Innovation & Wealth › Platforms, mobile & AI
The AI boom: from deep learning to ChatGPT
2012 (deep learning breakthrough) – today· Innovation· Innovations & industries
Well documentedSolid records, texts or physical evidence.
From 2012, 'deep learning' on powerful chips made computers far better at vision and language, culminating in chatbots like ChatGPT.
See it on the Innovation & Wealth map →What happened
In 2012 Geoffrey Hinton's Toronto team won an image-recognition contest with a neural network trained on Nvidia graphics chips. In 2016 Google DeepMind's AlphaGo beat Korean Go champion Lee Sedol (이세돌). Google's 2017 'transformer' design let models learn from vast amounts of text, and ChatGPT (2022) brought them to the public. Tech giants now spend hundreds of billions of dollars a year on AI computing. Chipmakers, cloud firms and shareholders gained first; artists, writers and publishers sued over use of their work.
Impact on society and the economy
It is the biggest technology bet since the internet, reshaping stock markets, energy demand, work and geopolitics; Hinton and DeepMind's Demis Hassabis won 2024 Nobel Prizes.
How we know
Research papers and benchmark results, company filings on spending and revenue, user surveys, and lawsuits such as The New York Times v. OpenAI (2023).
Still debated
Whether the spending is a bubble, whether current methods can reach human-level intelligence, and whether AI's risks, from misinformation to loss of control, are overstated or underrated.
Inside this (4)
Connections (7)
▸Nvidia (Jensen Huang) supplies the chips for The AI boom: from deep learning to ChatGPTWell documented
The connection
Training and running AI models takes trillions of simple calculations in parallel, which Nvidia's graphics chips do well. Its CUDA software (2006) made them programmable, and researchers adopted them after AlexNet (2012). Since ChatGPT its data-center chips, costing tens of thousands of dollars each, became the scarce input; it holds an estimated 80–90% of the AI chip market.
When and how
Santa Clara, California; Toronto in 2012; a demand surge from late 2022 led by Microsoft, Meta, Google, Amazon and OpenAI.
How we know
Nvidia's data-center revenue rose from about $15 billion in fiscal 2023 to $115 billion in fiscal 2025, and research papers name the hardware used.
Debates
Whether rival chips from Google, AMD, Amazon and Chinese designers will erode Nvidia's lead and its unusually high profit margins.
▸Google (Larry Page and Sergey Brin) published the transformer research behind The AI boom: from deep learning to ChatGPTWell documented
The connection
In 2017 eight Google researchers published 'Attention Is All You Need,' describing the transformer, a design that reads whole passages at once and improves with more data and chips. OpenAI's GPT, short for 'generative pre-trained transformer,' and nearly every later chatbot use it. Yet OpenAI, not Google, launched the first hit chatbot.
When and how
Google Brain in Mountain View, 2017; authors included Ashish Vaswani and Noam Shazeer; all eight eventually left Google, though Shazeer later returned.
How we know
The paper is among the most cited in computer science, with over 100,000 citations, and OpenAI's GPT papers build on it explicitly.
Debates
Why Google, which invented the core method, was slow to ship a chatbot: caution about errors, or fear of undercutting its search-ad business.
▸OpenAI and ChatGPT made mainstream The AI boom: from deep learning to ChatGPTWell documented
The connection
ChatGPT, released free on November 30, 2022, let anyone talk with a large language model in plain English. It reached an estimated 100 million users in two months, then the fastest growth of any consumer app. Its success sent Microsoft, Google, Meta and others racing to build rivals, and set off the surge of spending on chips and data centers.
When and how
San Francisco, November 2022; Microsoft invested about $10 billion more in January 2023 and built the technology into Bing and Office.
How we know
User estimates from analysts such as UBS and OpenAI's own figures; tech companies' earnings calls cite ChatGPT as the turning point.
Debates
Whether OpenAI can keep its lead as rivals catch up, and whether its business can ever cover its enormous computing costs.
▸The AI boom: from deep learning to ChatGPT sparked AI and the future-of-work debateWell documented
The connection
Earlier automation mostly hit factory and routine office work. Chatbots that write, code and summarize put skilled white-collar jobs in question on a new scale: a 2023 OpenAI and University of Pennsylvania study estimated about 19% of US workers have at least half their tasks exposed. That set off forecasts, union contracts and policy fights over who gets the gains.
When and how
Mainly the US and Europe from 2023; studies by OpenAI, Goldman Sachs and the IMF; Hollywood writers' 148-day strike won limits on AI use.
How we know
The surge of studies, contracts and hearings after ChatGPT is well documented; actual job effects so far are small and contested.
Debates
Whether exposure means jobs lost or jobs changed; early evidence is mixed, with some signs of fewer entry-level hires in exposed fields.
▸The AI boom: from deep learning to ChatGPT drives AI data center building boomWell documented
The connection
Training a leading model can take tens of thousands of chips running for months, and serving hundreds of millions of users takes even more. That demand turned data centers from routine IT buildings into giant, power-hungry campuses. The biggest tech firms' capital spending more than doubled from 2023 to 2025, and power, not land, is often the binding constraint.
When and how
Northern Virginia, Texas, Arizona, Ohio, Louisiana and abroad from 2023; builders include Microsoft, Amazon, Google, Meta, Oracle and OpenAI's Stargate partners.
How we know
Company capital-spending reports, utilities' load forecasts and the 2024 Lawrence Berkeley National Laboratory study of data-center power use.
Debates
Whether more efficient models, like China's DeepSeek in early 2025, will cut how much computing AI needs, or whether cheaper AI will simply raise total demand.
▸The AI boom: from deep learning to ChatGPT has swelled many fortunes in The billionaire boom and the inequality debateMostly agreed
The connection
Investors bid up shares of companies seen as AI winners, enriching their founders and early backers. Nvidia's Jensen Huang went from a few billion dollars in 2019 to well over $100 billion; Oracle's Larry Ellison briefly topped the world list in 2025; Zuckerberg, Page and Brin gained too. A handful of AI-linked stocks drove much of the US market's gains.
When and how
Mainly the US from 2023, plus Taiwan's chip industry and China's AI firms.
How sure are we?
Forbes and Bloomberg rankings, which track public share prices; the gains are well documented, though paper wealth moves daily.
Debates
Whether these fortunes are durable or a bubble like 2000, when many dot-com fortunes vanished within two years.
▸John Bogle's index fund gives ordinary savers a large stake in The AI boom: from deep learning to ChatGPTMostly agreed
The connection
An S&P 500 index fund owns companies in proportion to their market value, so the AI boom's winners, Nvidia, Microsoft, Apple, Alphabet, Amazon, Meta and Broadcom, make up a large and growing share of it. By 2025 the ten biggest stocks were about 40% of the index, so ordinary retirement savers automatically share both the gains and the risk.
When and how
US stock markets, 2023–2025, after ChatGPT's launch; tens of millions of Americans hold S&P 500 index funds through 401(k) plans and IRAs.
How sure are we?
S&P publishes index weightings daily, and fund data show passive funds overtook active ones in US stock-fund assets in 2019.
Debates
Is the index now dangerously concentrated in one theme, as at the 2000 dot-com peak? And do huge passive holdings inflate the biggest stocks' prices? Both are debated.