What is the story about?
The big promise of artificial intelligence to boost productivity is now facing one major issue for tech giants: very high cost.
While companies like Microsoft, Uber and Meta are seeing strong productivity benefits from AI coding tools, at the same time, they are also struggling with rapidly increasing spending on tokens, GPUs, servers and AI infrastructure.
Microsoft is reportedly cancelling most internal licences for Anthropic’s Claude Code by June 30 after the company saw very high usage costs, according to reports from The Verge and other media outlets. The company had introduced the tool to thousands of engineers in late 2025 for coding, debugging and software review work.
However, as more employees started using the tool daily, the billing increased sharply because of token-based pricing. Microsoft is now moving many developers toward its own GitHub Copilot CLI tools, mainly to reduce costs and improve integration inside its own ecosystem. The move is important because Microsoft itself has invested billions of dollars in Anthropic.
Ride hailing giant Uber also faced a similar situation, but on a much bigger scale. According to The Information, Uber finished its full 2026 AI coding budget by April itself after deploying Claude Code widely across engineering teams. CTO Praveen Neppalli Naga said around 84% of Uber’s nearly 5,000 engineers were using the AI tool by March, and almost 70% of committed code was being generated through AI.
The cost, however, became extremely high. Heavy users reportedly cost Uber between $500 and $2,000 every month. Naga himself said one two-hour coding session cost nearly $1,200. Uber had even created internal leaderboards to encourage more AI usage among employees.
Now, senior executives are starting to question whether the spending is giving enough business return.
Uber COO Andrew Macdonald recently said that while internal adoption is very strong, it is still difficult to connect AI-generated coding with major new consumer products or innovation at scale.
The issue is not limited to Microsoft and Uber only.
Nvidia Vice President Bryan Catanzaro told Axios that for his teams, “the cost of compute is far beyond the costs of the employees,” showing how AI infrastructure spending is becoming one of the biggest operational expenses for tech companies.
Meta is also under pressure after increasing its 2026 capital expenditure guidance to between $125 billion and $145 billion, mainly for AI data centres, GPUs and memory infrastructure. Although the company reported strong quarterly earnings, investors became worried about the scale of AI spending and the stock fell sharply after the announcement.
While companies like Microsoft, Uber and Meta are seeing strong productivity benefits from AI coding tools, at the same time, they are also struggling with rapidly increasing spending on tokens, GPUs, servers and AI infrastructure.
Microsoft is reportedly cancelling most internal licences for Anthropic’s Claude Code by June 30 after the company saw very high usage costs, according to reports from The Verge and other media outlets. The company had introduced the tool to thousands of engineers in late 2025 for coding, debugging and software review work.
However, as more employees started using the tool daily, the billing increased sharply because of token-based pricing. Microsoft is now moving many developers toward its own GitHub Copilot CLI tools, mainly to reduce costs and improve integration inside its own ecosystem. The move is important because Microsoft itself has invested billions of dollars in Anthropic.
Ride hailing giant Uber also faced a similar situation, but on a much bigger scale. According to The Information, Uber finished its full 2026 AI coding budget by April itself after deploying Claude Code widely across engineering teams. CTO Praveen Neppalli Naga said around 84% of Uber’s nearly 5,000 engineers were using the AI tool by March, and almost 70% of committed code was being generated through AI.
The cost, however, became extremely high. Heavy users reportedly cost Uber between $500 and $2,000 every month. Naga himself said one two-hour coding session cost nearly $1,200. Uber had even created internal leaderboards to encourage more AI usage among employees.
Now, senior executives are starting to question whether the spending is giving enough business return.
Uber COO Andrew Macdonald recently said that while internal adoption is very strong, it is still difficult to connect AI-generated coding with major new consumer products or innovation at scale.
The issue is not limited to Microsoft and Uber only.
Nvidia Vice President Bryan Catanzaro told Axios that for his teams, “the cost of compute is far beyond the costs of the employees,” showing how AI infrastructure spending is becoming one of the biggest operational expenses for tech companies.
Meta is also under pressure after increasing its 2026 capital expenditure guidance to between $125 billion and $145 billion, mainly for AI data centres, GPUs and memory infrastructure. Although the company reported strong quarterly earnings, investors became worried about the scale of AI spending and the stock fell sharply after the announcement.
