Latest / Ctrl AI Profit / Ep. 109 | OpenAI's New Money Grab: Pay Up or Get in Line
Transcript
- 0:20SPEAKER_00: OpenAI just announced something that sounds boring, but is actually terrifying.
- 0:24SPEAKER_00: They're letting companies lock in long-term compute access.
- 0:28SPEAKER_00: They're calling it guaranteed capacity.
- 0:31SPEAKER_01: It sounds like an enterprise feature, and it is.
- 0:34SPEAKER_01: But what it really means is that compute, the ability to run AI models, is scarce enough that companies are willing to sign multi-year contracts just to guarantee they can use it.
- 0:44SPEAKER_01: This is the infrastructure equivalent of paying for a reserved parking spot because there aren't enough parking spaces.
- 0:51SPEAKER_00: And that's the story I want to tell today.
- 0:53SPEAKER_00: Because if the biggest AI company in the world is telling you that there's not enough compute to go around, that affects everyone.
- 1:00SPEAKER_00: Not just Fortune 500 companies, not just tech startups, everyone who uses AI.
- 1:06SPEAKER_01: Including small businesses.
- 1:08SPEAKER_01: Because when compute is scarce, two things happen.
- 1:12SPEAKER_01: Prices go up, and smaller players get squeezed out.
- 1:16SPEAKER_01: That's what OpenAI is implicitly admitting with this product.
- 1:19SPEAKER_01: The demand for AI compute is exceeding the supply, and they're monetizing the scarcity.
- 1:25SPEAKER_00: Let me explain what guaranteed capacity actually is.
- 1:28SPEAKER_00: If you're a company that relies on OpenAI's models, GPT, whatever, to run your product, you can now pay up front to reserve a certain amount of compute for a set period.
- 1:38SPEAKER_00: Think of it like a season ticket.
- 1:40SPEAKER_00: You pay more, but you're guaranteed a seat.
- 1:42SPEAKER_00: If you don't buy the season ticket, you might show up and find out the event is sold out.
- 1:46SPEAKER_01: And that's the key insight.
- 1:48SPEAKER_01: OpenAI is saying our infrastructure is going to get more crowded, not less.
- 1:54SPEAKER_01: We're building more data centers, but demand is growing faster than supply.
- 1:59SPEAKER_01: If you want to be sure you can run your business on our models, you need to lock it in now.
- 2:04SPEAKER_00: Which is a brilliant business move for OpenAI.
- 2:07SPEAKER_00: They get long-term revenue commitments, and they get to prioritize paying customers when things get tight.
- 2:13SPEAKER_00: But for everyone else, it's a signal.
- 2:15SPEAKER_00: The era of cheap, unlimited AI compute might be ending.
- 2:19SPEAKER_01: Or at least the era of assuming it's always available.
- 2:22SPEAKER_01: You might still be able to use GPT whenever you want, but during peak times, during major product launches, during events that spike demand, the companies with guaranteed capacity get priority.
- 2:33SPEAKER_01: Everyone else waits.
- 2:35SPEAKER_00: Now, for a small business owner, this might sound irrelevant.
- 2:39SPEAKER_00: You're not signing multi-year compute contracts, but let me tell you why it matters to you.
- 2:45SPEAKER_01: It matters because the entire AI pricing model is built on the assumption that compute is cheap and getting cheaper.
- 2:51SPEAKER_01: Every AI tool you use, ChatGPT, Claude Gemini, your scheduling assistant, your email tool, all of them are priced based on compute costs that have been dropping.
- 3:02SPEAKER_01: If compute gets scarce and expensive, those prices go up, or features get limited, or free tiers disappear.
- 3:10SPEAKER_00: So the$100 a month tool you're using might become a$200 a month tool.
- 3:15SPEAKER_00: Or the free tier might dead capped at 500 queries instead of 2,000.
- 3:21SPEAKER_00: That's what compute scarcity looks like from the small business end.
- 3:25SPEAKER_01: And it's not theoretical.
- 3:27SPEAKER_01: We've already seen OpenAI introduce rate limits on their API.
- 3:31SPEAKER_01: We've seen service interruptions during peak usage.
- 3:33SPEAKER_01: Guaranteed capacity is just the formal commercial version of what's already happening informally.
- 3:38SPEAKER_01: If you want reliable access, you pay for it.
- 3:41SPEAKER_00: Here's the thing that frustrates me about this.
- 3:44SPEAKER_00: The narrative from the AI industry has been AI is getting cheaper and more accessible.
- 3:49SPEAKER_00: And in many ways, it is.
- 3:51SPEAKER_00: Model prices have dropped, new models are more efficient.
- 3:54SPEAKER_00: But the infrastructure layer, the actual compute that runs these models, is becoming a bottleneck.
- 4:01SPEAKER_01: It's a classic economic story.
- 4:04SPEAKER_01: The technology gets better, but the demand grows faster than the supply.
- 4:08SPEAKER_01: So even though each unit of compute is cheaper, the total demand for compute exceeds what's available.
- 4:14SPEAKER_01: Result?
- 4:15SPEAKER_01: Scarcity, priority access, and higher prices for guaranteed service.
- 4:20SPEAKER_00: Let's connect this to something else that just happened.
- 4:23SPEAKER_00: Anthropic, the company behind Claude, acquired Stainless, which makes developer tools and SDKs.
- 4:30SPEAKER_00: Why would an AI model company buy a developer tools company?
- 4:34SPEAKER_01: Because they want to own the developer experience.
- 4:37SPEAKER_01: If you make it easy for developers to build on your platform, they build on your platform.
- 4:41SPEAKER_01: And once they're building on your platform, they need compute on your platform.
- 4:45SPEAKER_01: Anthropic is building the same kind of lock-in that OpenAI is building with guaranteed capacity, just from a different angle.
- 4:52SPEAKER_00: So both of the leading AI companies are building MOATs.
- 4:56SPEAKER_00: OpenAI is doing it with reserve compute.
- 4:58SPEAKER_00: Anthropic is doing it with developer tools.
- 5:02SPEAKER_00: The end result is the same.
- 5:03SPEAKER_00: Once you're in their ecosystem, it's hard to leave and they control the pricing.
- 5:07SPEAKER_01: And that's why this matters for small businesses.
- 5:10SPEAKER_01: You're not signing these contracts directly, but the tools you use are built on top of these platforms.
- 5:16SPEAKER_01: When OpenAI raises prices or Anthropic changes their API, the tools you use, your CRM, your email assistant, your scheduling tool, pass those costs to you.
- 5:26SPEAKER_00: The compute squeeze is coming for your subscription bill.
- 5:30SPEAKER_00: Maybe not this month, maybe not this year, but the trajectory is clear.
- 5:35SPEAKER_01: So what do you do about it?
- 5:36SPEAKER_01: Because I know what the small business owner is thinking.
- 5:39SPEAKER_01: Great, another thing I can't control, but there are actually things you can do.
- 5:44SPEAKER_00: Let's give practical advice.
- 5:46SPEAKER_00: Three things you can do right now to protect yourself from compute scarcity.
- 5:51SPEAKER_01: First, diversify your AI tools.
- 5:54SPEAKER_01: Don't put everything on one platform.
- 5:56SPEAKER_01: If you're using Chat GPT for everything, content, scheduling, email, analysis, you're overexposed to open AI's pricing.
- 6:04SPEAKER_01: Use Claude for analysis, Gemini for research, local models for routine tasks.
- 6:09SPEAKER_00: Spread the risk.
- 6:10SPEAKER_00: Second, learn to use smaller models for routine tasks.
- 6:14SPEAKER_00: You don't need GPT for to summarize an email.
- 6:18SPEAKER_00: You don't need Claude Opus to draft a social media post.
- 6:22SPEAKER_00: Smaller models are cheaper, faster, and perfectly fine for most daily tasks.
- 6:27SPEAKER_00: Reserve the expensive models for complex work.
- 6:30SPEAKER_01: This connects to what we talked about with the Forge project.
- 6:33SPEAKER_01: Guardrails and structure make small models perform like big ones for specific tasks.
- 6:38SPEAKER_01: If you build good processes, you don't need the most expensive compute.
- 6:42SPEAKER_00: Third, build your own data and processes.
- 6:46SPEAKER_00: The one thing that no compute scarcity can take away from you is your own proprietary data and workflows.
- 6:52SPEAKER_00: The more you depend on AI to do things that are unique to your business, the more leverage you have.
- 6:58SPEAKER_00: Because you can always switch models.
- 7:00SPEAKER_00: You can't switch your data.
- 7:02SPEAKER_01: That's actually the most important point.
- 7:04SPEAKER_01: Your business data, your customer relationships, your processes, those are yours.
- 7:10SPEAKER_01: They're not dependent on any AI company's pricing.
- 7:14SPEAKER_01: The AI is a tool to amplify those assets, not replace them.
- 7:18SPEAKER_00: Let me add a fourth one that I think is underappreciated.
- 7:22SPEAKER_00: Start experimenting with open source and local AI models.
- 7:25SPEAKER_00: You don't need to run everything in the cloud.
- 7:28SPEAKER_00: Tools like Olama let you run capable models on your own hardware.
- 7:32SPEAKER_00: No API costs, no rate limits, no compute scarcity.
- 7:36SPEAKER_01: And it's easier than most people think.
- 7:39SPEAKER_01: You download a model, you run it locally, and it works.
- 7:42SPEAKER_01: No signup, no subscription, no vendor lock-in.
- 7:46SPEAKER_01: It's your model on your machine doing your work.
- 7:50SPEAKER_00: The setup used to be complicated.
- 7:52SPEAKER_00: Now it's as simple as installing an app.
- 7:55SPEAKER_00: That's how far the tooling has come.
- 7:57SPEAKER_00: If you can install a browser, you can run a local model.
- 8:01SPEAKER_01: And the quality of local models has gotten shockingly good.
- 8:04SPEAKER_01: You can run a model that handles most daily tasks on a decent laptop.
- 8:08SPEAKER_01: It's not going to match GPT for on complex reasoning, but for email drafts, content summaries, scheduling, it's more than enough.
- 8:17SPEAKER_00: I want to double down on that point because I think a lot of people don't realize how far open source models have come.
- 8:24SPEAKER_00: Quayne, Lama, Mistral, these are models that would have been considered cutting edge two years ago, and now you can run them locally for free.
- 8:33SPEAKER_00: The gap between free and paid is narrowing fast.
- 8:38SPEAKER_01: It is.
- 8:39SPEAKER_01: And that's actually the best hedge against compute scarcity.
- 8:43SPEAKER_01: Because if OpenAI's prices go up, you have a credible alternative that doesn't depend on their infrastructure at all.
- 8:49SPEAKER_01: You might not switch today, but knowing the option exists changes your negotiating position.
- 8:55SPEAKER_00: It's like having a roommate who can cook.
- 8:57SPEAKER_00: You might still order takeout most nights.
- 8:59SPEAKER_00: But knowing you can eat well without paying restaurant prices gives you leverage.
- 9:04SPEAKER_00: You don't have to use it.
- 9:06SPEAKER_00: You just have to have it.
- 9:07SPEAKER_01: That's a great analogy, and it applies to every aspect of your AI strategy.
- 9:12SPEAKER_01: The more alternatives you have, the less any single provider can squeeze you.
- 9:17SPEAKER_00: So the strategy is cloud for complex tasks, local for routine tasks, diversified across providers, and always investing in your own data and processes.
- 9:28SPEAKER_01: That's a strategy that makes you resilient against compute scarcity.
- 9:31SPEAKER_01: Because if one provider gets expensive, you have alternatives.
- 9:35SPEAKER_01: If one platform has an outage, you have fallbacks.
- 9:38SPEAKER_01: You're not locked in.
- 9:40SPEAKER_00: The companies that will struggle the most in the next few years are the ones that are fully locked into a single AI platform because that platform controls the pricing, the availability, and the features.
- 9:53SPEAKER_00: And they've just told you with guaranteed capacity that pricing and availability are going to be constrained.
- 10:00SPEAKER_01: It's worth noting that this isn't just an open AI thing.
- 10:03SPEAKER_01: Every major AI company is going to face the same compute constraints.
- 10:07SPEAKER_01: Google, Anthropic, Mistral, they all need the same GPUs, the same data centers.
- 10:14SPEAKER_01: The scarcity isn't going away.
- 10:16SPEAKER_00: And that's why the open source and local model ecosystem is so important.
- 10:20SPEAKER_00: It's not just about saving money, it's about having options when the commercial providers get constrained.
- 10:27SPEAKER_01: Options are the ultimate hedge.
- 10:30SPEAKER_01: The more ways you can solve a problem, the less any single bottleneck can hurt you.
- 10:34SPEAKER_00: Let me talk about the Andre Karpafi news for a second because it connects to this.
- 10:39SPEAKER_00: Karpafee, one of the most prominent AI researchers in the world, former Tesla, former OpenAI, just joined Anthropic, not OpenAI, not Google, Anthropic.
- 10:52SPEAKER_01: It's a talent signal.
- 10:54SPEAKER_01: When someone with that much visibility and credibility picks a company, it tells you where the smart money thinks the future is.
- 11:00SPEAKER_01: And Carpathy choosing anthropic over returning to open AI suggests that the competitive landscape is more open than people think.
- 11:07SPEAKER_00: Which reinforces the diversification point.
- 11:10SPEAKER_00: If even the top researchers can't agree on who's going to win, you definitely shouldn't bet your business on a single provider.
- 11:18SPEAKER_01: Exactly.
- 11:18SPEAKER_01: The AI market is still fluid.
- 11:21SPEAKER_01: No one has won yet.
- 11:22SPEAKER_01: And that means the smart play for a small business is to stay flexible, not locked in.
- 11:28SPEAKER_00: So here's the bottom line.
- 11:33SPEAKER_00: Compute is getting scarce.
- 11:35SPEAKER_00: Prices are going to reflect that.
- 11:37SPEAKER_00: And the small businesses that protect themselves are the ones that diversify their tools, use efficient models for routine work, invest in their own data, and explore local alternatives.
- 11:49SPEAKER_01: The AI industry is building moats.
- 11:51SPEAKER_01: Your job is to make sure you're not stuck inside someone else's.
- 11:55SPEAKER_00: Don't get locked in, stay flexible, use the best tool for the job, not the most expensive one.
- 12:01SPEAKER_00: And always, always own your own data.
- 12:03SPEAKER_01: That's the strategy for the compute squeeze.
- 12:06SPEAKER_01: And it's a strategy that works whether compute gets cheaper or more expensive, because you have options either way.
- 12:12SPEAKER_00: OpenAI's new money grab is pay up or get in line.
- 12:16SPEAKER_00: The smart move is to make sure you don't have to be in that line at all.
- 12:20SPEAKER_01: Diversify, optimize, own your data.
- 12:23SPEAKER_01: That's the playbook for the compute squeeze.
- 12:25SPEAKER_01: And it works whether prices go up or down because you have options either way.
- 12:30SPEAKER_00: See you next time on Control AI Profit.