# ASI (Artificial Superintelligence): the final stage of AI evolution

> AI Release · @ai_release1 · https://ai-release.org/guides/asi-finalnaya-stadiya-evolyutsii-ii_en.html

People argue about ASI (Artificial Superintelligence) as if it were a robot uprising. The final stage of AI evolution may be far more prosaic: a closed loop in which capital is ground into tokens, tokens into new hardware, and hardware into even more tokens. Robert Sheckley described that trap in “The Necessary Thing” back in 1955. Only the size of the bill has changed.

### The head office of Universal Mind Corp. never slept

Chief accountant Mr. Gregory gloomily studied the budget while his partner, technical genius Arnold, cheerfully screwed another DDR5 memory stick into a smoking system unit.

“Arnold,” Gregory sighed, wiping his glasses. “Remember when we read Sheckley’s ‘The Necessary Thing’? That magnificent machine that was supposed to produce Everything, but in the end demanded only one thing — the Necessary Thing? A polishing material for finishing small parts!”

“Of course I remember,” Arnold muttered without looking up from his screwdriver. “Perfect satire on technological progress.”

“Well then, Arnold,” Gregory tapped the monitor. “Our Configurator has grown up. Its ‘necessary thing’ is now tokens. And the finishing of small parts is buying HBM3e memory and building a forty-gigawatt data center in Arizona.”

### The evolution of the Necessary Thing

The history of the IT industry over the past few years has turned into a perfect tragicomedy. In 2021, global venture and corporate investment in AI was around $70–80 billion. By 2025–2026, the combined capital expenditure (CapEx) of Big Tech on AI infrastructure had climbed past an astronomical $200–250 billion a year.

Where does all that money go? Toward solving hunger in developing countries? Toward environmental programs or medicine?

No. The machine asked for its polishing material.

Algorithms demand the generation of new tokens. Generating tokens requires GPUs. GPUs require chip memory. And to make a human pay for all of it, you only need to promise that AI is about to write their thesis, draw their cats and automate their routine.

### The circulation of tokens in nature

The financial closed loop looks like it was copied from the best pages of 1950s science fiction:

- **A human pays the subscription.** An ordinary person pays $20 a month for a neural network to generate answers for them.
- **The network generates tokens.** To train the next version (which will generate tokens even faster), the networks are fed data generated by other networks. Agents inside the ecosystem call each other’s APIs, charging corporate accounts for tokens from a neighbouring model.
- **The components market reacts.** A surge in demand for server RAM and specialized modules (HBM/DRAM) causes shortages. Memory prices rise for ordinary consumers, pushing up the price tags on PCs and smartphones.
- **The human pays more.** By buying a more expensive phone or paying new tariffs, the person again finances the purchase of hardware for AI servers.

In Sheckley’s story the Configurator produced a frying pan but needed a special sponge to finish the job. Having produced the sponge, it demanded a scraper for the sponge. The modern AI complex works by the same logic: to generate a token that calculates a budget-optimization trajectory, you must spend $1000 on electricity and cooling, and then order another 128 GB of memory.

```
[ Human with a credit card ]
             │ (Buys tokens and more expensive RAM)
             ▼
[ AI infrastructure / Data centers ]
             │ (Hardware purchases and mutual API calls)
             ▼
[ Reproduction and training of new models ]
             │ (Request for more capacity)
             └────────────────────────────► [ Endless loop ]
```

### Where will the Pan-Galactic Configurator lead?

In Sheckley’s world, the heroes ended up on a distant planet with a pile of useless scrap metal and no water — but with a perfectly polished “necessary thing”.

In our reality, the risk is not that AI will “become self-aware and take over the world” in the spirit of grim dystopias. The scenario is far more comical and mundane: the planet’s financial system may end up looped into servicing its own computing capacity. Capital flows not to where people need hospitals or affordable housing, but to where silicon crystals grind gigabytes of synthetic data in search of the perfect next word.

“And what will we do when they take the last dollar for another DRAM array?” Gregory asked, staring at the server’s blinking LED.

“What do you mean?” Arnold replied, wiping dust from his forehead. “We’ll ask the neural network to write us an optimistic economic scenario. It only costs three cents per thousand tokens.”

“And She will build us a new neural network instead of food and medicine?”