The AI industry is entering a new phase of disruption as Meta and Microsoft reportedly reduce employees’ reliance on Anthropic’s Claude. The move demonstrates how quickly yesterday’s preferred AI tool can become tomorrow’s strategic dependency—and how fiercely technology giants want to control their own AI stack.
Microsoft had reportedly expected to spend at least $1 billion annually on internal Anthropic technology, but has cut that projection by more than one-third. Employees are increasingly being directed towards Microsoft’s own ecosystem, particularly GitHub Copilot, although engineers retain access to alternative models.
Meta is following an even more visible path. Claude Code usage reportedly fell from around 60,000 employees earlier this year to approximately 30,000, while Meta promotes internally developed alternatives such as MetaCode and Muse Code.
The immediate driver is economics. Generative and agentic AI can produce impressive productivity gains, but intensive usage consumes enormous volumes of tokens and compute. As autonomous agents undertake longer tasks, enterprise AI expenditure can rise rapidly even as the unit cost of inference declines.
But the deeper issue is strategic control. Companies developing competing foundation models cannot indefinitely depend on rivals for critical engineering workflows. Proprietary AI also generates valuable usage patterns, developer feedback and operational knowledge that can strengthen future models.
This creates an innovation paradox. Restricting employees to internal platforms may reduce costs, protect intellectual property and accelerate proprietary development, but it can also limit access to whichever external model performs best. Innovation could increasingly fragment into competing AI ecosystems where commercial strategy influences model selection alongside technical capability.
The probability of further disruption is therefore high. Enterprises will increasingly adopt multi-model architectures, switching between models according to performance, security, sovereignty, governance and cost. The next AI battle will not simply be about building the smartest model—it will be about who controls the enterprise AI ecosystem without restricting the freedom to innovate.
