AI Tokens 2.0: From Empty Promises to Real Compute

In January 2026, the crypto market is once again obsessed with artificial intelligence. But if you think we are simply back in 2024, you are mistaken. The rules have changed. Naive faith in "agents that will change the world" has been replaced by something far more cynical - and far more pragmatic.

While some traders are still burying portfolios full of last cycle's AI tokens, others are actively pouring liquidity into projects that openly say: "We are here for the fees." And in the background, an infrastructure heavyweight is quietly gaining momentum, promising to reinvent Bitcoin-style mining for the needs of neural networks.

Let's take a closer look at what is actually happening in the AI token sector right now.

Why the First Wave Faded Away (The AI16Z Lesson)

To understand the current trend, we need to remember why the previous one collapsed. In late 2024 and early 2025, the market was dominated by the "autonomous agents" narrative. The face of that era was AI16Z, a project that promised to create a DAO governed by an AI version of Marc Andreessen.

So what went wrong?

1. The "black box" problem. Investors bought tokens believing AI agents would be innovative money-making machines. In reality, most of these "agents" were little more than chatbots with wallets attached, burning more money on fees than they ever earned.
2. No cash flow. Token prices rose purely on the belief that the next buyer would pay more. Holders had no rights to intellectual property, no share in a real business, and often no business at all - the "agent's activity" frequently amounted to posting on X.
3. A collapse of expectations. It quickly became clear that AI agents were nowhere near ready to autonomously manage hedge funds. The bubble burst. AI16Z lost about 99% from its peak not because the technology was bad, but because expectations were completely detached from reality.

The market learned its lesson: promises of future profits are no longer enough. What is needed is a mechanism that generates revenue here and now. And such a mechanism has emerged.

The "Hype Tax": A New Monetization Model on Solana

On Solana, a new trend has formed that fundamentally changes the relationship between token creators and traders. It is usually described as creator fees.

How Creator Fees Work

In the old model, token creators made money by keeping part of the supply and selling it at the top - essentially dumping on their audience. The new model, implemented through platforms like Bags.fm and specific smart contract configurations (such as Solana's Token-2022 extensions), works very differently.

Now, the developer earns a percentage from every token transfer.

For example:

• creator fee: 1%
• you sell 1,000 tokens
• the token's smart contract automatically:
  • sends 990 tokens to the counterparty
  • withholds 10 tokens (1%) as a fee

That fee is sent to an address defined in the token's settings, usually controlled by the creator. Token trading is effectively turned into something akin to the art market, where artists receive royalties every time their work is resold.

The creator no longer needs the token price to "go to the moon". What matters is continued activity: trading, transfers, hype. High attention equals high income.

The New Wave Stars

1. Gas Town The most vivid example. Gas Town is a platform for orchestrating AI agents, but it became famous primarily because of its tokenomics. The token trades exclusively on DEXes, meaning every buy and sell is an on-chain transfer. As a result, the token creator earns fees on every trade. This is framed as an honest way to fund open-source development: "You speculate, I write code using your fees." It gives trading a new narrative and attracts participants. At its peak, daily trading volume exceeded $16.8 million.
2. RALPH A token monetizing the so-called "Ralph Wiggum Loop" - a methodology that forces AI coding agents to keep iterating until a task is completed. The technology is real, and the token generates real income for its developers. This creates a precedent for what could be called "trading-driven patronage".
3. PsyopAnime An example of the same model applied to AI-generated content. The project produces anime using AI. Buying the token - even on the secondary market - funds the creation of new episodes, while price growth attracts attention. Elon Musk even followed the project's X account, triggering a massive spike in trading volume (over $22 million per day) and, consequently, substantial payouts to the developers.
4. SOL Tomato A meme token built around "vibe coding" culture. Its creators launched a Claude-based AI agent with exactly one task: autonomously take care of a real tomato growing inside a biodome. It controls lighting, temperature, and watering without human intervention, while the community watches. This is a pure hype play, with no complex technology involved - and yet the model works.

This approach is more sustainable than the classic pyramid schemes of the past. It is transparent: you know you are paying a "tax" to the creator. At the same time, it turns crypto trading into something resembling a casino, where the house (the developer) always wins, and players just push chips around hoping to come out ahead.

Gonka AI: "Bitcoin for AI" or Just Another Pyramid?

Against the backdrop of fee-based meme tokens, Gonka AI and its token GNK stand apart. This is a completely different kind of project. Gonka AI positions itself as a Layer-1 blockchain for AI computation.

What Is the Core Idea?

Gonka AI uses a concept it calls Proof-of-Work 2.0, or transformer-based PoW. The idea is similar to Bitcoin's at a high level, but with a crucial difference. In Bitcoin, miners burn electricity hashing meaningless numbers. In Gonka, miners (GPU owners) perform real work: servicing neural network inference and training requests. Those computations themselves act as proof of work that secures the network.

Gonka aims to aggregate GPUs from around the world into a single network and sell their compute power for AI workloads. In essence, it is a decentralized AI infrastructure where GPU owners rent out resources and model developers pay for them using GNK tokens. The team claims this can be cheaper than AWS or Google Cloud, while also preserving privacy.

Gonka can be compared to Render, Cocoon, and Qubic, but there are important differences:

• Render was originally built for graphics rendering and only later adapted to AI workloads. Gonka, by contrast, was designed from scratch for machine learning.
• Cocoon rewards its nodes in TON, but it has never clearly stated that users will also pay for compute in TON (and I suspect payments there will ultimately be denominated in something more stable).
• Qubic allows CPU mining, while Gonka requires GPUs - and not consumer-grade GeForce cards, but hardware on the level of H100 and above. This makes the barrier to entry significantly higher than in Qubic.

Despite these differences, Gonka shares one important trait with Qubic.

Qubic configured its system for parallel mining with another cryptocurrency (Monero) to prove that mining a new AI-focused coin could be more profitable. And indeed, miners massively switched to the Qubic pool. In August 2025, the pool reached 52.7% of Monero's hashrate, carried out several block reorganizations, and declared this proof that "useful mining" would inevitably replace "useless mining".

The Monero network has not fully recovered from that shock. Kraken temporarily disabled XMR deposits and later required 720 confirmations. At Rabbit.io, we used to process XMR swaps after 10-12 confirmations. Now we also have to wait longer - not 720 confirmations like Kraken did at one point, but still 20-25, meaning swaps can take 40-50 minutes or more.

Unlike Qubic, Gonka presents itself as a competitor to Google and Amazon, not to PoW blockchains. But since Gonka requires GPUs - hardware far more common among miners than CPUs - it may become a "black hole" for compute power, pulling resources away from other networks.

What Qubic did to Monero deliberately, Gonka could do to other GPU-mined networks simply through market efficiency. Using GPUs solely to secure blockchains like Ethereum Classic, Ravencoin, Ergo, or Alephium may become economically irrational if those same GPUs can earn more solving AI tasks for paying customers.

Who Buys GNK, and Why?

1. Compute consumers. In theory, these should become the main buyers. Developers who need cheap AI compute may turn to Gonka instead of AWS or Google Cloud.
2. Future compute consumers. Today, Gonka is still scaling and cannot offer truly massive capacity. GNK is cheap, and developers who plan to launch AI products in the future may buy now with the intention of using it later. This is risky, but the potential cost savings could be substantial.
3. Miners. They need GNK as collateral. Without staking collateral, a miner receives only about 20% of the rewards for providing compute. The collateral functions much like in Proof-of-Stake systems, guaranteeing honest behavior and uptime. It is partially slashed if a miner refuses tasks or performs them poorly.
4. Speculators. They see Gonka as "Bitcoin for the AI era" and buy GNK to resell later to end users or other speculators at a higher price.

Is There a Catch?

Gonka looks like a legitimate infrastructure project. Large investors take it seriously:

• In 2023, Gonka AI raised funds from Coatue Management (an OpenAI investor) and Slow Ventures (an early Solana investor).
• In 2025, Kazakhtelecom, Kazakhstan's largest telecom company, joined the network.
• In December 2025, Bitfury joined as an investor. Bitfury previously produced Bitcoin ASIC miners and now owns Axelera AI, a neural network chip manufacturer.

Notably, Bitfury did not just invest in the business - it bought GNK tokens at $0.60 each. Today, the token trades at around $1.90.

However, the network currently has no paying customers. Its entire economy rests on mining and expectations. If, in the future, Gonka attracts many miners but few real clients who need AI compute, the GNK price will likely collapse as miners dump rewards on the market.

For now, the price is driven by the belief that decentralized AI will outperform corporations burdened with expensive data centers and heavy overhead. But we saw a similar story last year, and that bubble burst when DeepSeek showed that cutting-edge AI results could be achieved at far lower costs than previously assumed.

No one knows whether another breakthrough is waiting around the corner, one that could render Gonka's business model uncompetitive.

Conclusion

The previous wave of AI tokens collapsed under the weight of speculation and the absence of real products. The new wave offers more concrete mechanisms and, in some cases, real technology - but speculation remains the primary source of demand.

There are tokens genuinely connected to technical innovation, alongside novel monetization schemes. The risks remain high.

Projects like Gonka promise a breakthrough, but for now they are still experiments riding a powerful narrative rather than proven, sustainable businesses. Gonka is clearly aiming to build a full-fledged crypto economy around GPU rental. Whether that effort will succeed is still an open question.