AI’s new “digital divide” is widening—while Microsoft flips to pay-as-you-go agents

Microsoft has begun charging for its new AI agent using a pay-as-you-go model, a pricing shift the company says is the first of its kind in two decades. The change was announced alongside the launch of Copilot Cowork, and it is explicitly tied to the soaring cost of running artificial intelligence. At the same time, reporting highlights that corporate AI spending is creating a new inequality gradient inside firms and across workforces. Data cited in the cluster suggests the median company spends about $11.38 per worker on AI, while the top 1% can spend roughly $7,450 per worker. The geopolitical angle is that AI is becoming a strategic input to productivity, cybersecurity posture, and competitive advantage, and access is increasingly governed by budgets, model availability, and internal restrictions on how employees can use enterprise systems. When only the largest players can afford broad model access and permissive usage policies, they can accelerate automation, decision support, and software development faster than mid-tier competitors. That dynamic can translate into market concentration, labor displacement pressures, and a widening capability gap that governments may later try to address through procurement, industrial policy, or regulation. The cluster also notes that economic analysis of AI’s impact remains comparatively scarce, with “AI-pilled economists” trying to close the evidence gap—an important factor because policy responses often lag behind real-world adoption. Markets and the economy are likely to feel the effects through cloud and enterprise software demand, as well as through the cost structure of AI services. A move to usage-based billing can shift spending from predictable annual budgets to variable operating expenses, increasing volatility for firms that already report burning through AI allocations quickly. The example of Uber spending its annual AI budget in four months signals that AI cost inflation is not theoretical; it is showing up in corporate P&Ls and could pressure margins in sectors with high labor intensity or heavy customer interaction. In practical terms, investors may reprice exposure to AI infrastructure and tooling, while enterprise customers may seek cost controls, model governance, and tighter procurement—potentially affecting demand for AI compute, data services, and productivity suites. What to watch next is whether pricing changes like Microsoft’s pay-as-you-go model become an industry standard and whether they are paired with stronger cost-management tooling and usage governance. Watch for more corporate disclosures of “budget burn” timelines, including whether firms revise forecasts, cap agent usage, or renegotiate enterprise terms. Another key indicator is the emergence of more rigorous economic measurement of AI productivity and labor effects, since the cluster suggests the analytical base is still thin. Trigger points include further rapid budget depletion announcements, new restrictions on model access inside companies, and any policy or regulatory proposals aimed at AI access, security, or competition—any of which could accelerate or slow the widening divide.

Geopolitical Implications

• 01

  AI access is becoming a strategic differentiator; unequal enterprise adoption can widen industrial and security capability gaps between firms and, indirectly, between states via supply chains and procurement.

• 02

  Usage-based pricing may accelerate consolidation toward large vendors and large customers that can absorb variable AI costs, reinforcing market power.

• 03

  If budget burn continues, governments may face pressure to intervene through industrial policy, standards, or governance frameworks to ensure secure and affordable AI deployment.

Key Signals

• —More enterprise disclosures of AI budget depletion timelines and revised adoption targets
• —Whether other major vendors follow Microsoft’s pay-as-you-go model and how they price compute, data, and agent actions
• —Internal policy trends: restrictions on model use, audit requirements, and role-based access to enterprise models
• —Emergence of credible economic studies quantifying AI productivity, labor displacement, and cost curves