A quick follow-up to the I/O post.
Buried in the rapid-fire agentic-product reveal at I/O was a single pricing slide that nobody I read covered in any depth: Google cut AI Ultra from $250/month to $100/month. Effective immediately. Existing subscribers grandfathered for a billing cycle, then auto-migrated to the new price.
That’s a 60% cut on the flagship consumer tier of the second-most-valuable AI product on earth, announced in the middle of a subscription war where everyone else is holding price. It deserves more attention than it got.
The competitive board, after the cut
The consumer top-tier prices, as of this week:
- ChatGPT Pro: $200/month
- Claude Max: $200/month
- Google AI Ultra: $100/month (was $250)
Two months ago, AI Ultra was the most expensive mainstream consumer AI subscription on the market. Today it is the cheapest of the three top tiers, by half. The reframe is total.
Why Google did this
I see three plausible reads, and unlike the Trump EO post I think these are not mutually exclusive — they reinforce each other.
1. They needed Gemini Spark to land in consumers’ hands. Spark is Google’s general-purpose agentic surface — the equivalent of the agentic mode every other lab has. Adoption for an agent-first product matters more than ARPU for an agent-first product, because the data flywheel from agent traces is worth more than the marginal subscription. Cutting the price to expand the user base is a defensible strategic trade if you believe the agent-trace data compounds. Google clearly believes it.
2. They are pricing against a future cost curve, not a current one. Gemini 3.5 Flash, which is what Ultra runs on for most workloads, is dramatically cheaper to serve than the equivalent Anthropic or OpenAI flagship. Google’s TPUs let them set a price floor the competitors can’t follow without compressing margin. The $100 number is the floor that’s hard to undercut without losing money. Setting that floor publicly is itself the move.
3. Workspace bundling. AI Ultra at $100 plus Workspace Individual at $10 is materially cheaper than ChatGPT Pro alone. For the small-business / pro-sumer segment that buys both productivity software and AI, the bundle is the play. The pricing cut is the cover for the bundle.
What ChatGPT and Claude do next
The obvious move is to match or hold. Holding is hard. The mid-tier segment — people paying $200 for Pro or Max — was, in my experience, already complaining about whether they’re getting their money’s worth, and is now being shown a competitor with a comparable agent at half the price.
OpenAI has the harder problem. Their margin pressure is well-documented (the IPO S-1 will make it more so), they don’t have a TPU-equivalent cost-of-goods advantage, and Pro at $200 was already the highest-priced consumer tier in the market. They’ll either:
- Hold the price and lose net subscribers,
- Cut and compress margin going into an IPO, or
- Re-segment the product (Pro stays $200, introduce a new “Plus Max” at $100-120 with reduced compute allocation).
The third is the move I’d bet on. It buys them the optics of not cutting while functionally cutting.
Anthropic is in better shape. Claude Max is sold disproportionately to enterprise individuals and power users who are largely insensitive to a $100 price differential — they need the model. The bigger pressure is on Claude Pro at $20, which is now competing against AI Ultra’s much-broader feature set for only a 5x price differential. Anthropic will probably hold Max and feature-up Pro.
Why this is the I/O headline that should have been
The agentic-product slide gave us five new things to build into demos. The pricing slide reset the consumer AI subscription market.
Builds get out-shipped. Prices don’t get un-cut. The latter is the more durable change.
This is the story I’d point to in twelve months when someone asks what actually shifted at I/O 2026.