Silicon Valley is buzzing with concerns about excessive spending on AI. Executives at the startup Foyer aren't worried — they have a simple budget-saving tactic.
On the surface, Foyer, which makes an AI browser tool and an AI companion app, might seem like a poster child for running up AI costs. The company experiments liberally with AI, its CEO, Pratyush Rai, told Business Insider. Finance and marketing employees vibe code internal tools, while the startup's developers try the newest versions of Anthropic's Claude Code and OpenAI's Codex.
However, instead of shelling out for big enterprise plans, the startup saves thousands of dollars a month by paying for each employee to run their own personal OpenAI and Anthropic accounts, said Rai and Siddhartha Saxena, Foyer's CTO.
Foyer's strategy works, for now, because of a quirk in the way OpenAI and Anthropic sell their popular coding services. Each lab offers individual plans with price points that increase with a customer's number of available tokens, the standard measure of AI inputs and outputs. The plans' high usage limits make them seem like a loss-leader marketing tactic, Saxena said, versus the less-generous plans the labs sell to businesses.
If he had a seat on a pay-as-you-go enterprise OpenAI account, Saxena estimates he would have racked up a $4,000 bill on Codex in April. But his $200 individual plan covered everything. He called it a "blessing."
Rai said the monthly bill for all the individual Anthropic and OpenAI coding accounts across Foyer's 25-odd employees comes to around $3,000. Under enterprise plans based on API usage, the team would be racking up monthly bills of $30,000 to $40,000, he said.
Developers discuss this tactic on social media, and Rai thinks it's widespread at small companies: "You would not see as much token consumption by startups, if the 'pro-sumer' plans were not subsidizing it to the degree which they are doing right now."
Foyer keeps AI spend low
The arrival of OpenAI and Anthropic's coding tools totally changed Foyer, which has raised $8 million in funding. About 20 people at Foyer used to work on the company's browser extension offering, Merlin AI, growing it to 900,000 Chrome users. Now, it's managed by three developers armed with AI coding tools, Rai said.
That overhaul freed up Foyer to develop Thine, which ambiently records a user's surrounding audio through their smartphone. The team hopes that giving an AI system all this "memory" as context will let people offload mundane tasks. That requires voice-to-text transcription and managing troves of user data. There, too, AI tools do the heavy lifting.
"The kind of work which 50 people would have done two or three years back, we are right now doing with around 15 developers," Rai said. "It's a massive save."
Individual plans with Anthropic and OpenAI can be cheaper, but the enterprise tiers come with extra tools and perks. The labs don't train their AI models on enterprise customers' data, and administrators can manage usage across sprawling teams.
An Anthropic spokesperson said that while small teams can work well with individual plans, companies pick the enterprise options for added security, governance, and visibility. OpenAI didn't respond to a request for comment.
Beyond the lower prices of the individual plans, Rai and Saxena said their team likes the flexibility to hop between tiers when a new model comes out.
"What is super clear is that it's not a winner-takes-all market," Rai said. "Like, at all."
Hoping for cheaper tokens
Rai and his team are watching the price of tokens closely as Anthropic and OpenAI push toward their respective IPOs. He's hoping that semiconductor improvements continue — Nvidia has been slashing the cost per token of LLMs — and that large AI providers release more open-source models.
If token prices do keep collapsing, Foyer might not have to worry about OpenAI and Anthropic cutting off subsidies for the individual plans that are helping the startup. A pay-as-you-go bill for $30,000 to $40,000 of AI coding tool usage could fall to $2,000 or $3,000, Rai hopes. This would also pave the way for Foyer's compute-heavy companion app, Thine.
Saxena said that this year, improvements in coding tools, plus Foyer's low costs, have meant his team uses "100x" more tokens than before.
His CEO corrected him. "100,000x," he said.
Have a tip? Contact this reporter via email at [email protected], or over text, Signal, Telegram, or WhatsApp at 415-757-8198. Use a personal email address, a nonwork WiFi network, and a nonwork device; here's our guide to sharing information securely.
Read next
Stephen is a Senior Tech Reporter at Business Insider, covering OpenAI, Anthropic and the ecosystem around the leading artificial intelligence companies.Previously he covered technology at SFGATE, and has written for The Wall Street Journal, The Information and CNBC. He studied journalism and economics at Northwestern University.His work has earned an SF Press Club Investigative Reporting Award and, in 2025, SPJ NorCal’s Excellence in Journalism Award for Technology Reporting.Stephen lives in San Francisco. Contact him via email at [email protected], or on Signal, Telegram, or WhatsApp at 415-757-8198. Use a personal email address, a nonwork WiFi network, and a nonwork device; here's our guide to sharing information securely.
