The End of Subsidized AI
The subsidy era is ending. The intelligence era is beginning. Mike Molinet & Govind Kavaturi break down what Claude dropping token subscriptions and Anthropic's Mythos model mean for every builder's margins and strategy.
Mike Molinet & Govind Kavaturi

Last week we talked about humans at the edges. Intent. Verification. Exceptions. The middle is AI. The edges are you.
This week: two things happened that seem unrelated. They're not.
The Two Signals
Signal 1: Claude removed token subscription support for OpenClaw.
If you were running AI agents on Claude Pro Max at $200/month, that's over. API pricing now.
Heavy usage? $2,000/month. Maybe more.
The cost of running agents with the most intelligent AI model went 10x overnight.
Signal 2: Anthropic hinted at Mythos, a new super intelligent model.
A model so far ahead of current benchmarks they're delaying release. Not for hype. Because they're genuinely nervous about how disruptive it could be.
Way above Claude Opus 4.6. Way above GPT 5.4.
Not a step. A leap.
Two signals. One message.
The training wheels are coming off.
The Subsidy Era
The last two years of AI pricing have been a lie.
Not a scam. A strategy.
OpenAI, Anthropic, Google. They subsidized usage to win market share. Lost money on every API call to get you hooked.
$200/month for unlimited Claude? That was never sustainable. That was customer acquisition.
ChatGPT Plus at $20/month? Subsidized.
Free tiers everywhere? Subsidized.
Cursor at $20/month running heavy inference? Subsidized.
The AI you've been building with was never that cheap. You were paying with someone else's money.
We built on prices that were never real.
Why It's Ending Now
Three reasons.
The race is narrowing.
Two years ago it was a land grab. Get users at any cost. Now the winners are emerging. OpenAI. Anthropic. Google. Maybe a few others.
When you're fighting for share, you subsidize. When you've won, you monetize.
Investors want returns.
Anthropic raised $50B. OpenAI raised $122B last week.
Those checks didn't come with infinite patience. Investors want a path to profitability. Subsidies are the first thing to go.
The models are getting expensive.
Mythos-level intelligence doesn't come cheap.
Training costs in the billions. Inference costs scale with capability. The smarter the model, the more compute it burns.
Someone has to pay. It's about to be you.
What This Means For Builders
Your costs are going up. Maybe 2x. Maybe 10x.
If your margins were thin, they might go negative.
If you were dependent on one provider's pricing, you're exposed.
This isn't speculation. OpenClaw users woke up to 10x costs this week. No warning. Just a policy change.
That's the new reality. Pricing can shift overnight. The subsidy can disappear whenever the provider decides.
The Mythos Signal
Anthropic is sitting on a model they're afraid to release.
Think about that.
Not hyping it. Not teasing a launch date. Actively delaying because of what it might do.
Benchmarks way above current. Capabilities that change what's possible. So disruptive they're being careful.

This isn't marketing. This is a company genuinely nervous about what they built.
What happens when Mythos releases?
- Everything we know about capability changes.
- Pricing for that tier will be astronomical.
- The gap between frontier and open source will widen.
The subsidy era ending isn't just about cost. It's about access.
The most powerful AI might not be available at any price. Or only at prices most builders can't afford.
The Two Futures
Future A:
You built on subsidized pricing. Assumed $200/month was real. Margins depended on cheap inference.
Costs 5x. Margins gone. You stop building.
Future B:
You saw this coming. Mixed models. Built flexibility. Priced for reality.
Costs rise. You're ready. You keep building.
The difference is what you do in the next 6 months.
The Framework
Here's what to do.
Audit your AI costs today.
Not what you budgeted. What you're actually paying. Per user. Per task. Per month.
Model your costs at 5x.
What breaks? Where do margins go negative? What survives?
Identify commodity tasks.
Simple summarization. Basic extraction. Template generation. Move these to open source. Llama. Mistral. Qwen. DeepSeek.
Identify reasoning tasks.
Complex analysis. Multi-step logic. Creative problem-solving. Keep these on frontier models. Pay for what matters.
Build provider optionality.
GPT 5.4 still offers token subscription at $200/month for advanced planning. That's an option.
Claude API is an option. Local models are an option. Having options is the point.
Don't depend on one provider's goodwill. They're a business. They'll act like one.
Price for reality.
If AI costs 5x tomorrow, do your margins survive?
If you're charging $10/month and spending $8 on inference, you're one policy change away from done.
Build margin. Real margin. Not subsidy-dependent margin.
The Easy Part Is Over
The last two years were the easy part.
Cheap models. Subsidized APIs. Low barriers. Everyone could play.
That era is ending.
What's coming is harder. More expensive. More competitive. More real.
But here's the thing.
When subsidies end, tourists leave.
Builders who only worked with subsidized pricing? Gone.
What's left are real builders. Real margins. Real leverage.
Less competition. Stronger survivors.
If you make it through the transition, the other side is better.
Vol 1: The opportunity exists. Economics changed.
Vol 2: Smart operators doing dumb work. Patterns to spot.
Vol 3: Stop automating. Start eliminating.
Vol 4: The game changed. Train like it matters.
Vol 5: Agents will run daily life. Build the access.
Vol 6: $100M solo founder. The grind behind the highlight reel.
Vol 7: The launch economy. 1 million launches. 10,000 winners.
Vol 8: Trust is the new moat. Build it in from day one.
Vol 9: Stop selling AI. Sell the outcome, not the motor.
Vol 10: The human at the edges. Intent. Verification. Exceptions.
Vol 11: The end of subsidized AI. The intelligence era is beginning.
The subsidy era got us here. It won't take us further.
Mythos is coming. Pricing is rising. Access is narrowing.
The builders who thrive will be the ones who planned for this.
What happens to your margins when the subsidies are fully gone?
—Mike & Govind
Building on AI APIs? Audit your costs this week.