Two enterprise AI platforms introduced new pricing structures effective September 1, 2026 — Microsoft Foundry and Google’s Gemini Enterprise Agent Platform. Neither is a coordinated move; the shared date appears to be coincidence. But both changes point the same direction: enterprise AI pricing is shifting from flat, uniform rates toward usage- and region-based tiers.
Microsoft Foundry: regional premiums on new models
Per Microsoft’s own announcement on its Foundry blog, deployments outside the US now carry a price premium over Global pricing:
- EU Data Zone: 9% above Global pricing
- Other Regional deployments outside the US: 7–16% above Global, depending on region
- New APAC Data Zone (newly launched): 20% above Global
The premium applies differently depending on how a customer buys. For Standard pay-as-you-go deployments, it only applies to models launched on or after September 1, 2026 — customers who stay on models they’re already using see no change. For Provisioned Throughput (PTU) customers, the increase applies across the board to any EU Data Zone or non-US Regional PTU, regardless of when the underlying model launched.
Google: Gemini Enterprise Agent Platform starts billing agent state
Google’s Gemini Enterprise Agent Platform began billing for three components on the same date, per its own published pricing page: Sessions (which manage conversation history and state), Memory Bank, and Skills Registry. Sessions billing breaks into two parts — storage, billed as Agent Storage at $0.30 per GiB-month, and read operations, billed at $0.085 per Agent Compute vCPU-hour for every 3 million reads. It’s a usage-based model rather than a flat per-session fee, prorated to actual hourly usage per project.
What this means for buyers
Both changes push cost accounting further into engineering’s hands. Microsoft’s regional premiums mean the same model can cost differently depending on data residency choices made for compliance reasons, not performance. Google’s shift means agent memory and session state — previously bundled or unmetered in early agent-platform pricing — are now line items that scale with how much conversational history an agent retains. Both changes reward architectures that are deliberate about what they store and where, over ones that default to the broadest region or the longest retention window.
What’s confirmed and what isn’t
- Confirmed: both effective dates and the Microsoft percentage premiums, per Microsoft’s own Foundry announcement.
- Confirmed: the Google Sessions billing rates, per Google Cloud’s own published pricing page.
- Not independently verified: how the two announcements’ shared date affects real-world enterprise spend at scale — neither company has published aggregate cost-impact figures.
