Microsoft’s Cost-Cutting AI Push Collides With a $190 Billion Bet

Satya Nadella has sent an unusual message to Microsoft’s workforce: stop burning expensive frontier models on mundane tasks. The directive, which employees have christened “Tokenmaxxing,” marks a striking reversal for a company that plans to spend roughly $80 billion on AI infrastructure in fiscal 2026 alone. The total capital expenditure for the current financial year could reportedly reach $190 billion, putting the efficiency drive into stark context.

The shift is not purely about philosophy—privacy concerns are also reshaping the company’s internal AI habits. Since June 11, Microsoft staff have been barred from using Anthropic’s new Claude Fable 5 model internally, because Anthropic stores user inputs and outputs for 30 days by default, and up to two years when violations are reported. That conflicts with Microsoft’s standard of zero data retention. External customers can still access Claude Fable 5 via GitHub Copilot and Azure AI Foundry, but Microsoft’s legal team is now weighing the risks for sensitive client data and intellectual property. No conclusion has been announced.

The efficiency drive extends to the hardware side, where Microsoft is leaning on smaller, optimised models—using techniques such as distillation and quantisation—to maintain performance while cutting costs. Nadella put it bluntly: “Frontier models should not be applied to non-frontier problems.”

Nvidia Joins the Hardware Pivot

Even as the company reins in AI spending on the software side, it is forging deeper ties with Nvidia to revive its sluggish Windows device business. On the Computex show floor, Nvidia CEO Jensen Huang unveiled the “RTX Spark,” a new AI superchip built on the Arm architecture and co-developed with MediaTek and Microsoft. The chip is designed to run personal AI assistants directly on laptops and compact desktop PCs.

Microsoft will embed the RTX Spark into its upcoming “Surface Laptop Ultra,” which promises up to 128 GB of memory and substantial local compute power. Partners including ASUS, Dell, and Lenovo are also preparing devices around the chip, with availability expected in the autumn. The goal is to ignite a new wave of Windows 11 machines and arrest a two-percent decline in Windows licence and device revenue.

Should investors sell immediately? Or is it worth buying Microsoft?

The broader business, however, continues to hum. Third-quarter fiscal 2026 results beat expectations: revenue hit $82.89 billion, earnings per share came in at $4.27, and Azure grew 40 percent year-on-year. The annualised AI revenue run rate now exceeds $37 billion. Yet the hardware slump and the looming Xbox margin squeeze are weighing on sentiment.

Xbox Margins Pinched, Markets Wary

Microsoft’s gaming division is facing planned job cuts after the fiscal year ends on June 30. Margins for the Xbox segment have reportedly slipped to around three percent, despite the $69 billion Activision Blizzard acquisition. Storage costs have quintupled in two years, adding further pressure. On June 11, a cautious Oracle outlook on AI investment costs dragged down the entire tech sector, adding a third headwind.

The stock closed at €336.95 in recent trading, roughly 30 percent below the October 2025 high of €478.10 and well beneath both its 50-day and 200-day moving averages. The relative strength index sits at 37.5, signalling oversold territory. Since the start of the year, the shares have lost about 16 percent of their value.

Investors will be watching the July 29 release of fourth-quarter results closely. That report will show whether Nadella’s efficiency campaign is already trimming costs—or whether the gargantuan investment bill continues to gnaw at margins. For now, the market wants proof that the new AI laptops can revive hardware sales, and that Microsoft’s twin push for cost discipline and chip innovation can translate into a convincing financial narrative.