> For the complete documentation index, see [llms.txt](https://docs.thetpa.ai/deai-tokenized-processing-architecture/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.thetpa.ai/deai-tokenized-processing-architecture/the-problem.md).

# The problem: paying for AI today

AI is sold per token, through separate accounts and card-based subscriptions at each provider. That model fits people with cards and one or two tools. It fits poorly for everyone else, for software agents, and for the developers who build new models.

* **Fragmented billing.** Using several models means several accounts, several balances and several invoices.
* **Card dependence.** Payment runs through Visa or Mastercard. About 21% of adults worldwide have no account at a bank or mobile-money provider ([World Bank Global Findex 2025](https://www.worldbank.org/en/publication/globalfindex): 79% of adults have one) **\[Verified source]**.
* **Prepaid packages.** Flat subscriptions charge the same whether a customer uses a little or a lot, so light users pay for capacity they never use.
* **Agents cannot hold cards.** AI agents that call other models, tools and agents need a way to pay per call, with no human checkout.
* **Model builders cannot own what they publish.** A developer who fine-tunes a model rents an endpoint from a platform. The endpoint is not theirs to keep or sell, and most hosting platforms pay nothing back when others use it. Hosting a fine-tuned model on its own also tends to need a dedicated GPU, which makes small specialist models expensive to offer ([S-LoRA](https://arxiv.org/abs/2311.03285) describes this cost).

## Blockchains cannot run AI, and GPU marketplaces only match

Crypto offers digital payment, but it does not solve the rest of the problem **\[Analysis]**:

* **Blockchains record transactions; they do not compute.** Ethereum, Solana and similar chains were built to agree on a ledger. Every node repeats every step and must reach exactly the same result. An AI model needs billions of GPU calculations per answer, and two GPUs do not return bit-identical results, so these chains cannot run AI. AI that "uses" them runs somewhere else, with no link between the payment and the work.
* **GPU marketplaces only connect buyers and sellers.** Decentralized compute networks list GPUs and match them to jobs. They do not give a model builder an owned endpoint, a price in US dollars, or a payment that is split automatically between the people who built, served and secured the work.
* **Nobody proves the work.** When a payment and an AI job live in different systems, the buyer has to trust the provider's word that the model they paid for is the model that ran.

What is missing is a chain where the AI endpoint, the GPU power behind it, the price, the payment and the proof are one system: a compute layer with a blockchain layer on top that controls, proves and pays for it. That is what DeAI TPA and TPAI Chain set out to build.

## Market context

AI is expected to become one of the largest economic categories. PwC estimated in 2017 that AI could add up to **$15.7 trillion** to world GDP by 2030, about 14% more than without it ([PwC, *Sizing the prize*, 2017 estimate](https://www.pwc.com/gx/en/issues/data-and-analytics/publications/artificial-intelligence-study.html)) **\[Verified source]**. A more recent study by McKinsey estimates that generative AI alone could add **$2.6–4.4 trillion a year** to the global economy ([McKinsey, *The economic potential of generative AI*, June 2023](https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/the-economic-potential-of-generative-ai-the-next-productivity-frontier)) **\[Verified source]**.

**The starting market is much smaller.** Payments made by AI agents are still small but growing. Keyrock counted about $73 million settled by agents across 176 million on-chain transactions in the year to April 2026, mostly in stablecoins ([Cointelegraph](https://regional-front.cointelegraph.com/news/ai-agents-using-stablecoins-as-default-settlement-layer-keyrock)) **\[Verified source]**. That is the realistic beginning that TPAs target: software paying for AI per call, in digital money. DeAI TPA's thesis is that a growing share of AI usage will be paid for this way **\[Hypothesis]**. The market is early: this is an opportunity and a risk at the same time.
