China's DeepSeek has shaken up the AI world with the release of an open-source AI model that reportedly outperformed OpenAI's in several benchmarks. Even more startling is the company’s claim that its AI technology was developed for only $5.6 million.
This figure has raised eyebrows, especially as companies like OpenAI and Anthropic have spent hundreds of millions annually to develop their large-language models. Meanwhile, tech giants such as Microsoft have guided for an $80 billion expenditure in 2025, and Meta has projected spending between $6 billion and $65 billion this year, much of it directed toward Nvidia’s GPUs.
As an investor in OpenAI and Anthropic through Fundrise, as well as an owner of most of the U.S. big tech stocks, DeepSeek's performance has me intrigued.
Necessity Is the Mother of Invention
Founded in 2023 by Liang Wenfeng, a former chief at AI-driven quant hedge fund High-Flyer, DeepSeek has adopted an open-source approach to AI development. This strategy enables the global developer community to inspect, enhance, and innovate upon its software.
DeepSeek claims its R1 model matches or exceeds OpenAI and Meta’s leading products in benchmarks like AIME 2024 (mathematical tasks), MMLU (general knowledge), and AlpacaEval 2.0 (Q&A performance). It also ranks highly on UC Berkeley’s Chatbot Arena leaderboard. All of this is hard to believe with such limited resources.
The company’s mobile app, launched in early January 2025, quickly rose to the top of iPhone download charts in countries like the U.S., Australia, and the U.K. What partly sets DeepSeek apart is its AI model, R1, which explains its reasoning before delivering responses—a key differentiator from competitors like OpenAI’s ChatGPT.
How did a small startup with fewer than 200 employees and a budget that’s half of what many personal finance enthusiasts consider ideal for retirement manage to compete effectively with U.S. giants? The answer might lie in necessity. When something becomes imperative, innovation often follows.
Do Whatever It Takes To Survive
As a parallel, consider Financial Samurai—just a two-person team (my wife and me) operating on a modest budget, yet still managing to compete effectively against larger sites with teams of writers, editors, and freelancers. I wrote this article from 4:30 am to 6:15 am PST while on holiday skiing in Palisades, Lake Tahoe because, since 2009, I feel a responsibility to the Financial Samurai readership to keep you informed.
If we ever lost everything and needed to rebuild our net worth by earning millions online to support our children, I’m confident we could. I would gnaw off my arm and then gouge out my eyeballs if it meant keeping my children alive! There is nothing a parent wouldn't do for their children.
However, if Financial Samurai were to perform at the same level as DeepSeek, it would be like this site generating as much traffic as The New York Times—a media giant with ~1,700 journalists and 5,800 total employees. Such an achievement would be virtually impossible, which is why I find it hard to believe DeepSeek only spent $5.6 million without receiving substantial support from the Chinese government.
The Other Side of the Coin
U.S.-based Alex Wang, the 28-year-old CEO of Scale AI, told CNBC:
“The Chinese labs have more H100s than people think,” referring to Nvidia’s GPUs, which are restricted from export to China. “My understanding is that DeepSeek has about 50,000 H100s—which they can’t talk about, obviously, because it violates U.S. export controls.”
The logical conclusion seems to be that DeepSeek has far more resources than it’s disclosing to the public. The estimates are between $500 million – $2 billion when you count up the cost of all the GPUs.
Once the initial panic subsides, those with insider knowledge will likely reveal the true extent of DeepSeek’s capabilities and support.
What I Think Will Happen And How I Plan To Invest In An AI War
It’s clear no U.S. AI company will sit idle while their future—and fortunes—are at stake. Here’s what I predict:
- The U.S. will embrace open-source AI models for greater efficiency and faster innovation, including that of DeepSeek's. Investing in artificial intelligence by corporations and investors will continue, but at an even rapid pace.
- Nvidia and other AI chip manufacturers may face a temporary decline of up to 20-25%, followed by a rebound as AI adoption accelerates thanks to Jevon's paradox. The Jevons Paradox states that, in the long term, an increase in efficiency in resource use will generate an increase in resource consumption rather than a decrease.
- The Trump administration will take additional measures to protect the U.S. AI industry. Its announcement of a $500 billion AI infrastructure investment—led by Oracle, OpenAI, and SoftBank—signals how seriously the U.S. views this race.
- Big tech stocks like Microsoft, Meta, Amazon, and Palantir could drop by up to 10-15%, but they’ll rebound as lower AI costs lead to higher profits down the road.
Buying The Dip Due To Greater Adoption Of AI
In light of these trends, I’m buying the dip in U.S. big-cap tech stocks and private AI companies. Lower costs mean greater AI adoption and ultimately higher profitability for these companies. Alibaba on January 29, 2025, says it has released an AI model that surpasses DeepSeek. Great!
Apple stands to be one of the biggest beneficiaries of DeepSeek’s advancements. With its massive ecosystem and late entry into heavy AI CAPEX spending, Apple is well-positioned to capitalize on lower costs, increased AI adoption, higher future earnings, and improved customer satisfaction. I'm also talking my book as Apple is my largest public stock holding.
Please note, this is not investment advice for you. It simply reflects what I am doing with my own money. Risk assets carry no guarantees, and you must take full responsibility for your financial decisions.
Real Estate May Also See An Uptick in Demand
If the S&P 500 faces a prolonged 10%+ slump over the next 3–6 months, Treasury bond yields will likely decline as investors seek the safety of risk-free returns.
Lower Treasury yields would, in turn, lead to reduced mortgage rates, driving greater demand for U.S. real estate. This could remind investors of the growing disconnect between residential commercial real estate values and the stock market since early 2023. Additionally, many may revisit the idea of converting volatile stock market gains into tangible assets, like real estate, that offer both stability and everyday utility.
I’ll continue dollar-cost averaging into the S&P 500, private AI companies, big tech, and residential real estate. It's precisely during moments of market panic that disciplined investing becomes most critical. When you focus on long-term goals—whether retirement or securing a future for your children—deploying capital during downturns becomes much easier to do.
One thing is certain: the AI and investment landscapes are evolving rapidly, creating both risks and opportunities for those who stay informed.
Readers, what are your thoughts on DeepSeek's open-source AI model? Do you believe the company truly spent just $5.6 million to compete effectively with the likes of OpenAI, Anthropic, Google, and Meta? Is this the beginning of the end for the Magnificent 7, or do you think U.S. big tech will innovate their way through the challenge? And with the current panic selling, how are you deploying your capital?
Again, this article is not investment advice for you. It simply reflects what I am doing with my own money. We have completely different financial circumstances. Invest at your own risk.
Subscribe To Financial Samurai
If you're looking to gain more exposure to AI, as I am, check out Fundrise's venture capital product, which invests in private AI companies. I'm dollar-cost averaging in over the next three years. Fundrise is a long-time sponsor of Financial Samurai.
Listen and subscribe to The Financial Samurai podcast on Apple or Spotify. I interview experts in their respective fields and discuss some of the most interesting topics on this site. Your shares, ratings, and reviews are appreciated.
To expedite your journey to financial freedom, join over 60,000 others and subscribe to the free Financial Samurai newsletter. Financial Samurai is among the largest independently-owned personal finance websites, established in 2009. Everything is written based on firsthand experience and expertise because money is too important to be left up to pontification.
I used to work in Silicon Valley during the dot com bubble of 2000. On one hand, the last time I heard so much hype about a new technology was in the late 1990’s. On the other hand, AI remains a language model, meaning, it reads and compiles information but is NOT able to think. So any answer given by an LLM today is probability-based and not given from actual thinking. As such, there is no killer app in sight and there might never been an indispensable need for AI like Sam Altman is selling. The AI story has been adopted by the top tech companies for growth, but this might end up like the “Metaverse” Marc Z. was selling just a couple of years ago that failed. So while AI will have a future, the probability this is a huge tech bubble about to implode rose substantially this past week with the Deepseek business model (open source and significantly cheaper).
Good points. However, what if AGI eventually happens? Could be huge.
****
AGI stands for Artificial General Intelligence in the context of AI. It refers to a level of artificial intelligence that can understand, learn, and apply knowledge across a wide range of tasks at a human-like level. Unlike narrow AI (which is designed for specific tasks like playing chess or recognizing images), AGI would have the ability to reason, solve problems, and adapt to new situations without being explicitly programmed for each scenario.
AGI remains a theoretical concept, as current AI systems (including large language models) are still categorized as narrow AI, excelling in specialized domains but lacking true general intelligence.
Yes, AGI would be the end goal, and only then would we see mass adoption of a “killer app”. But as of now, we are not anywhere near. Sam Altman speculates that AGI will happen in less than 10 years as he benefits from the hype. But most experts don’t see it happening for at least another 20-30 years, so this current AI momentum might just be Silicon Valley year 2000 2.0. Tech companies have the mindset of growing forever, and AI is picture-perfect to sell their growth story.
I think this the so called panic has been overblown. In about two weeks time NVDA should be back near 140. A good time to add to the position.
Jevon’s Paradox is the right way to look at it. Cheaper, more efficient AI requiring less compute will massively boost applications and demand for AI, which in the long run (just a year or two) will boost demand for chips. Nvidia will be fine. This is like cheap lower-margin PCs replacing large high-margin mainframes — the computing market grew bigly after that.
No way they copied it using US open source from Meta and US chips from NVDA for 6M! Elon laughed at that. It’s not really about the result or competition. It’s all about the capex. If any of the mag 7 adjust downward, we could see things get REALLY ugly with tech stocks. Everything related to compute is getting crushed from semis to nuclear energy. The next two weeks will be telling with earnings looming and the inevitable capex inquiries/reporting. That said, hard to envision a scenario where TSLA and AMZN aren’t the two mag 7 that ultimately benefit the most from here. The potential tam for robotics and autonomy are off the chart for TSLA and hard to find a better buy than AMZN of the mag 7 in consideration of all the ways they can/will benefit from AI/robotics, etc. Plus, AWS is probably the single greatest business ever created from an earnings standpoint. Their valuation also seems cheap when layered over the other mag 7. Apple is probably the least exciting mag 7 in my view at this stage and could be part of the reason why Buffet dumped so much at the current valuation.
Deepseek has likely created another chatbot. The low training cost is impressive, but businesses need AGI to increase productivity. Meta, Microsoft, OpenAI aren’t spending billions in order to make a really good automated receptionist.
Most likely subsidized by the Chinese government-just like the Chinese government subsidizes most of their industries.
Seems to be the most likely case indeed. Big tech CEOs will be reporting soon, and I’m curious to hear about their thoughts.
This is just lazy blaming with a racist taste for good stuff made by smart and hardworking Chinese younger generations. Or claiming they steal, copy, not fair, etc. Rather than focusing on one’s own work. Typical American attitude: blaming others for own’s underahievment or mistakes. Some of us are lucky to ride with the US stock market but fundamentally nowhere in the world could have such massive number of motivated productive well-trained STEM workers as in China. I am not talking about the useless leadership and ppt skills that the US business world values but the actual ability to solve problems and get things produced. That takes decades to accumulate. I don’t see how US could compete in this level. Elon Musk tried to whip the US workers in a very toxic way but it wouldn’t work and that was why he loves and can’t live without his most productive giga factory in Shanghai.
just seems like a great buying opp to me. I’m not waiting to buy more NVDA. Starting now.
also bought more NVDA today, 50 PE and 50% expected growth rate gives 1.0 PEG ratio
Fascinating. Thanks for highlighting this news in such a timely manner as I am behind the ball. I find it hard to believe their spend is that low in comparison to the US. So I believe the conspiracy theories that they actually have a lot of resources they aren’t disclosing. We may never know the whole truth but I do believe the US firms will make changes and find ways to stay afloat and compete effectively. Lots to watch!
I think it’s built on pirated OpenAI actually. One of the early versions. They are really good at creating copies. I’m in MAICON and they covered this in December.
All of this is fascinating.
Mostly definitely. Everybody copies everybody in tech. The thing is, with open-source, that’s what it’s all about, developers coming together to make it better.
The lesson is the classic first-mover spending boatloads to create, and then getting copied by the second-mover, who can do so more cheaply and efficiently.
Hence, there is now HUGE opportunity for AI entrepreneurs to enter the ecosystem and develop something great for the consumer. Ultimately, we the consumers, benefit.
As investors, it’s so hard to pick the exact winner(s). So I like buying a fund with a number of different names. But I also like Apple, as the slow, incumbent, second mover who doesn’t seem to have spent billions on AI GPUs, and will now benefit from cost savings and rapid AI adoption.
There will be and is huge demand for the entire AI infrastructure being built. This will be a blip on the chart when it’s all said and done. The rate of innovation is so rapid that NVDA will be selling every chip they can make. Good luck trying to run the full Deepseek R1 model on anything running less than a ~$10k chipset and even then the performance will be slow and generally poor.
All this does is provide all of us reading this with a great AI model to leverage for daily use and push the industry even faster to innovate.
Once literally every SAAS company is running AI models in the background to process core parts of their software it will require huge amounts of resources that simply don’t exist today.
Buy the chips and the dip
I agree, Jevon’s Paradox in action. I wonder what this means for Databricks, another name I own that recently raised a large round at a higher valuation. I must find out!
I’m currently in Mexico and I downloaded Deepseek and tried to create an account. Trying using a Google account generated an error. I then tried using an email address and after an hour I still have not received the required code to setup the account. Maybe all the hype has overloaded their system as their service status is having problems.
Interesting feedback, thanks Alan! I’m now downloading DeepSeek for privacy concerns at the moment, given it’s no better than OpenAI and Anthropic. It should force OpenAi and Anthropic to provide more value for its monthly $20/fee or whatever… or give more away to the free users.
At the end of the day, this is a net positive for consumers and adoption of AI. So firms like Apple should benefit, as they have a huge ecosystem and haven’t dumped massive amounts into AI Capex yet. A win for the slow, 2nd mover!
I’m going out on a limb here but I think this is a temporary, knee-jerk reaction. Which will be stopped by another knee-jerk reaction from the current administration in Washington.
We’ll be back to normal by the end of the week. It’s going to take more than this to kill the momentum that’s been building in markets for well over a year.
Nothing to look at. And feel free to laugh at me if I’m wrong, it’s all good.
Great read Sam, and very timely. Thank you for explaining this emerging topic in detail. Makes me wonder if there is any real moat around the AI business model.
Not sure how you turned this around so quickly, Well Done!
You’re welcome! As an investor in AI, I’m highly interested in what’s going on.
I was following everything that was going on with DeepSeek Sunday evening after getting in some runs at Palisades, Tahoe. Passed out by 10pm, and woke up at 4:15 am to see what was going on and got to work writing. I generally try and write before the wife and kids wake up, so I can free up time to spend with them and do whatever.
Let me go put in some buy orders now. Fascinating stuff going on in AI land!
Nice read, Sam and very timely. I personally think the Magnificent Seven especially Nvidia will face erosion in stock price that will probably endure over a quarter until the industry figures out how to respond. So Nvidia down, Apple up for the short term. OR the new commander-in-chief could issue some executive order that may ban/prevent DeepSeek’s use Stateside?…who knows.
So short Nvidia and long Apple? :)
On a side note, 4:15 am is aggressive, man! It’s awe-inspiring. Does it remind you of your I-banking days? :) I used to work in sell-side equity research in a previous avatar myself, but those days are long gone. No way can I arise at 4:30am like I used to, to get on those Asia-Pac calls back in the day. Not at my age. And I’m a bit younger than you!
Cheers,
Indian Tiger
I definitely wouldn’t short NVIDIA after today’s sell-off. But I am glad my Apple is rebounding after poor performance YTD.
I got up almost every morning by 4:30am during the pandemic in 2020-2021 to write and get stuff done before the family woke. It was bedlam with both kids at home all the time and a 2 year old. It was maybe just adrenaline as the protector of the house during uncertain times that enabled me to regular wake early.
This morning, it was too exciting not to get up and write something before hitting the slopes. I got 5 runs in so far after staining some doors and trim at my condo and getting lunch. Now I’m gonna do several more runs and take a tap by 3pm and hit the hot tub after!
So long as I can nap, waking up by 4:30am is not a problem. Responding on the chairlift actually!
Remind me what you’re up to again?