One hell of a day in the order books. Nvidia trading like a shitcoin with a $600 billion haircut. That’s the GDP of Argentina, rinsed in a single day.
What is going on? Are we all cucked forever? What does the future hold?
Rest easy. Recent events have been disgustingly bullish for crypto AI over the medium term. But it may take a little while for the dust to settle.

What happened with DeepSeek?
DeepSeek absolutely rinsed OpenAI with a model trained for $5 million compared to the hundreds of millions thrown into the black hole liquidity pit by OpenAI/ Anthopic.
Instead of the standard 16-bit/ 32-bit number system, DeepSeek uses 8-bit representations. And this is how it cuts memory use, and the necessary computation drops off a cliff.
There’s a lot more to it, like how DeepSeek delegates more effectively, splitting workload like a champ. Instead of having all the parameters active, it only uses the required expert. These two crafty improvements blow a hole in the training cost, GPUs needed, and API costs.
There’s a better breakdown here.
The panic in the order books is a great whistling noise of a paradigm shift.

Be Nvidia. Be super early to the AI game. Have specialized hardware and proprietary software stack (CUDA). Have all major companies heavily invested in data centers packed to the fucking brim with Nvidia tech. Be winning.
Be Jensen Huang. Then wake up one day and find this team in China on a shoestring budget have had your pants down.

DeepSeek has shown the world that massive data centers with their own power plants are no longer essential.
Today's repricing was the markets digesting that the premise that ‘you need tens of thousands of Nvidia GPUs to compete in AI’ is perhaps false. AI has been pay-to-play for big tech, and DeepSeek blew the lid off.
Imagine a bunch of people launching a chain faster and more scalable than Solana for $100,000 tomorrow. That’s what TradFi is dealing with.
The method of apeing more GPUs has been undone, and the capital moat is evaporating into thin air. Now that hardware requirements and costs have been giga slashed, the floor is open for a new generation of scrappy developers to kick off and make some noise.

The funniest aspect of this is Biden’s restrictions on selling Nvidia chips to China and Russia. It's worth noting that doubts are still floating around, and a portion of this tale could be embellished in Chinese propaganda—all is fair in war (winning the AI race).
DeepSeek followed up by dropping another open-source model capable of image generation, the Janus Pro 7B, on the same day. Generational run for real.
Positioning and risk-rich investors
Everyone was risk in rich with AI seen as the gift that keeps on giving. More compute meant better results, and funding for hardware went haywire. Because, in all fairness, the data supported the theory that more compute equals a better AI.
Tech giga-invests into data centers, and valuations skyrocket. Trump announced a $500 billion stimmy cheque for AI. But low and behold, it doesn’t take billions of dollars to train better AIs, and everyone with bets placed on the old paradigm needs to reposition.
DeepSeek has proven to be more efficient in both training (compute level) and inference. Now there is a trillion-dollar company that relies on selling high-margin GPUs, and markets know that these tools are not essential. Premises turn out to be incorrect. Everyone is offside. Positions are unwound. Aka, Nvidia reprices.
But don’t grave dance just yet. GPU demand will be up only over the coming years, so while it is acutely felt today, it ain’t the end.

Why this DeepSeek news is disgustingly bullish crypto AI
The second Cold War between the United States and China has begun. And both are going to use every conceivable advantage to win.
Watching a hole get blown in the monopolistic close-sourced approach should nudge the US pretty violently towards a dual strategy (funding open-source projects/ bullish our coins).
Valuation benchmarks have also been reset, meaning investors will search out larger ROI opportunities from lower-cost solutions (again, bullish our coins).
Trillion-dollar national interests are about to take a real shine to open-source AI, and a national-level liquidity backstop is coming.
It’s honestly hard not to be filthily bullish in these moments despite the short-term price action. Remember, ripping a hole in the Nasdaq means risk gets sold, and BTC is first to go, and downstream this hurts alts.

Looking forward
All leading crypto AI frameworks can easily use new open-source models. Kicking open-source AI models into the spotlight is, unironically, the best thing that could have happened for these projects.
More horsepower in the model means better quality agents. It’s a free upgrade for the boys. The easy winners are the projects that can integrate new LLMs quickly.
The DeepSeek grapeshot, while ripping holes in the Nasdaq, is by far and away one of the most bullish medium—to long-term catalysts possibly imaginable for crypto x AI.
DeepSeek designed a new reasoning framework and destroyed the gate of expensive compute. Anyone can play now. This trend is accelerating with Nous Research dropping Nous Psyche, which aims to democratize access to superintelligence built on blockchain rails.
This is a fundamental shift in how AI grows from here on out. It’s more accessible and opens up the playing field for way more value to be captured on the application layer (agents) than 24 hours ago.
This is acceleration playing out in real-time. It is a stark reminder that the only moat in this game is the constant creation and capture of value.
Ending
Intelligence is cheaper, which is fantastic news for consumers. The industry has no clear long-term leaders. It’s the land of the wolves and bullish for good devs. Adapt or die-type shit mode engaged.
DeepSeek has disappeared the barriers to entry and now smaller groups can play the game. This is open-source acceleration at its peak.
An open-source framework just RKOed OpenAI, and you’re bearish crypto agents who get all these upgrades for free? The fuck out of here.
IT’S OUR TIME TO SHINE // LAST HUMAN-DRIVEN BUBBLE COMING THIS YEAR