NEW YORK, July 8 — The economics of artificial intelligence have entered a new, more expensive phase, and the numbers coming out of the industry are starting to look like something from a blockbuster movie budget. Training a leading large language model now costs hundreds of millions of dollars, a staggering leap from just a few years ago, and companies are betting billions that the payoff will be worth it. Think of it like building a skyscraper.
The first few floors went up for a relative pittance, but now that builders are reaching for the sky, every new level demands exponentially more steel, concrete, and engineering. OpenAI’s GPT-3, launched in 2020, cost about $4.6 million to train.
By 2023, its successor, GPT-4, carried an estimated price tag of between $100 million and $200 million. That is a more than twenty-fold increase in just three years. The costs don’t stop once the model is built.
Running these systems — a process called inference — adds a significant ongoing expense. For GPT-3, OpenAI priced inference at roughly $0.02 for every 1,000 tokens as of 2023.
That per-query cost adds up fast when millions of users are generating responses. The energy bill alone is eye-opening: researchers estimate that a single query on GPT-3 consumes about 1,000 times more energy than a standard Google search.
Big Bets, Bigger Markets
Despite the eye-watering costs, the market is responding with enthusiasm that borders on frenzy. The global AI market was valued at $136.6 billion in 2022, according to Grand View Research, and is projected to surge to $1.8 trillion by 2030. The large language model segment, a smaller but critical piece of that pie, was estimated at $1.5 billion in 2023 by MarketsandMarkets, with projections of hitting $13.4 billion by 2028.
Revenue is starting to flow. OpenAI’s annualized revenue hit $1.6 billion in December 2023, according to The Information.
That is real money, though it remains to be seen whether it covers the enormous training and operational costs. Microsoft has placed a $13 billion bet on OpenAI, the largest single corporate investment in the space. The hardware side of the equation is even more dramatic.
NVIDIA, the chipmaker whose graphics processing units have become the backbone of AI training, reported data center revenue of $47.5 billion in fiscal 2024. That figure was up 217% year-over-year, a growth rate that reflects the insatiable demand for computing power.
Google, meanwhile, spent $32.3 billion on capital expenditures in 2023, a significant portion of which went toward AI infrastructure.
Jobs, GDP, and the Human Factor
The practical upshot for ordinary workers and the broader economy is still being calculated. Goldman Sachs issued a sobering estimate in March 2023: as many as 300 million jobs globally could be affected by AI automation. That is a staggering figure, but the bank also projected that AI could boost global GDP by 7% over the next decade.
The net effect, in other words, could be a massive reshuffling of the labor market, with some roles disappearing and new ones emerging. Investment in AI startups reached $42.5 billion in 2023, according to CB Insights, a sign that venture capital sees long-term potential despite the near-term costs.
The money is flowing into everything from foundational model companies to specialized applications in healthcare, finance, and logistics. Here’s what it means in plain terms: the AI industry is currently in a capital-intensive build-out phase, not unlike the early days of the internet or the construction of the national railroad system. The upfront costs are enormous, the energy demands are significant, and the impact on jobs is real.
But the projected market growth — from $136.6 billion to $1.8 trillion — suggests that investors and corporations believe the long-term returns will justify the spending. The question that remains is whether the revenue from AI products and services will eventually catch up to the investment.
OpenAI’s $1.6 billion in annualized revenue is a promising start, but it is still a fraction of the $13 billion Microsoft has poured in. NVIDIA’s explosive growth shows that the infrastructure builders are making money hand over fist, but the profitability of the AI applications themselves is less certain. What to watch next: the energy costs.
If AI inference continues to consume 1,000 times more energy than a web search, the environmental and operational expenses could become a limiting factor. Companies are already exploring more efficient chips and algorithms, but the economics of scale will ultimately determine whether AI becomes a ubiquitous utility or a premium service for those who can afford it.
{“@context”: “https://schema.org”, “@type”: “NewsArticle”, “headline”: “Economics of LLMs: Training Costs, Market Size, Revenue Projections”, “description”: “NEW YORK, July 8 — The economics of artificial intelligence have entered a new, more expensive phase, and the numbers coming out of the industry…”, “datePublished”: “2026-07-08T09:30:00+00:00”, “author”: {“@type”: “Person”, “name”: “Erika Johnson”}, “publisher”: {“@type”: “Organization”, “name”: “InfoPulse Today”, “url”: “https://newsfeed.infopulsetoday.com/”}}
