How Much Does It Cost to Develop and Train AI? Here's the Current Price, According to the CEO of an $18 Billion AI Startup. There's a sky-high bar to creating AI, and an expert says it will get even more expensive.
By Sherin Shibu Edited by Melissa Malamut
Key Takeaways
- Dario Amodei, the CEO of $18 billion AI startup Anthropic, says it takes about $100 million to train AI at the lower end of the spectrum.
- He says there are models in training today that cost "more like a billion" dollars to train.
- Amodei disclosed these figures in a podcast episode of 'In Good Company' with Norges Bank CEO Nicolai Tangen.
Most startups won't be able to afford to sign up for the AI race, the CEO of $18 billion Anthropic admits.
In a recent "In Good Company" podcast interview with Norges Bank CEO Nicolai Tangen, Anthropic CEO Dario Amodei went into the nuts and bolts of AI development — including how much it takes to get an AI model off the ground.
Amodi said that training an AI model costs around $100 million, but "there are models in training today that are more like a billion."
Related: Model From OpenAI Rival Anthropic Shows 'Metacognition': Report
Amodei estimates that current prices will soon be considered on the lower end compared to what it will cost to develop competitive AI in the next few years.
"I think if we go to $10 or $100 billion, and I think that will happen in 2025, 2026, maybe 2027... then I think there is a good chance that by that time we'll be able to get models that are better than most humans at most things," Amodei stated.
Anthropic CEO Dario Amodei. Photo by Chesnot/Getty Images
The sky-high price of developing AI sets a high bar for startups that want to make their own models to take on OpenAI or Anthropic.
On average, U.S. startups raised about $59 million in fourth-stage Series C funding in Q1 2023 — a far cry from the $450 million Anthropic raised in its Series C in May 2023.
Anthropic has raised over $8 billion to date.
Related: Hugging Face CEO Says More AI Startups Want To Get Acquired
Amodei acknowledged that many companies will not be able to participate in the AI race.
"Of course, I think there's going to be a vibrant downstream ecosystem and there's going to be an ecosystem for small models," he said.
Anthropic's March release of Claude 3 Opus, the latest version of its flagship AI chatbot Claude, sparked a conversation about "metacognition" or AI's potential to monitor its activity and self-correct. An Anthropic engineer revealed at the time that Opus seemed to know when engineers ran a test on it.
Amodei also stated in the interview that Anthropic is working on Nvidia-competing AI chips with Google and Amazon, two companies that have invested up to $2 billion and $4 billion respectively in Anthropic.
The cost of AI could also mean environmental setbacks. A Google report last week revealed its emissions rose by nearly 50% in four years because of AI.
Related: AI Drives Google's Greenhouse Emissions Up By 48% Since 2019
