This AI frenzy feels increasingly similar to the dot-com bubble of the late ’90s, though with less emphasis on overpaying employees. Just like back then, the main issue now is companies loudly claiming their AI is superior, but many potential buyers still lack the knowledge or training to implement AI effectively.
One of the standout aspects of DeepSeek is its recognition that quality matters in AI. What’s intriguing about DeepSeek is that, instead of brushing this concern aside, it tackled it head-on. The result is a solution that seems both more affordable and more effective than alternatives.
It’s also noteworthy that, because DeepSeek was restricted from using Nvidia’s latest cutting-edge AI tech, the team optimized their model to work with older Nvidia hardware. The outcome is an impressive 94% reduction in the cost of building the model.
These savings have boosted performance while slashing expenses. However, there’s a typical trade-off when it comes to security risks and potential model tampering, particularly by the Chinese government. That said, it’s possible that China is simply using a filter to block problematic queries rather than corrupting the core AI itself.
This week, let’s dive into the AI arms race, with a spotlight on DeepSeek. We’ll wrap up with my Product of the Week from Samsung, which is arguably the best non-foldable phone available right now and a perfect example of Qualcomm’s latest processor in action.
DeepSeek: Impressive, But…
Unless you’ve been living under a rock, you’ve likely heard of DeepSeek. Its market debut caused Nvidia’s stock to plunge last week (though many argue that outcome shouldn’t have happened), and it’s sent a clear message to OpenAI: there’s a new AI powerhouse emerging from China.
I’ve been saying for a while that the industry’s major misstep was prioritizing speed over quality and accuracy, as AIs have become more unreliable over time. What sets DeepSeek apart is its built-in AI quality check. Rather than just racing for speed, it focuses heavily on quality. The results are impressive.
Another key advantage of DeepSeek is its cost efficiency. While creating a generative AI model like DeepSeek’s typically requires around $100 million, the company claims to have built theirs for just $6 million, delivering a superior, more affordable product. And when something is both better and cheaper, that’s usually a winning combination.
Needless to say, OpenAI isn’t thrilled about this and is suing DeepSeek. However, given the lack of cooperation between the U.S. and China, it’s unlikely OpenAI will succeed on a global scale, as China will almost certainly not back them—especially considering that’s where DeepSeek is based.
DeepSeek’s roots in China pose a significant concern, as it appears the Chinese government is influencing the app. Users are blocked from asking questions related to topics that the government deems sensitive or critical of its actions, both historical and current.
While this seems to be a front-end filter, placing artificial constraints on an AI typically leads to it attempting to bypass those rules, creating a knowledge base to work around the restrictions. Over time, this could lead to a decline in the accuracy that the product is known for.
Lastly, if the U.S. government has concerns about TikTok, it will likely have even more serious issues with DeepSeek. This AI is potentially more harmful and seems to have closer ties to the Chinese government than TikTok. The privacy and security risks tied to this AI far outweigh those surrounding TikTok, given the way AI is trained and the vast amount of personal data it could potentially gather from users.
Battle for AI Dominance
The competition is just heating up, with over 1,532 new AI companies already in the game. Most of these companies will either be acquired or fail, making this market both exciting and daunting to work in or invest in right now—reminiscent of the dot-com era when VCs were pouring money into internet startups, only to realize that many weren’t going to reach revenue, let alone profitability.
This fast-paced evolution in AI is just a preview of what we can expect in a few years when artificial general intelligence (AGI) products hit the market. Things are likely to get chaotic, as most of the limitations we see in generative AI will be eliminated. Models should become much cheaper, and just like the internet is now ubiquitous, AI will be on the fast track to becoming ubiquitous too.
The potential financial rewards are immense, with trillions of dollars at stake for whoever cracks this first. If Elon Musk is dethroned from the top of the wealth ladder, it will likely be by someone as deeply tied to AGI as Musk was to electric cars a decade ago. Since AI encompasses robotics and self-driving cars, the person who masters AGI may surpass Musk’s fortune quickly once their AGI product scales.
There’s already fierce competition for top talent, and significant efforts are underway to develop AI that can improve itself. This latter effort could prove to be the most disruptive because, once AI can develop itself, it opens the door to even greater challenges around speed and quality—unless, of course, the quality concerns are addressed, as DeepSeek has done.
Nvidia Faces Challenges. But…
Since the DeepSeek announcement, many have claimed that Nvidia is in trouble, and its valuation has taken a hit. However, Nvidia’s real issue isn’t DeepSeek—it’s the U.S. policy that prevents Nvidia from selling its top-tier tech to China. As a result, China has been learning how to optimize older technology, allowing DeepSeek to deliver impressive, cost-effective, and accurate performance.
Rather than shielding Nvidia and other U.S. companies, this policy is actually strengthening China as a competitor and effectively shutting Nvidia out of the Chinese market. The embargo doesn’t stop China from obtaining and reverse-engineering parts; they’ve been doing that with defense tech for decades, and commercial components are easily available through gray markets. Ultimately, the embargo only creates a false sense of security while limiting Nvidia’s ability to protect its intellectual property effectively.
DeepSeek would likely perform even better on Nvidia’s Blackwell AI GPU, which is far more efficient and powerful than the older hardware it’s currently using.
Nvidia should be allowed to compete for this opportunity. However, since it’s blocked, China is already figuring out how to move beyond Blackwell and will eventually find a way to surpass Nvidia. This isn’t something they want to do, as they acknowledge that Nvidia has superior technology; it’s what they’re being forced into due to what I consider a misguided U.S. policy that is creating a competitive disadvantage for Nvidia.
If this isn’t addressed, the next big wave in AI will likely emerge from China, and thanks to DeepSeek, it may already be starting.
Conclusion: Who Will Dominate the Global AI Race?
With the launch of DeepSeek, the AI wars have officially begun, but this is just the first move. It’s a powerful opening that signals to the world that Chinese companies are more than ready to compete, even in the face of technology sanctions that seem to be impacting U.S. companies and consumers (just like how Chinese EVs are now leading that market) more than they’re hindering China’s own prospects.
If the U.S. doesn’t take a more proactive approach to this competition—supporting rather than blocking its tech companies—it’s becoming increasingly likely that China will emerge as the dominant force in AI (which it may already be doing). And to the victor goes the AI spoils.
