Internal Leak! The "Emperor's New Clothes" Behind TAO's $5.8 Billion Valuation: After Subsidies Fade, Who Will Pay for This AI Computing Power Myth?

Brothers, today we’re not talking about candlestick charts or macroeconomics. Let’s discuss something that has recently been elevated to the status of a deity—Bittensor, also known as TAO. The current price hovers around $275, with a market cap reaching $2.6 billion. If we look at fully diluted valuation, it’s even more shocking—$5.8 billion! Grayscale has endorsed it, and Jensen Huang has publicly praised it. There are 21 million tokens in total, and it even mimics Bitcoin’s halving schedule. Sounds pretty sexy, right? It just finished its last halving at the end of last year, reducing daily issuance from 7,200 to 3,600. The number of subnets has quadrupled in a year—from 32 to 128. The Templar team has used decentralized compute to run a competitive 72B large model. All these achievements are impressive, I admit.

But here’s the problem.

Is this thing really worth so much? As crypto traders, we ultimately care whether it can make money, whether it can generate real demand and revenue. It can’t just rely on hype and faith forever, right? Today, I’ll strip down the shiny exterior and see what’s really underneath.

First, you need to understand how the TAO network operates. Simply put, there are four types of participants: market makers (subnet owners), workers (miners), evaluators (validators), and stakers (those who put up collateral). Every day, 3,600 new TAO tokens are minted and distributed among them. Miners and validators get the largest share—each 41%. Subnet owners also take home 18%. The top ten subnets control over half of the network’s output. TAO is the network’s native currency—everything requires it. The more active the subnets, the greater the demand for TAO.

Sounds logical, right?

But here’s a critical flaw: on the supply side, all data—how many tokens are issued daily, how they are split, halving rules, staking ratios (around 70%)—are all transparent and on-chain. But on the demand side? It’s a complete black box! All AI services—such as inference and computation—happen off-chain. Nothing is recorded on the blockchain. You have no idea which subnets are truly profitable or how much they’re earning. You can only guess by looking at who staked more TAO or whose token prices are higher, or by listening to project hype. This black box isn’t temporary; it’s inherent to the structure! The blockchain only records token transfers, not API calls. There’s no solution.

So let’s look at what demand-side signals we can piece together.

The most prominent example is Chutes (SN64). It consumes 14.4% of all TAO issuance, dominating the network. What does it do? Provides open-source model inference services. Its pricing is sky-high—85% cheaper than AWS, and cheaper than Together AI. The data is astonishing: 400,000 users, over 5 million requests daily, processing 9.1 trillion tokens. It’s a top provider on OpenRouter, with some models outperforming centralized solutions.

Doesn’t that sound invincible?

But its low cost isn’t because of superior decentralization or efficiency. It’s simply subsidized by TAO inflation! Based on its market share, it can claim about 518 TAO daily, worth roughly $5.2 million annually. But how much real revenue does it generate from external users? The team estimates just $1.3 million to $2.4 million per year—without audits. Do the math: for every dollar paid by users, the network needs to print 22 to 40 TAO to cover the subsidy! This subsidy ratio is crazier than the most reckless internet burn campaigns.

Remove this astronomical subsidy, and based on its daily processing volume of 101 billion tokens, its true cost is about $1.41 per million tokens. How does that compare to centralized markets? Together.ai’s LLaMA 3.3 70B Turbo costs around $0.88; DeepSeek V3 costs between $0.4 and $0.8; some smaller models can go as low as $0.18. See the difference? Without subsidies, Chutes would be 1.6 to 3.5 times more expensive than centralized options! The so-called 85% cost advantage is actually a cost disadvantage. Its low price is entirely paid for by TAO holders through inflation, subsidizing users.

What happens after the next halving (expected late next year or early the following year)? TAO issuance will be cut in half again. What then? Service prices might double, driving users away; miners might feel rewards are insufficient and stop mining; or the subsidy gap could grow even larger, making things uglier. Some say early internet companies also burned money to gain market share. But Uber and AWS burned money to build moats, user habits, and ecosystems. What about subnets on Bittensor? All models are open-source, interfaces are standard, users can switch instantly at zero cost if prices rise. Once subsidies stop, users will scatter immediately—no loyalty, no retention.

Even more concerning, Rayon Labs, which operates Chutes, also controls two other subnets. Together, they could take nearly a quarter of all TAO issuance. And these two don’t report any external revenue. Essentially, a small group controls the entire network’s money printer.

What about other subnets? Some are more viable. For example, Targon (SN4), which provides enterprise GPU compute for confidential workloads, has a projected annual revenue of about $10.4 million, a valuation of $48 million, and a market-to-sales ratio of 4.6—quite healthy in crypto. But remember, this is forecasted revenue, not confirmed financials. Then there’s Templar (SN3), which produced the 72B model. Its market cap is $98 million, but external revenue is zero—product still in development. The remaining 120+ subnets mostly rely on subsidies, with little or no revenue, or are still in early stages.

Adding up all confirmed external income, the total is at most $3 million to $15 million annually. Ironically, Chutes alone receives about $52 million worth of subsidies per year—more than the entire network’s external revenue. Dividing the $2.6 billion market cap by this revenue yields a P/E ratio of 175 to 200; using the fully diluted valuation of $5.8 billion, it’s nearly 400! Compare that to centralized AI infrastructure companies, which are valued at 15–25 times forward revenue, or high-growth SaaS firms maintaining over 50x. TAO’s valuation multiple is 4 to 10 times that of aggressive growth stocks.

What does this huge gap tell us?

It indicates that the market’s current valuation of TAO isn’t based on its real value creation. It’s purely speculative—driven by scarcity stories (halving), institutional endorsements (Grayscale ETF), AI hype, and the dream of decentralized AI. These factors can push prices up, but it’s crucial to realize that the idea of “Bittensor as a profitable AI service network” is a complete disconnect from reality.

And even if you believe it can succeed, the road is treacherous.

Going upward, there’s the mountain of self-hosting. All models are open-source, weights are public. With a single H100 GPU, running a 70B model costs only $40–$50 a day. Deployment tools are becoming easier, and institutions can do it cheaper themselves. Next-gen chips will further cut inference costs.

Going downward, giants like Microsoft, Google, Amazon, and Meta will spend over $200 billion on AI in 2025 alone. TAO’s annual incentive budget? About $36 million—less than a week’s worth of AI infrastructure spending by Microsoft. These giants have hardware priority, dedicated data centers, and cash flow from other businesses to subsidize AI. They can lock in enterprise clients and use their ecosystem to maintain dominance. Professional service providers are also subsidizing open-source models with VC funds, engaging in price wars. Meanwhile, Bittensor’s subnets are caught in this narrow squeeze—costs can’t go down, prices can’t go up, and they bear additional friction costs like token conversion losses, validator expenses, and network latency.

Most critically, there’s no moat. All technology is open-source, papers are public. Today’s good subnet can be copied tomorrow—no proprietary tech, no network effects, no switching costs, no brand loyalty. The community claims that incentives are the moat, but that moat relies on continuous money printing. With each halving, the water level drops, until it dries up.

So, what are you really trading when holding TAO?

At a $2.6 billion valuation, you’re not trading its tiny few million dollars in actual revenue. You’re trading Bitcoin-like scarcity narratives, the potential ETF windfall from Grayscale, the hype-driven AI sector, and a long-term bet on decentralized AI’s future.

If you’re in for these stories, that’s fine—speculation is part of the game. But if you truly believe the hype—that Bittensor is now a profitable, revenue-backed AI service network capable of challenging centralized giants…

Brother, I suggest you wake up.

Look at the data, the terrifying gap between subsidies and revenue, and the survival environment squeezed from all sides. This dream looks way too expensive right now.

(Deep night reflection—more clarity, more shock. The market always rewards the best storytellers, but in the end, the ones who pay the bill are those who believe the stories.)