This past weekend, a buddy of mine lost access to ChatGPT. No warning, no explanation—just a corporate silence that left his 125+ employee firm in the lurch. They were early AI adopters, heavy users in marketing services, and the outage sent them scrambling for alternatives. No violations were reported, service was restored, but the message was clear: something, someone, somewhere decided their pipe needed cinching.

Besides the lost queries, their fallback options were the usual suspects: Gemini, Perplexity, Claude. A forced consolidation of choices—exactly what Marc Andreessen has been railing against on Rogan and Lex Fridman. The Biden administration, under the guise of AI safety, has effectively herded AI power into a few hands, stifling competition. There’s a now-famous meme of Sam Altman being asked, “What if ten smart guys with $10M wanted to compete with OpenAI?” His answer: “I should say yes, go for it. But in reality, don’t bother—this is an arms race.”

Yet, DeepSeek arrives and the market corrects dramatically.

DeepSeek’s Disruption

DeepSeek, a Chinese AI firm spun out of a quantitative hedge fund, upended expectations by delivering open-weight models that rival top-tier American AI labs—at a fraction of the cost. Suddenly, the assumption that China lagged years behind in AI development seemed absurd. More like months. Maybe weeks. Nvidia’s stock tumbled 17%—the largest single-day loss of market cap in history. AI’s value chain suddenly felt a lot less stable.

The claim that DeepSeek trained a ChatGPT-competitive model on just $6M (compared to OpenAI’s estimated $100M for GPT-4) sent shockwaves through the industry. Was China years behind, or had American firms been running a bloated, inefficient arms race? The answer, as always, is somewhere in between.

My Experience with DeepSeek

Curious, I put DeepSeek through its paces. It was long on promises, but short on delivery. The “Can you do it?” prompts often yielded “No, actually I can’t” responses. In fairness, it hasn’t been trained on my work or writing style, but the cracks were evident.

Still, its efficiency gains were undeniable. DeepSeek’s models, particularly R1, a reasoning model competitive with OpenAI’s, sparked a gold rush in open-weight AI. Y Combinator’s Garry Tan called it the model of choice for top U.S. universities. Amazon and Microsoft integrated it into their cloud offerings overnight. The market’s shift was seismic.

The Bigger Picture: Innovation vs. Control

The U.S. approach to AI development increasingly mirrors the way it handled social media—a regulatory patchwork that lets a few giants dictate the rules. OpenAI, Anthropic, and Google Gemini operate under heavy government scrutiny. Meanwhile, DeepSeek’s open-weight release throws a wrench in this ecosystem.

But will China actually allow DeepSeek to remain open-source? The CCP has a long history of co-opting successful tech ventures for state purposes. TikTok’s success didn’t make it immune to Beijing’s directives. If DeepSeek truly becomes world-class, does it remain free to the world, or does the party tighten its grip?

What’s clear is that innovation thrives when control is loose. The AI bottleneck we’ve seen in the U.S. has less to do with safety and more with consolidation. The power to shape the future of AI has been handed to a select few, and DeepSeek’s sudden rise underscores how fragile that structure really is.

We may not know yet whether DeepSeek is the real deal, but the lesson is clear: When one company—or one government—controls access to AI, innovation chokes. And this past weekend, when my buddy’s company went dark, that reality became painfully obvious.

PS; the memes on this one are fire! [Above lots of credit Trung Phan]