💸 Microsoft's AI costs more than hiring

Good morning, tech readers. Microsoft has been pushing its employees to use AI as aggressively as possible. Then its own internal data leaked showing the technology costs more than the workers it was meant to replace.

The same week, the Pentagon signed a $9.7 billion deal to consolidate all of its Microsoft licenses. Has the AI sticker shock finally arrived, or is enterprise spending just getting started?

In today's recap:

LATEST DEVELOPMENT

Microsoft's internal research now suggests deploying AI tools costs more than keeping humans on payroll for many tasks, and enterprise America is feeling it. Axios reports one company burned $500 million in a single month after failing to cap Claude licenses, and Uber's COO said AI costs are getting "harder to justify." Microsoft itself just canceled most of its Claude Code licenses, in part over costs.

The enterprise AI enthusiasm wave is breaking against real cost data, and the pressure isn't coming from critics, it's coming from Microsoft's own numbers. If the correction sticks, companies will shift from "tokenmaxxing" to targeted AI use, which could compress revenue for the broad AI tooling layer faster than anyone expected.

→ Axios

Your browser gets you to a page. Norton Neo gets you to the answer. The first safe AI-native browser built by Norton moves with you from idea to action without slowing you down. Magic Box understands your intent before you finish typing. AI that works inside your flow, not beside it. No prompting. No copy-pasting. No switching apps.

Built-in AI, instantly and for free. Privacy handled by Norton. Built-in VPN and ad blocking protect you by default. No configuration. No extra apps. Nothing to think about.

Fast. Safe. Intelligent. That's Neo.

Download Norton Neo

The Defense Department just awarded Microsoft a five-year, $9.69 billion enterprise software agreement called the Core Enterprise Technology Agreement, consolidating Microsoft 365 subscriptions, cloud access, and on-premises licensing across the military services, intelligence community, and US Coast Guard into a single contract. The deal doesn't add new spending. It rolls up fragmented go-it-alone purchases that had quietly piled up across years of siloed procurement.

The same company whose AI cost data is rattling CFOs just locked in the US military for nearly a decade of guaranteed revenue. If the Pentagon's consolidation model spreads to federal civilian agencies, Microsoft's enterprise base gets harder to dislodge even as ROI questions mount in the private sector.

→ Reuters

Stanford researchers analyzed 4.2 million job applications processed by pymetrics, an AI hiring platform used by 156 employers with $225 billion in combined annual revenue. They found that 26 percent of Black applicants and 15 percent of Asian applicants submitted applications to positions where the algorithm discriminated against their group. If those candidates had advanced at the same rate as the most recommended group, roughly 40,000 more people would have moved to the next screening stage.

The study names a specific mechanism: algorithmic monoculture. When dozens of employers use the same hiring model, a single system's bias propagates across every company simultaneously. If regulators pick this up, the EEOC's four-fifths rule could be applied to AI hiring vendors for the first time, forcing transparency and independent audits that most platforms don't currently provide.

→ Stanford HAI

The FAA declared a mishap investigation after SpaceX's latest Starship test. The booster separated on schedule, but its engines failed during the return burn, sending it into an uncontrolled splashdown in the Gulf of Mexico instead of the planned landing. The spacecraft itself completed the mission, releasing 20 mock satellites before a planned Indian Ocean splashdown. No injuries or property damage were reported.

SpaceX's case for a $2 trillion valuation rests heavily on Starship reaching operational cadence. Every grounding extends the timeline for the reusable architecture that's supposed to undercut every other launch provider on cost. If the booster failure traces to a systemic issue, the space datacenter and lunar logistics plays that depend on Starship get pushed further out.

→ AP News

ByteDance is developing proprietary central processors to support its AI infrastructure rollout, three Reuters sources confirmed. The company is running two tracks in parallel: one based on Arm architecture, one on the open-source RISC-V instruction set. It hasn't committed to either yet. The push comes as Intel and AMD have raised CPU prices by 10 to 35 percent quarter-over-quarter, and Intel is quoting Chinese customers delivery lead times of up to six months.

When China's largest consumer tech company moves to custom silicon, it's not just a cost play, it's infrastructure independence. ByteDance joins Google, Amazon, and Microsoft in concluding that custom chips beat the complexity of designing them. If RISC-V gets traction here, it opens a credible path for Chinese firms to sidestep Western CPU dominance entirely.

→ Reuters

QUICK HITS

🧡 Enjoyed this issue?

🤝 Recommend our newsletter or leave a feedback.

Cheers, Jason

Connect on LinkedIn, & Twitter.