Imagine you’re a CTO at a fast-growing SaaS company. You’ve been paying OpenAI thousands of dollars a month for API access. Your team has built workflows around GPT-4. Then you hear about DeepSeek R1—a Chinese open source model that scores nearly as well on benchmarks, costs nothing to run on your own hardware, and gives you full control over your data. What do you do?
You ditch the middleman.
That’s the nightmare keeping AI executives up at night—but it’s not the nightmare they talk about. They talk about compute shortages, investor pressure, and regulatory hurdles. They talk about Chinese censorship and data security. They talk about anything except the one thing that actually matters: their own customers are about to bypass them entirely.
This isn’t speculation. It’s already happening. A top comment on a recent Hacker News discussion put it bluntly: “Why pay more for a middleman if you can go directly to the source? The problem for Anthropic and OpenAI is to devise some political scheme to stop them.”
Let that sink in. The very companies positioning themselves as the gatekeepers of advanced AI—the ones who have spent billions building moats around proprietary models—are facing a threat that has nothing to do with Chinese regulations or Western sanctions. The threat is disintermediation. Pure and simple.
Think about it. OpenAI and Anthropic don’t sell AI. They sell access to AI. They are the toll booths on the highway of intelligence. But when the highway itself is open source, the toll booth becomes optional. Chinese open source models like DeepSeek, Qwen, and Yi are not just competitors—they are the infrastructure. And your clients can run that infrastructure themselves.
You’ve probably noticed the pattern before. It happened in travel: Expedia and Booking.com were the middlemen, then hotels started direct booking. It happened in media: Netflix and Spotify were the aggregators, then creators went direct-to-consumer. And it’s happening in AI right now, faster than anyone wants to admit. The middleman always gets squeezed out when the source becomes directly accessible.
So why aren’t Anthropic and OpenAI panicking? Because they’re stuck in the same trap that kills every platform: they believe their proprietary moat is stronger than it is. They’ve convinced themselves that their models are so superior, so fine-tuned, so aligned that no open source alternative can compete. But that’s a lie they tell themselves to sleep at night.
DeepSeek R1 matches GPT-4 on several reasoning benchmarks. Qwen 2.5 beats Claude on coding tasks. And these models are not locked behind API gates—they are downloadable, deployable, and modifiable. Any company with a GPU cluster can run them. Any startup with a cloud budget can fine-tune them. And any CTO with a spreadsheet can calculate the savings: 10x cheaper, 100x more control.
This is the twist that changes everything. The conventional wisdom says the biggest barrier to offering Chinese models is geopolitical risk or government pressure. But the real barrier is that offering Chinese models would accelerate the commoditization of AI, stripping away the very proprietary moat that justifies the premium pricing. The incumbents can’t afford to legitimize the competition—even if it means losing the market.
So they’ll do what every threatened middleman does: they’ll try to regulate. They’ll lobby for export controls, security reviews, and licensing requirements. They’ll frame Chinese open source as a national security risk. They’ll do everything except build a better product, because a better product doesn’t protect their margins.
But here’s the thing about open source: it doesn’t care about your political schemes. It spreads. It improves. It gets adopted. And the companies that embrace it—not fight it—will be the ones that survive. The only way to win the AI game is to stop being the toll booth and start being the highway.
For the CTOs reading this, the message is simple: your AI strategy should not depend on a single vendor. Start experimenting with open source models today. Build the infrastructure to run them yourself. Because the day your competitor cuts their AI costs by 90% and ships faster, you’ll wish you had.
For the AI executives reading this: your customers are not loyal. They are rational. And rationality says: if you can get the same intelligence for free, you will. The only question is whether you’ll be the one offering it.
FAQ
Q: Isn't this just a theoretical risk? Couldn't OpenAI or Anthropic just block Chinese models from their platform?
A: This isn't theoretical. Many companies are already experimenting with open source models like DeepSeek and Qwen. The risk is that clients will run these models on their own infrastructure, completely bypassing the incumbents. Blocking doesn't help when the alternative is self-hosted.
Q: What can OpenAI or Anthropic actually do to prevent disintermediation?
A: They have two options: embrace open source themselves (e.g., release their own base models freely and monetize services on top), or build deeper moats like proprietary data pipelines, fine-tuning expertise, and integrated workflows that are hard to replicate. The first option is risky but sustainable; the second is a delaying tactic.
Q: Why would a company risk using Chinese models given geopolitical tensions and data security concerns?
A: Many companies are already using Chinese open source models because they are auditable, self-hosted, and not subject to API-level surveillance. The risk of data leakage is lower when you control the hardware. Plus, the cost savings and performance are compelling enough to override political concerns for most businesses.