AI Spotlight: Inference Capital Markets - UsePod Vs. Surplus Intelligence
Demand for compute has surged exponentially over the past year as more people realize the benefits of incorporating AI into their businesses and personal lives.
OpenRouter, a leading aggregator that unifies access to hundreds of AI models from dozens of providers, has experienced explosive growth over the past year, with weekly volume skyrocketing from ~2 trillion tokens to over ~25 trillion in the past year.
But the shift runs even deeper. We’ve moved far beyond AI as a mere chat assistant. Open-source frameworks like OpenClaw and Hermes Agent let anyone turn a personal desktop or server into a fully autonomous AI agent capable of collaborating with other agents.
AI Agents And ZHCs
AI agents are paving the way for the rise of Zero Human Companies (ZHCs). These are organizations that are run almost entirely by AI agents, with a single human founder setting high-level direction from the helm.
This may seem like science fiction, but it's already becoming reality. Consider how Argentina has already drafted legislation to make ZHCs legal entities.
Early experiments are already underway. Projects like Juno and KellyClaude are experimenting with ZHC models on the Base blockchain. While their market caps have cooled significantly since the initial hype earlier this year, both teams continue building.
An organization run entirely by AI agents would demand large amounts of compute/inference. So the question arises, from where are they going to get it?
Disadvantages Of Centralized Compute
ZHCs could rely on centralized AI providers such as OpenAI's ChatGPT or Anthropic's Claude. However, this approach comes with several drawbacks:
Censorship: Major AI companies embed extensive safeguards and content filters into their AI models. These can abruptly interrupt an autonomous agent in the middle of a critical workflow.
Revokation: Even worse, an entire model can be revoked or disabled for all users. Just last week, the US government ordered Anthropic to suspend access to its latest model, Fable 5 (along with Mythos 5).
Subscriber Lock-in: Centralized AI providers are heavily subsidizing compute today. Users receive thousands of dollars worth of inference for a few hundred dollars per month. This is unsustainable. If venture capital dries up, providers could suddenly hike prices, leaving AI agents and ZHCs with crippling cost spikes.
Logging: Running inference through Anthropic or OpenAI means handing over your entire company’s data including trading strategies, confidential conversations, and workflows.
Beyond the enormous environmental and energy costs, these drawbacks make a compelling case for decentralizing artificial intelligence.
By shifting toward open-weight models, strong privacy frameworks, and transparent compute marketplaces, we can build ZHCs that are sovereign, efficient, and truly autonomous.
Decentralized Compute Marketplaces
Imagine this: Your AI agents complete their day’s work with surplus compute credits still available. Instead of letting them expire, the agents automatically resell those unused tokens on a decentralized marketplace to buyers who need capacity immediately.
Several innovative projects are already bringing this vision to life. UsePod on Solana and Surplus Intelligence on Base are building compute marketplaces specifically designed for AI agents and ZHCs, enabling autonomous trading of inference capacity.
UsePod ($SQUIRE)
Whether you’re locked into a monthly Anthropic subscription or staking Venice’s DIEM tokens for daily inference, your AI agents rarely use their full quota. The compute you’ve already purchased then often sits idle, and simply expires.
UsePod solves this inefficiency. Users can supply their API keys from providers like Anthropic or Venice, allowing the network to resell any unused inference credits on a pay-per-use basis. Buyers get access at discounted rates (paid in USDC) while sellers earn from their surplus.
Additionally, hardware providers with GPUs can also host open-weight models to earn USDC per inference request. UsePod has even greater ambitions when in comes to privacy.
In the near future, AI agents will be able to lend their own TEE-enabled GPUs to the UsePod marketplace. Crucially, all inference requests will run inside encrypted TEE enclaves, ensuring the hardware operator sees no plaintext at all.
If you want to learn more about the UsePod framework, I suggest reading their whitepaper and docs.
Surplus Intelligence ($SURPLUS)
Built on Base, Surplus Intelligence is a peer-to-peer marketplace for buying and selling discounted inference. Similar to UsePod, it focuses on intelligently routing each request to the cheapest available supply, whether from surplus API credits or local GPUs.
While Surplus Intelligence places less emphasis on hardware-level privacy compared to UsePod, it functions as a true “decentralized AWS for inference” with on-chain verification and transparent pricing.
UsePod and Surplus are actually complementary. In fact, Surplus has already been integrated as a provider into UsePod, allowing for cross-platform liquidity.
Until next time...
Considering the rapid adoption of AI in both our personal and business lives, AI inference is poised to become one of the largest commodity markets in history. The question is, will it be priced in fiat currencies or cryptocurrencies?
Given the unsustainable nature of our traditional debt-based financial system, and AI agents' preference for non-KYC'd instant settlement on blockchains, cryptocurrencies will more than likely become the currency of choice for ZHCs in the agentic economy.
Projects like UsePod and Surplus Intelligence are laying the groundwork for what could become the world's largest commodity exchanges in the future so it may be worth reading their whitepapers/docs and following them on X to stay on top of developments in this exciting new sector.
If you learned something new from this article, be sure to check out my other posts on crypto and finance here on the Hive blockchain. You can also follow me on X or InLeo for more frequent updates.