Google Cut AI Ultra by 60%. The Meter Is the Story.

Google charged $250 a month for AI Ultra. Then they cut it to $100. That is a 60% price drop in under a year. But the price is not the story. The story is what replaced it: a compute-based billing model that charges you for what you actually use, not a flat fee for all-you-can-eat access. The subscription got cheaper. The usage got metered. And every AI company on the planet is watching.

I think this is the most important pricing move in AI since OpenAI set the $20/month anchor with ChatGPT Plus in February 2023. Here is why it matters, what it means for your wallet, and how to position yourself on the right side of the shift.

The Metered Intelligence Model

Here is the framework you need to remember: the Metered Intelligence Model.

For three years, AI subscriptions worked like gym memberships. You paid a flat monthly fee. Some people used the service constantly. Most people barely touched it. The heavy users were subsidized by the light users, and the company ate the difference. That model is dying.

Google's move from flat pricing to built-in AI credits is the first major consumer-facing signal that AI is becoming a utility, not a product. Think electricity, not Netflix. You do not pay one price for unlimited kilowatt-hours. You pay for what you consume. The Metered Intelligence Model says the same logic now applies to AI inference.

The reason is simple math. A user running Gemini 3.5 Flash with a 1M token context window and 65,000 max output tokens burns orders of magnitude more compute than someone asking for a recipe. At $250 flat, Google was losing money on power users and overcharging casual ones. The Metered Intelligence Model fixes that by aligning price with cost.

This is not just a Google thing. It is the economic gravity that every AI provider will eventually feel.

The $250 to $100 Pipeline: How Google Turned a Pricing Failure Into a Monetization Playbook

Let me show you exactly how this played out, because the sequence matters.

Google launched AI Ultra in May 2025 at $249.99 per month. That is 12.5x the price of ChatGPT Plus. The pitch was "highest usage limits" and access to the most capable Gemini models. For first-time users, Google offered 50% off the first three months, bringing the effective intro price to about $125.

The problem? Almost nobody thought it was worth it. PhoneArena reported that "supporters of Google's Ultra plan are not that numerous." Independent reviewers found that the functional difference between Ultra and the $20 Pro plan was minimal. One reviewer's summary was blunt: "Basically everything in Pro is in Ultra. Go get Pro."

So Google had a product at a premium price point with weak differentiation. That is the hard way to learn a lesson about value.

Then came the March 2026 quota reduction. According to the Google AI developer forum, users woke up to dramatically lower usage limits on the same $249.99 plan. One subscriber described it as "a massive reduction in effective limits," not a restructuring. The community was furious.

Here is where it gets interesting. Google responded by doing two things at once. First, they slashed the headline price. Digital Trends reported a new $100 tier alongside a price cut on the top plan, bringing Ultra down to around $200. Second, they reframed the entire billing model around "built-in AI credits" consumed through their Antigravity platform. Google AI Pro became the tier for "practical builders" with generous Flash limits. The smarter read: Google used the backlash as cover to restructure its entire monetization engine.

Think about what actually happened. Google took a flat subscription, broke it into metered compute credits, lowered the sticker price to reduce friction, and created a system where heavy users pay more as they scale. That is not a discount. That is a completely different business model.

My read is that Google deliberately front-loaded the pain. Launch high, absorb the criticism, then "generously" cut prices while quietly installing the metering infrastructure that will generate far more revenue long-term. The $100 price tag feels like a deal compared to $250. But the real economics now live in the credit consumption rate, not the monthly fee.

This is the playbook every AI company will study. The old way was selling a fixed product at a fixed price. The new way is selling access to a meter. The meter is where the money lives.

It is unclear whether Google planned this exact sequence from the start or improvised after the backlash. But the outcome is the same: they now have a billing system that scales revenue with usage, and a price point that looks consumer-friendly on the surface.

Here is the part most people miss. When you break down the Ultra bundle, about $150 of the original $250 was arguably non-AI value. Google One's 30 TB storage tier costs roughly $150 per month on its own. YouTube Premium runs $14.99. Strip those out and the AI-specific cost at the original price was maybe $85 to $90. At the new $100 price point with bundled storage and YouTube, Google may actually be charging less for the AI compute than before, while making the metering system do the heavy lifting on revenue.

That is stupid easy to miss if you only look at the headline number.

2031

Three signals inside the same shift

Pull back from the monthly pricing drama and look at where this lands in five years.

The AI subscription market is splitting into two lanes. Lane one: cheap front-door plans for casual users. $20 a month, generous enough for daily questions, light content creation, and basic research. OpenAI, Anthropic, and Google all offer this tier today, and it is not going away. Lane two: compute-metered premium access for builders, agencies, and power users. This is where Google AI Ultra now lives, and it is where the real revenue will compound.

This split mirrors what happened in cloud computing between 2010 and 2020. Amazon Web Services did not win by offering one price for everyone. AWS won by metering every API call, every gigabyte of storage, every second of compute. The customers who used the most paid the most. The customers who used the least paid almost nothing. That model generated $90 billion in annual revenue by 2023.

The asymmetric advantage here belongs to companies that control both the models and the metering infrastructure. Google has Gemini, TPU clusters, and the billing system. Microsoft has OpenAI models, Azure, and Copilot licensing. Anthropic has Claude but relies on AWS for infrastructure. Meta gives Llama away for free, which is a counterpositioning play that forces everyone else to justify their compute margins.

The strategic question for the next five years is not "which model is best." It is "who owns the meter." The company that controls how AI usage is measured, priced, and billed will capture the most value. Models will commoditize. Metering will not.

There is a contrarian case worth considering, though. Maybe compute-based billing stays confined to developers and enterprises. Maybe consumers never accept metered AI the way they accepted metered electricity. The evidence is mixed. OpenAI's biggest consumer success is still the flat $20 ChatGPT Plus plan. Anthropic's Claude Pro is also flat-rate. People hate bill shock. They hate complexity. A world where you need to budget your AI tokens like cell phone minutes in 2005 might generate enough backlash to keep flat tiers alive for retail users.

But for builders? For agencies? For anyone running AI workflows at scale? The meter is coming. And the companies that learn to sell compute efficiently will build the flywheels that define the next decade of AI economics.

One case study worth watching: Nvidia nearly went bankrupt in the early 2000s before GPUs became essential infrastructure. The companies building metering systems today might look similarly fragile. But if AI usage grows the way cloud usage grew, the metering layer becomes the most valuable real estate in the stack.

What to Build This Weekend

You do not need to wait for the industry to settle its pricing debates. You can start positioning yourself right now. Here is how.

First, sign up for Google AI Pro at $20 per month if you have not already. It gives you access to Gemini 3.5 Flash with generous limits. Use it to understand what "compute-based" feels like in practice. Pay attention to when you hit quota walls. That friction is the future of AI pricing, and understanding it now gives you an edge.

Second, pick one tool from this week's digest and build something small with it. Stitch lets you prompt your way to a complete UI design. No code required. No design degree needed. Open it up, describe an app layout in plain language, and see what comes out. Your goal is not perfection. Your goal is one finished thing.

Third, test the monetization angle. Honen v1.2 lets you turn rough notes or expertise into a polished AI course. If you know something well enough to explain it in bullet points, you can have a sellable course structure by Sunday night. The tool does the formatting. You supply the knowledge. That is the kind of asymmetry that compute-based pricing rewards: low input cost, high output value.

Fourth, if you are already running AI workflows for clients or for your own business, audit your usage. How many tokens are you consuming per project? What does that cost at current API rates? What would it cost under a credit-based system? Do the napkin math now. When metered billing becomes the default, the builders who already understand their unit economics will price their services correctly. Everyone else will guess and lose margin.

Things will break. You will hit limits. The tools will behave weirdly sometimes. That is normal. The point is to get your reps in before the pricing shift forces everyone to pay attention. The people who understand compute economics in 2026 will have a massive head start over those who figure it out in 2028.

The meter is coming. Learn to read it before it reads you.