Jensen Huang said two things, one has already been realized, and no one has noticed the other yet.

First statement: In the All-In Podcast, he praised Bittensor as the modern version of folding@home, a shared computing power model with over 4 million participants, contributing 4.8 million CPU cores and 280k GPUs. TAO surged directly by 90%, reaching over 300. The decentralized AI track has officially been recognized by the mainstream, indicating that Bittensor has the potential to become the largest distributed computing network in human history.
Bittensor proves that the market is willing to buy into the narrative of decentralized AI training, and TAO has opened the valuation ceiling for the entire track.
But Jensen Huang’s second statement, no one has seriously paid attention to it until now: the agent economy scale will reach trillions of dollars. This year at GTC 2026, he repeated this point for the second time.
Recently, everyone has been busy chasing after Binance’s life and working on the clone season RAVE M, no one is asking where this tens of trillions of dollars in agent value is actually running.
First, agents need to perform tasks, invoke tools, and manage funds. They require computing power for reasoning, memory for learning, and on-chain settlement for trading. All three are indispensable.
And I recently discovered that 0G is currently the only platform that has completed all three layers and has its mainnet online:
• Verifiable computing power (TEE-enclosed inference)
• Persistent memory (decentralized storage at 2GB/s)
• On-chain settlement (EVM-compatible L1)
In decentralized training, 0G is also continuously advancing. Last July, they trained a 107-billion-parameter model using the DiLoCoX protocol, which can run on a normal 1Gbps network, reducing communication costs by 95%. They co-developed with China Mobile, passed peer review on arXiv, and are now preparing to publicly retrain and fully open-source.
In the ecosystem, Ghast AI is currently in internal testing. An on-chain native AI assistant that can be used with a browser plugin, with AI memory that can be tokenized into tradable on-chain assets. This is a very cool on-chain version of Openclaw.
On the funding side, Nasdaq-listed company ZeroStack has locked 280k dollars for 21% of 0G’s supply, voting with real money.
Additionally, 0G Labs @0G_labs’ Apollo accelerator has also been launched, offering each team up to 2 million USD + Google Cloud credits + Stanford mentors. Interested parties can also check it out for funding support.
TAO proves that decentralized AI has a market. But Huang’s second statement about a trillion-dollar agent economy might be the real direction worth serious research. @0g_CN