For the last few years, artificial intelligence has been sold as corporate survival. Companies across sectors promised faster work, leaner teams and higher productivity through AI tools. Employees were pushed to adopt AI coding assistants, automation software and large language models at breakneck speed.
The result was visible everywhere including layoffs in tech, hiring freezes in media and pressure on workers to “do more with less”. AI was positioned as the great cost-cutter of the modern workplace. But now, a new problem is emerging inside corporate boardrooms: AI itself is becoming extremely expensive.
The AI bill is arriving faster than expected
What companies initially treated as an investment boom is slowly turning into a cost-management crisis. Businesses are now discovering that running AI at scale requires enormous spending on cloud infrastructure, GPUs, software licences and network capacity.
According to the 2025 State of AI Cost Management report, 84% of enterprises said AI infrastructure costs were already hurting gross margins by more than 6%. Around 26% reported profit erosion of 16% or more.
Even more worrying for finance teams, 80% of enterprises admitted they miss AI cost forecasts by more than 25%, while nearly a quarter underestimate spending by over 50%. Only 15% of firms can currently forecast AI costs within a 10% error range. For companies that embraced AI as a path to efficiency, the numbers are creating an uncomfortable contradiction.
Workers are now asking where the AI profits are going
As companies navigate rising AI expenses, workers are beginning to focus on another issue entirely: if AI is increasing productivity, why are employees not sharing in the gains? That tension is already spilling into labour disputes.
In South Korea, Samsung Electronics narrowly avoided a strike after union workers pushed for greater AI-linked profit sharing. The dispute reflected a broader fear among employees that companies are using AI to expand margins while wages stagnate.
In California, Governor Gavin Newsom has ordered studies into “universal basic capital”, a concept aimed at giving ordinary people ownership stakes in the AI economy. Even AI companies themselves, including OpenAI and Anthropic, have publicly discussed versions of wealth-sharing models tied to future AI profits. The political debate around AI is slowly shifting from innovation to distribution.
The AI era is entering a more uncomfortable phase
For much of the past two years, AI conversations were dominated by hype — faster products, trillion-dollar valuations and fears of human replacement.Now the conversation is becoming more grounded. Finance teams are worried about runaway infrastructure costs. Investors are beginning to question whether AI spending is producing real returns. Workers are demanding a share of productivity gains. Governments are exploring redistribution models. And companies that once pushed AI adoption without hesitation are quietly pulling back in places where costs are spiralling.s
Microsoft’s AI pullback shows a bigger shift
One of the clearest examples is emerging from Microsoft. The company had aggressively rolled out Anthropic’s Claude Code across teams including developers, designers and project managers, encouraging widespread experimentation with AI-assisted coding. Employees reportedly embraced the tool rapidly.
But within months, Microsoft allegedly began cancelling many direct Claude Code licences and redirecting employees towards GitHub Copilot CLI instead. The decision does not indicate a breakup between Microsoft and Anthropic. Their multi-billion-dollar partnership remains intact. Even the world’s largest tech firms are beginning to rethink how much AI spending is sustainable internally.The AI adoption race is no longer only about capability. It is now about controlling the bill.
Uber spent its annual AI coding budget in four months
The problem is not limited to Big Tech.Ride-hailing giant Uber recently revealed that its entire 2026 AI coding tools budget had already been exhausted within the first four months of the year.The company reportedly used internal leaderboards to rank teams based on AI tool usage, pushing employees towards heavier integration.The excitement around productivity gains is now colliding with financial reality.
Companies are retreating from pure cloud AI models
Another major trend is beginning to take shape that is companies are trying to regain control over AI infrastructure costs. The report found that 67% of enterprises are actively planning to “repatriate” AI workloads from public cloud systems back to private or hybrid infrastructure. Another 19% are evaluating similar plans.
Today, 61% of companies already operate hybrid AI systems combining public and private infrastructure. Yet only 35% properly include on-premise AI costs in financial reporting, creating massive blind spots in budgeting and forecasting. This tells us that the industry is entering a second phase of AI adoption — one focused less on experimentation and more on cost discipline.