Is TAO Just Exit Liquidity? Shay Settles the Argument

In this video, Shay of Deciphering Crypto tackles one of the most uncomfortable questions hanging over Bittensor right now: is the entire dTAO subnet system just a clever way for insiders to dump on retail? He walks through the argument honestly, lands on a nuanced verdict, and gives viewers one concrete metric to use as their own filter going forward.

The question that set everything off

• On April 10, 2026, Bittensor lost ~$900 million in market cap in two hours.
• The trigger: Sam Dare of Covenant AI, one of the most respected builders on the network, announced he was leaving. But before the public announcement, he liquidated around 37,000 TAO worth of subnet alpha tokens across Templar, Grail, and Basilica.
• Co-founder Jacob Steeves pushed back on Dare’s “decentralization theater” accusation but admitted he had personally sold subnet alpha tokens, saying he reserved the right to buy and sell: “that’s how dTAO works.”
• That admission is what gave exit-liquidity critics their receipt.

The mechanics critics are pointing at

Shay walks through why dTAO is structured the way it is, because the design itself is what makes the accusation plausible:

• Root staking (Path 1): stake TAO into Subnet 0, earn TAO. Like an index fund.
• Subnet staking (Path 2): you’re not really staking, you’re swapping. You hand over TAO and receive a subnet’s alpha token.
• The kicker: the APY for root staking is around 18%, while subnet staking is around 35-100%, meaning root staking carries reduced influence inside Bittensor economics. The system mechanically nudges holders to swap TAO for alpha tokens.

The bull read: the design rewards conviction in specific subnets. The bear read: it forces TAO holders into riskier tokens issued by anonymous teams with thin liquidity and no product.

Shay’s answer: both are right, depending on which subnet you’re looking at.

The counter-evidence the bears ignore

• ~$43M in estimated Bittensor subnet revenue in Q1 2026 (though Shay notes this figure is debated.
• Chutes (Subnet 64) runs serverless AI inference with real paying users.
• Covenant 72B, the 72-billion parameter open-source model, was a verifiable engineering milestone even after the team’s messy exit.
• And the network is responding: starting conviction locks go live soon, forcing subnet owners to lock tokens with publicly visible unlock dates. If a founder is heading for the exit, the market can reprice before the dump happens.

The one metric Shay tells viewers to actually use

This is the most actionable part of the video: Net TAO Flow.

• Net TAO flow = TAO staked in minus TAO unstaked out, per unit of liquidity.
• Positive net flow = real capital is entering. People believe the work.
• Negative net flow = the market is walking away regardless of what Twitter says.
• You can check it free on taostats.io.
• Shay’s opinion: it’s not perfect, but it’s “100 times better than judging by Discord hype.”

Read more about TAO flow:

Shay’s verdict

• For weak subnets: no revenue, anonymous teams, dead GitHub, just Twitter noise. Yes, the alpha tokens absolutely function as exit liquidity.
• He cites a March 2026 Philip Maymin study showing the larger top-tercile subnets lost around 50% annualized, while smaller ones swung wildly in both directions. The “safer-looking” subnets weren’t the better bets.
• For Bittensor as a whole, no, exit liquidity doesn’t tell the full story. The network is producing real AI infrastructure, expanding toward 256 subnet slots, gaining regulated exposure products via Grayscale, and patching its incentive holes with conviction locks.
• Practical takeaway:
  • Want Bittensor exposure without the roulette? Root stake.
  • Want to pick a specific subnet? Check net TAO flow, check GitHub activity, check whether anyone is actually paying for the product. If two of three are missing, walk away.

Bottom line

Shay’s closing line: dTAO is both a real market-discovery system and a place where insiders have at times used the design to extract value. Both can be true at once.

Full video here: