizes that emissions should be treated as a cost, noting that a network earning $10 million while issuing $100 million in tokens is essentially masking a massive gap. The ultimate test for these platforms will be their ability to provide the reliability, auditability, and financial controls that enterprise-grade clients demand.

The Future of Decentralized AI

As the sector matures, the focus is shifting toward verification and honest output. Whether through random rechecks of completed jobs or hardware-level validation, the industry is racing to make honest work the most profitable path for participants. For investors, the signal is clear: look past the marketing dashboards and focus on on-chain lease data and repeat-customer metrics. The projects that will endure are those where the token serves as a bridge for value capture, not just a temporary subsidy for hardware.