Microsoft and Meta Scale Back Internal Use of Anthropic's Claude AI to Cut Costs

Microsoft and Meta, two of Anthropic's largest enterprise customers, are reportedly scaling back their internal usage of the Claude AI model. This move is primarily motivated by cost-cutting initiatives and a strategic pivot towards leveraging their own internally developed AI tools. Microsoft, for instance, has cut its internal spending forecast for Anthropic by over a third from an initial projection of at least $1 billion annually. Similarly, Meta's internal usage of Claude Code has seen a significant decline, with user numbers dropping from approximately 60,000 to 30,000, partly due to layoffs and the adoption of their own Muse Code and MetaCode. This development is highly significant for practitioners in cloud, DevOps, and AI, as it signals a maturing phase in AI adoption where cost optimization is becoming a paramount concern. The initial enthusiasm for readily available, powerful external AI models is now being tempered by the realities of scaling costs. For many organizations, the allure of quick integration and advanced capabilities offered by third-party AI services can quickly turn into substantial, unpredictable expenditures, especially as usage grows. This trend directly impacts FinOps teams, who are increasingly tasked with managing AI spend, a responsibility that 98% of FinOps teams now handle, up from 31% just two years prior. The broader context here is the accelerating trend of enterprises seeking greater control and cost efficiency over their AI investments. While external models offer immediate access to cutting-edge AI, the long-term operational costs, particularly for high-volume or sensitive workloads, can become prohibitive. This has led to a dual strategy: optimizing the use of external models through techniques like prompt caching, model routing, and right-sizing, while simultaneously investing in proprietary AI development. The goal is to reduce reliance on variable, token-based pricing structures and to align AI capabilities more closely with specific business needs and cost structures. The FinOps Foundation's 2026 survey highlights that organizations are being asked to fund AI investment through optimization savings, further emphasizing this shift. In practice, this means practitioners should be rigorously evaluating the total cost of ownership for AI solutions. This includes not only the direct API costs but also data transfer, storage, and the operational overhead of managing external integrations. Organizations should prioritize implementing robust AI FinOps frameworks that provide visibility, accountability, forecasting, optimization, and governance across AI consumption. Furthermore, exploring strategies like model right-sizing, caching, batching, and setting budgets for token usage are crucial for controlling AI costs. For workloads that are core to their business or require significant customization, investing in internal model development or fine-tuning open-source alternatives may prove more cost-effective in the long run, as demonstrated by Meta and Microsoft's strategic shift. This also means closely monitoring the performance-to-cost ratio of AI models, ensuring that the qualitative output justifies the token cost.