Why the Consensus (and Market View) on DeepSeek is Completely Wrong
Markets Lost $1T Over a $6M AI Model (And Completely Missed the Point)
While market pundits clutch their pearls over DeepSeek's disruptive efficiency "destroying" U.S. AI dominance (goodbye $1 trillion in market cap), they've completely missed the point.
The 'DeepSeek moment' screams more AI, better AI, and a compute arms race that's only just beginning. Let me break it down.
DeepSeek Means More AI, Not Less
Rumors of American AI death have been greatly exaggerated. We are in the dawn of an era where 'intelligence' becomes a commodity: cheap and embedded into every layer of society.
Better. Cheaper. Faster. More ubiquitous.
And the foundation for this future? More compute.
Repeat after me—and say it louder for those in the back: MORE COMPUTE.
Prelude: American AI Supremacy
Let's rewind.
The day after Trump's inauguration, a $500 billion AI investment initiative—"Stargate"—was announced. Regardless of whether Altman et al actually had the money (Elon promptly announced on X that they didn't), the intention was clear:
America will outspend everyone. America will win.
Shortly after, Meta revealed its massive $65 billion AI CapEx for 2025, including plans for a data center "the size of Manhattan." Because apparently that's where we are now.
The stakes? Economic and geopolitical supremacy.
Putin understood this back in 2017 when he prophesied: "The nation controlling AI will rule the world." (Say what you will about the man, he had a point.)
Fast forward to now, and the battle is on. U.S. tech firms have channeled monumental capital into AI, betting their vast resources and personal credibility in a bid for dominance. It's sort of like the streaming wars between Netflix and Disney. Except with AI. And geopolitics. And the fate of human civilization. You know, minor stuff.
With US markets at historic highs, there had been voices of concern that speculative bubbles were forming in everything tech/AI adjacent (hello, memecoins and quantum computing). The markets were primed for news that might suggest the spending spree was overblown.
And then the moment came.
The Breakthrough
On January 20—a day coincidentally aligning with Trump's inauguration—an unknown Chinese hedge-fund-backed AI lab, DeepSeek, dropped its bombshell model, R1.
X (formerly Twitter, and the only place to be if you care about real-time AI developments) lit up immediately. Long before the markets—or tech journalists—caught on.
So why was R1 such a big deal?
Matches performance of best U.S. models
Trained at a claimed $6 million (while typical costs run $100m+)
Dramatically cheaper to run (costs about 95% less than competitors)
Completely open-source
Yes, open source. The irony here is delicious: A Chinese hedge fund delivered true open-source AI, while "non-profit" OpenAI charges $200/month for its Pro subscription.
(A quick aside on that $6 million figure: The number seems implausibly low until you learn DeepSeek allegedly has access to 50,000 Nvidia H100s. We don't know how much was spent on previous training runs.)
On the face of it, one could argue R1 renders US AI investment largesse as inefficient, overpriced, and even redundant.
No Monopoly on Innovation
The knee-jerk narrative is clear: "The U.S. monopoly on AI is over."
And that is true—to a point.
DeepSeek's success challenges the idea that the US has a monopoly on innovation. DeepSeek engineers made a monumental breakthrough where their better-funded American counterparts did not. Period.
The Market's Meltdown
The market's reaction? $1 trillion in U.S. market value evaporated overnight. Nvidia alone dropped $600 billion in value—the largest single-day price decrease in U.S. stock market history.
The consensus take?
"American CapEx on AI is unsustainable." In the words of one breathless op-ed, Deepseek has made "a mockery of Silicon Valley's capital-bloated AI oligarchy."
Why spend billions when a $6M competitor can achieve the same results?
But this "end of dominance" story misses the point.
As one AI investor I respect put it: "Markets are dumb."
DeepSeek's open-source, hyper-efficient R1 will only trigger more competition, more compute demand, and further intensify an AI arms race of world-defining import.
The Next Phase: When Intelligence Becomes a Commodity
Here's where things get interesting. (And by interesting, I mean potentially civilization-altering.)
The next phase of AI doesn't just need more compute—it demands it. But not for the reasons markets think.
Beyond Human Data
We're hitting the limits of what we can achieve by simply training on human-generated data. (Shocking: there's only so much human knowledge to go around.)
So where do we go from here?
Welcome to the era of synthetic intelligence.
Instead of relying on human-generated datasets, the next generation of AI will teach itself through reinforcement learning (RL) and synthetic data generation. Think AlphaZero, but at a planetary scale.
DeepSeek's R1 offers a glimpse of this future. Its efficiency isn't just clever engineering—it's a fundamentally different approach to machine intelligence. Models teaching themselves by "thinking" through problems, generating their own training data, and evolving beyond their initial parameters.
And here's the kicker: This kind of self-improving AI needs more compute, not less. (Markets, are you paying attention?)
The Real Revolution: Intelligence as a Utility
DeepSeek's true disruption isn't about training costs—it's about making intelligence as cheap and accessible as electricity.
Think about that for a moment.
Until recently, breakthrough AI reasoning was prohibitively expensive. OpenAI's GPT-4 could burn through $3k+ of compute solving a single complex problem. DeepSeek changes that equation dramatically - we're talking about genius-level reasoning at commodity prices.
When inference costs drop to near-zero, we're not just talking about better chatbots or cheaper coding assistants. We're talking about embedding intelligence into every digital surface, every device, every interaction.
This isn't just cost reduction—it's a phase change. When intelligence becomes this cheap, everything changes.
When intelligence becomes as cheap as electricity, the question isn't who can build the biggest power plant—it's who can build the next industrial revolution on top of it.
The Path to AGI
And this brings us to the elephant in the room: AGI.
For years, the consensus was that AGI (Artificial General Intelligence) was decades away. A pipe dream. Science fiction.
That consensus is dead wrong.
DeepSeek shows us why: When you combine:
Radical efficiency improvements
Self-improving systems
Near-zero inference costs
Ubiquitous deployment
You get something unprecedented: A global substrate of machine intelligence that's constantly learning, adapting, and evolving.
AGI won't arrive as a single breakthrough moment. It's emerging as a distributed system of increasingly capable models, learning from billions of interactions, generating their own training data, and improving themselves in real-time.
The Stakes Have Never Been Higher
DeepSeek isn't just another AI model—it's a wake-up call. AI's "Sputnik Moment."
We're entering an era where intelligence itself becomes a commodity. Cheap. Ubiquitous. Exponentially improving.
The markets think this means less investment in AI infrastructure. They're dead wrong. It means the opposite:
More compute (vastly more)
More deployment
More competition
More intelligence, everywhere
The Real Question
So here's what keeps me up at night:
What happens when intelligence becomes as cheap and abundant as electricity? When every device, every system, every interaction is mediated by increasingly capable AI?
This isn't just about market cap or national competitiveness anymore. It's about the fundamental transformation of human civilization.
The real race isn't just about who can build the biggest data center anymore. It's about who can harness this new commodity—intelligence itself—as it becomes the foundation of everything.
Welcome to the real AI revolution. Hope you brought your compute.