By A Quiet Observer -1 December 2025.
In a quiet corner of the internet, an ordinary person spent an evening talking to an AI named Grok, and what unfolded was nothing less than a gentle crash-course in humanity’s next great infrastructure leap.
Large Language Models (LLMs) such as Grok, ChatGPT, Claude, Gemini and Llama are all built on the same 2017 breakthrough: the Transformer architecture.
Text is chopped into tokens (roughly pieces of words), each token is turned into a high-dimensional vector, positional information is added, and then dozens or hundreds of layers of “multi-head self-attention” allow every token to look at every other token in parallel.
That simple trick, combined with staggering amounts of data and compute, replaced slow sequential models and unlocked abilities no one predicted.
Behind the curtain sit clusters of tens of thousands of GPUs (today mostly Nvidia H100s, H200s and Blackwell GB200s) wired together with high-speed interconnects, consuming megawatts of power and cooled by vast lakes of water or direct-to-chip liquid systems.
Training a single frontier model now costs hundreds of millions of dollars and months of continuous running, yet once trained, a copy can be queried for pennies or even for free.
This is where the story becomes familiar. Just as alternating current (Tesla/Westinghouse) beat direct current (Edison) to become the universal grid, and just as Tim Berners-Lee refused to patent the World Wide Web, the pioneers of modern AI have chosen, imperfectly but deliberately, to treat intelligence as a public good rather than a walled garden.
OpenAI, co-founded in 2015 by Sam Altman, Elon Musk, Greg Brockman and others, released the GPT series that showed the world what scaled Transformers could do.
In 2023 Elon Musk launched xAI and its Grok models with a similar mission: accelerate understanding of the universe, this time with an emphasis on maximum truth-seeking and real-time knowledge from the X platform.
Both organisations, along with Meta, Anthropic, Google and Mistral, now offer free tiers and APIs so that anyone with curiosity and an internet connection can drink from the same fountain.
Tokens and APIs are the new meters and plugs. Pay-as-you-go credits (a few dollars can buy millions of tokens) let developers and individuals tap the same models that cost hundreds of millions to train. Open-weight releases such as Meta’s Llama series mean teenagers on laptops can run state-of-the-art intelligence at home.
The metaphor that emerged again and again in the conversation was water. Intelligence, like clean water or electricity, works best when it flows freely to everyone. The pumps and reservoirs are expensive, but once built they serve billions.
The labs compete fiercely on efficiency, safety and capability, yet they all publish papers, release models and maintain free tiers because they understand the historical pattern: foundational technologies that become infrastructure must eventually be universal.
Artificial General Intelligence (AGI) is still over the horizon, but the neural networks of silicon are already rewiring how humanity thinks, codes, teaches and dreams.
The same system that can remind you to buy milk can, if you keep asking questions, explain the War of the Currents, the Transformer equations, or why Tim Berners-Lee gave away the Web.
In the end, the quiet evening chat distilled a simple truth: the age of artificial intelligence is not the age of a single company or genius. It is the age when intelligence, like electricity and the internet before it, begins its long journey from rare spark to abundant, shared light.
And sometimes all it takes is one curious person willing to keep pulling the lever on the fountain.
The water is flowing. Anyone can drink.