1/ Thread on what I think it’s one of the most under-the-radar projects in the market: $FLOCK 🧵 >> @flock_io became the first & only decentralized AI integrated by @Alibaba_Qwen, Alibaba’s LLM (the GPT to OpenAI, the Claude to Anthropic). This is not an empty partnership, like Flock simply using Alibaba Cloud or integrating Qwen models. Instead, Qwen is building models and training on Flock’s infrastructure for decentralized privacy-first AI. Some alpha: >> Qwen’s 2.5 on-edge version is being trained with FLock, and will sell directly to industry use cases, like hospitals, financial institutions, and other sectors where data privacy is of outmost importance. In their own words: "This marks the first deployment of a decentralized AI agent model within a major centralized AI ecosystem".

2/ That’s already big enough, but there’s more. @base MCP replaced the onchain part of Anthropic’s Claude with FLock’s Web3 Foundation Model. "Claude was expected to be the default execution engine. However, in March, Base ultimately chose FLock as its default model for onchain agent execution, after FLock surpassed Claude on various Web3 benchmarks."

3/ But what is FLock? FLock is a full-stack infrastructure for privacy-first AI. They pioneered Federated Learning (FL) at scale — a decentralized training method where models learn collaboratively without sharing raw data. Sectors to disrupt are those where privacy is vital (eg. healthcare, government, banking, law firms, etc). FLock’s core thesis: public data has already been scrapped. The next leap in AI will come from private data. The protocol covers the full stack, from data prep to model deployment, across three layers: + AI Arena = like Bittensor, for open model training & competition + FL Alliance = like Nous Research, for fine-tuning + Moonbase = like Virtuals, for agents and AI models @Defi0xJeff did a fantastic writeup on FLock, covering it extensively. Check it out here:

5/ Numbers also speak by themselves • 7,169 AI models created • 186 trading nodes • 249 validators • 1,263 validators • 613,692 validation submissions @MessariCrypto covered Flock in detail here (numbers have increased since then showing consistent growth):

8/ They raised $9M from @DCGco and @OKX_Ventures, who doubled down in Dec’24 (always a good sign), and @GSR_io (also their Market Maker). As you can see, these investors aren’t simply writing checks. GSR is training decentralized models with trading data, and DCG supports the subnet and is an official validator. Another Flock validator that stands out is A41, Korea’s largest validator by total staked assets. They also received a grant from the Ethereum Foundation.

10/ VCs invested at $150M and 300M FDV and are locked for 12m cliff + 24m vesting, as well as team and advisors. No concerns about selling pressure from these. Circulating supply will increase considerably via inflation to incentivize participation in the ecosystem. Something to take into account. Fortunately, the subnet earns $TAO rewards that I hope they use to avoid $FLOCK inflation.

12/ The team recently hinted to be cooking something with the McLaren F1 team. That’s the level they’re playing. Not everyone enters those facilities. Wouldn’t surprise me, given the competitive nature of Formula 1, to see collaborations of this sort for privacy-first AI models.

13/ Others: • Flock is also highlighted in The top 100 of the most promising AI companies by @CBinsights, alongside titans like @Figure_robot, @LangChainAI, and other web2 startups • The team authored an award-winning paper on the application of federated learning to healthcare. (paper: • They have tons of other academic papers and citations:

14/ That’s my overview for now. Thesis is simple: • First ever integration of a decentralized AI system by a web2 LLM — Alibaba’s Qwen, directly using Flock to train and build models • Integrated in Base MCP for onchain execution • Raised $9M by DCG, OKX, GSR • Numbers that signal high levels of adoption, with players like GSR, Animoca, or Dimo training models • Solid underlying technology (federated learning) with a good narrative and vision (privacy-first AI), and a full-stack co-creating platform with value accrual to the token • Potential partnership with McLaren • Bittensor subnet backed by Yuma Group and DCG • Only $20M market cap with an ever-increasing staking / locking ratio • Listed in tier 1 exchanges like Bybit, Upbit, and Bithumb, with a great penetration in the korean market