When news of DeepSeek hit recently, friends and family came to me asking: What does this mean? Is this the end of NVIDIA? Is AI taking over? Should I be worried about my job? The stock market, predictably, went into a frenzy.
But taking a step back, the real story behind this isn’t about the “fall of NVIDIA” or some existential AI threat. The real story is that AI development just got a lot more accessible which is great for consumers and developers —and that changes everything.
DeepSeek
First – what is DeepSeek? (I thought this was a good primer from Stratechery.)
DeepSeek is an AI model that came about as a side project from a company in China. It competes with the likes of OpenAI and Anthropi, and it is open source. But two things made DeepSeek particularly interesting:
- It claimed to train its final model at a fraction of the cost that OpenAI and Anthropic have spent training their models. Whether that’s entirely true is debatable, and many people on the internet think they’re lying or that they’re hiding something, but whether they’re right or wrong actually doesn’t matter IMO.
- It was built with NVIDIA’s H800 GPU chips – not their most powerful chips.
This second point is where things get interesting. For years, NVIDIA’s A100 and H100 GPUs have been considered gold and necessary for AI training—so much so that their availability (or lack thereof) has shaped the AI industry. I have portfolio companies who had been clamoring for chips and couldn’t get any. But DeepSeek proved that they didn’t need NVIDIA’s most powerful chips (they chips they used had the same level of computer power but less bandwidth). In fact, because they are based in China, where it’s impossible to get access to A100 and H100 GPUs, they were able to train their model on less powerful computer chips, which was considered near-impossible. This is why NVIDIA’s stock price is down – no longer do AI companies need their most powerful chips if you can just take a page out of Deepseek’s playbook.
I think, for NVIDIA, though, this doesn’t mean extinction. Powerful chips will always be in demand, and people will just use them to do more. (This is not investment advice!) But DeepSeek’s success opens up a whole new market. It was previously thought that if you want to develop expensive AI models, you would not only need tech talent but also a lot of money to buy compute power. No longer is that true, and NVIDIA is no longer the gatekeeper.
The Real Winners: Developers and Entrepreneurs
For the first time, any developer can, in theory, develop new AI models. This is a game available to anyone—not just deep-pocketed tech giants. Open-source combined with less expensive infrastructure means that a new wave of developers and startups can build AI-models without requiring hundreds of millions in funding.
But the broader trend of decreasing costs has been happening for decades already. In the 90s, launching a tech company required massive capital. Startups were building their own servers in closets. But when cloud computing came along, AWS made infrastructure cheap and accessible. No longer did you need to raise $5m to own and run your own infrastructure. Now, AI is hitting a similar inflection point. Developers can develop AI models without NVIDIA’s chips, and in many cases, without raising millions in VC money. That means startups can bootstrap or seed-strap in ways they never could before.
In fact, we’ve been seeing this trend for a while now. Many of today’s AI entrepreneurs aren’t raising huge rounds of funding. Instead, they’re launching lean, capital-efficient businesses that can reach $1M-$2M in revenue without taking on significant outside investment. This is a fundamental shift. AI isn’t just for the big players anymore.
The Real Losers: Venture Capitalists
Ironically, the group that should be most worried isn’t NVIDIA, other big tech companies, AI startups, or independent developers—it’s VCs.
For decades, venture capital thrived on the high cost of starting a company. In the early internet days, founders needed millions to build and scale — those servers were not cheap.
But the cost to build a business has been coming down over the years. But VCs were still needed, because tech became a huge roaring industry, and there were not enough engineers to support the industry. The cost to hire engineers in Silicon Valley zoomed up. There was more demand for engineering talent than supply. And this is what has kept venture capitalists in business.
But then a decade of coding bootcamps, a new generation of students entering computer science in droves, and the proliferation of no-code tools have brought that cost down again. These days, many of my founders are using AI tools to write code for them. For most software companies, you don’t need specialized computer software knowledge to build a multi-million dollar business.
The initial wave of AI put a momentary blip in that trend, because so much capital was required to train AI models. But now, when it no longer takes a ton of money to build a viable AI startup, what happens to the venture capital industry?
- Startups don’t need as much money. AI tools make it easier for small teams (sometimes just one or two people) to launch and scale products. People are using AI tools to write code and increase productivity.
- The infrastructure costs have now decreased. We saw this with server costs in the early internet wave, and now we’re seeing this with AI training costs.
- You don’t even need to be an engineer. Many of these tools are so user-friendly these days that you don’t even need to be an engineer by training to build and run a software company.
So what ends up happening with the confluence of these trends?
- Markets are more crowded. With lower barriers to entry, there’s more competition. If 200 startups are building the same AI-powered tool, it’s hard for one to achieve dominance—and hard for a VC to get a 100x return. But, founders can still make money. You can still have 200 companies in a busy space, where even if you’re in the long tail, you can make millions of dollars per year and do well for yourself. Moreover, these outcomes for founders would probably be similar to building a billion dollar business with venture capital money and taking a small portion of that exit home. That’s great for entrepreneurs! Just not great for VCs.
- Companies struggle to find moats. If anyone can spin up an AI-powered product in a weekend, it’s difficult to build a defensible business. This can actually be ok if you’re a 1-2 person shop. You’ll still have business even if retention isn’t perfect. But again, this isn’t a good situation for a VC to invest in.
Founders can still make great money—but VCs are finding it harder to generate the outsized returns they depend on.
Where Investors Can Still Win
So, if venture capital is struggling in this new landscape, where does investment still make sense in software?
- Super-early-stage bets: While it’s getting cheaper to launch a startup, some founders still need pre-seed capital—particularly those who haven’t yet reached revenue yet. Many founders will need some level of capital to survive and experiment with before they get to ramen profitability. (I’m obviously talking my own book here.)
- Big, capital-intensive ideas: While small AI startups are thriving, some ambitious projects that go beyond software still require massive funding. Think AI hardware, biotech, space, or deep-tech startups where money itself is one lever of a competitive advantage. This is a great place for large VCs to play. Very few competitors can go after these opportunities and they can be defensible because of the capital and knowledge moat.
- Uniquely defensible businesses that have deep ties into workflow. There are going to be software businesses that are so entrenched in workflow that even if they are copied, the sheer distribution edge is enough to win. However, many of these opportunities will likely reside with existing large software incumbents.
- International opportunities. US investors have long shied away from global markets, because they don’t understand them. But ironically, this is where the greenfield opportunity lies. You can find great companies that have limited competition and favorable valuations. And, now that companies require less capital for software companies, the lack of capital in these regions is a much much lower risk.
The Future: AI Everywhere
DeepSeek is just the beginning. AI is about to become pervasive.
The cost of building AI-powered products is dropping fast, and that means AI isn’t just something that happens in big tech labs—it’s something that happens everywhere.
- Developers will integrate AI into everyday applications, making businesses and workflows dramatically more efficient.
- AI-powered startups will proliferate globally. This has already been happening for a while.
- There’ll be a lot of 1-2 person startups, and some of them will become massive with very limited capital raised. This is where we’ll see the billion dollar single-founder startups emerge.
Yes, NVIDIA will continue to sell chips. Yes, OpenAI and Anthropic will continue improving their models and will lose pricing power, because they will have serious competition. But, they will all be fine. Consumers will be great. AI development is no longer restricted to the elite few. We’re just at the beginning and that’s pretty exciting.