Transcription
The biggest thing I learned from researching Deep Seek is that Wall Street still doesn't understand AI. I think we're going to look back at Deep Seek R1 as one of the most important moments in market history, right up there with the release of ChatGPT, which is what caused Nvidia's meteoric rise.
In this video, I'm going to highlight four other stocks that are set to win big thanks to Deep Seek's AI breakthroughs that no one else is talking about. Your time is valuable, so let's get right into it.
First things first, this video has a bit of a twist. I'm going to highlight two stocks if you believe everything about Deep Seek and two if you don't. Don't worry, I'll go over their big AI breakthroughs in this video, so you have everything in one place. I'm also not here to waste your time, so here are the stocks that I'll be covering up front.
If you believe that companies are about to spend a lot less on AI chips and compute infrastructure than software companies like CrowdStrike and Palantir, here are the way to go. But if you believe that companies will spend even more on AI chips because they're about to get a lot more bang for their buck, then Meta Platforms is already one of the biggest and most profitable AI hardware investors, and Taiwan Semiconductor builds all the chips.
These are very different companies, so to understand why I'm splitting them this way, we need to understand how Deep Seek just changed the AI landscape in the first place.
Starting with the elephant in the room, Deep Seek R1 was allegedly trained for around $5.5 million, which is 35 times less than the $200 million it cost OpenAI to train GPT-4. It's also 100,000 times less than the proposed budget for Project Stargate.
Also, Deep Seek R1 was allegedly trained using H800 GPUs, which are slower Nvidia chips that complied with trade restrictions on China before they were banned by U.S. sanctions. So the big question that investors are asking is, why are companies like Microsoft, Google, and Amazon spending tens of billions of dollars per year on AI if Deep Seek can train a bleeding-edge model for so cheap?
So the technology actually doesn't exist. So it's—wait, wait, wait—the technology, William, here is the technology. I've asked you to simply make it smaller. Okay, sir, and that's what we're trying to do, but honestly, it's impossible. Tony Stark was able to build this in a cave with a box of scraps.
Believe it or not, it doesn't matter whether Deep Seek is telling the truth about their costs. What actually matters are the innovations in Deep Seek's technical papers that American AI labs have already reproduced. I broke them down in my previous video, but let me give you a very quick overview here before diving into the stocks.
A lot of Deep Seek's innovations focus on saving memory since that's one of the reasons training AI models take so many GPUs. Deep Seek developed a few math techniques that allow them to use 8-bit numbers instead of 32, predict multiple words instead of one at a time, use compressed data instead of a set of raw data, and share parameters and prompts between multiple models that work together.
There's a lot more to it than that, but the end result of stacking all of these techniques is around a 45x efficiency improvement, which is why Deep Seek's API pricing is so cheap compared to OpenAI.
The second breakthrough is considered one of the holy grails of AI. Deep Seek R1 developed reasoning capabilities by itself, meaning nobody programmed the model to generate long chains of thought, verify its work step by step, or allocate more compute power to harder problems. The AI model naturally evolved these behaviors based on clever reward functions set by its developers.
Said another way, the developers gave the model lots of problems with objective solutions and told it what a good answer looks like, but the model taught itself to think in a way that gets those answers. And again, this was all verified by AI labs in the U.S. already.
This breakthrough is important because it addresses a huge problem with transformer models: hallucinations. Most LLMs try to generate words that make sense with the words that came before, making it hard for them to notice their mistakes. But chain-of-thought reasoning breaks inference into multiple steps, which opens a window for AI models to check their answers and change their approach when things go wrong.
That's why reasoning models like OpenAI's GPT-4 and Deep Seek R1 top the charts in a wide variety of benchmarks, except Deep Seek's API is over 25 times cheaper.
So what does that mean for AI stocks? Are companies about to spend 25 times less, or will they spend even more since they're getting a lot more bang for their buck? One thing's for sure: more people are about to use more new AI tools to generate a lot more data, which means cybersecurity is about to be a major issue for everyone across the board.
In fact, CrowdStrike's stock just hit an all-time high because Deep Seek itself got hit with a cyber attack. Speaking of security, I just found out that dozens of online data brokers were selling my personal data. If you've been getting more spam phone calls, texts, or emails lately, they might be selling yours too.
That's why I joined Delete Me, the sponsor of this video. Delete Me is a hands-free subscription service that will remove your personal information from hundreds of online data brokers. You just sign up, enter your information, and let their experts get to work.
They've reviewed over 44,000 listings for me so far. After seven days, you get a privacy report showing everything they've done. I just got my eighth report, and Delete Me found 400 new pieces of my personal information across 34 different data brokers—not just my name and address, but my family's and my fiancé too.
And here's the best part: they'll keep scanning these websites even after they remove my information. I even get my own privacy adviser if I want to talk to a real person and make a custom removal request.
So if you care about your data and your family's privacy, you can get 20% off any consumer plan with my code SYMBOL20 by going to join.deleteme.com/symbol2 or by using my link in the description. A big thank you to Delete Me for supporting the channel and keeping my family's data safe.
All right, the reason that CrowdStrike's stock hit an all-time high when Deep Seek got hit with a cyber attack is because CrowdStrike's Falcon platform just achieved a perfect score in a real-world ransomware test when it stopped all known and unknown threats with no false positives.
CrowdStrike focuses on endpoint detection and response, which means securing workstations, laptops, smartphones—really any device that connects to a specific network. Their Falcon platform can be broken down into three parts.
It has a library of cloud-based modules to run things like antivirus scans, manage firewalls, detect malware, lock down cyber threats, and a lot more. It also has a threat graph that keeps track of a company's different networks, devices, users, and permissions, which CrowdStrike can then compare to actual network traffic and deal with any discrepancies as they come up.
The third part is the Falcon agent, which is a tiny piece of software that runs on every device and sends data back to CrowdStrike so it can run different cloud modules and update their threat graph. CrowdStrike reported 29% revenue growth on their latest earnings call.
CrowdStrike's revenue recently passed a billion dollars per quarter, and that growth should only accelerate as Deep Seek's breakthroughs make AI models faster and cheaper for hackers and other malicious actors.
On the flip side, Palantir builds some of the best AI software platforms out there, from Gotham and Foundry to Apollo and AIP, making Palantir a no-brainer if you think that Deep Seek will shift AI spend from hardware to software.
Palantir's Gotham and Foundry platforms use AI to enable data science across some of the biggest enterprises on Earth, from governments and militaries to massive commercial companies. They do this by creating digital twins for organizations, which are also called ontologies.
At a high level, an ontology is a map of a company's resources and how they all relate—kind of like a social network. For example, an army is a network of hundreds of personnel, assets, facilities, and their objectives.
Their relationships include everyone's rank, their chain of command, who has access to what information, everyone's current orders and locations, who's responsible for every computer, vehicle, and so on. That's a lot of information to track, especially when things are always changing in real time.
Palantir platforms help companies build and use these ontologies to do things like analyze data, track their resources, and optimize their operations. But one challenge with Gotham and Foundry, at least in my experience, is the learning curve can be pretty high.
That's where AIP comes in. At a high level, Palantir's artificial intelligence platform lets users interact with their company's ontology using prompts, kind of like they would with a much more capable and secure ChatGPT.
This seriously lowers the barrier of entry to using Palantir platforms, which is why their growth exploded after they started pushing AIP. One of the great features of AIP is companies can use whatever AI model they want.
While I seriously doubt that any government or Fortune 500 company is about to plug all their data into a Chinese model like Deep Seek, I wouldn't be surprised to see Palantir build some of Deep Seek's innovations directly into AIP, just like they did with ChatGPT to make AIP in the first place.
Palantir's earnings are only a few days away, so consider liking this video, subscribing to the channel, and turning on notifications so you'll know as soon as I publish my analysis.
All right, CrowdStrike and Palantir are both great bets if you believe that Deep Seek is going to shift corporate spending from hardware to software. But one company that's going to win either way is Meta Platforms.
Meta just had their fourth-quarter earnings call, and they posted one of their biggest beats in earnings per share yet. On the call, Mark Zuckerberg made it clear that he thinks investing heavily in AI infrastructure is going to be a long-term strategic advantage and that Deep Seek has only strengthened that conviction.
Meta Platforms also plans to implement some of Deep Seek's advancements into their open-source LLaMA family of models. Mark Zuckerberg echoed what I said in my previous Deep Seek video: even if compute resources do end up moving away from AI model training—and that's a big if—Deep Seek R1 shows the power of test-time scaling.
That's where AI models use more compute power to think about harder problems, break them down into parts, test multiple solutions, and even ask more specialized AI models for help, all of which drastically ramp up the number of tokens being generated before providing a response during inference.
In Zuckerberg's own words, this doesn't mean you need less compute because you can apply more compute at inference time in order to generate a higher level of intelligence and a higher quality of service.
Don't fall into the trap of dismissing Mark Zuckerberg. He's one of the longest-running founders and CEOs in Silicon Valley, driving one of the biggest and most profitable companies on the planet with billions of monthly active users and millions of corporate customers already engaging with AI on his platforms today.
In fact, he expects Meta's AI assistant itself to reach one billion users this year. There are only 8 billion people on Earth, so that's a pretty big deal.
While Meta Platforms is positioned to win either way, Taiwan Semiconductor will win big if Mark Zuckerberg is right about a rise in demand for AI hardware. But if Deep Seek just lowered AI training costs by 25x, why would hardware spend go up?
There are actually many different ways to answer this. You've probably heard of Jevons Paradox, which says that as something gets cheaper, demand rises faster than costs come down, so overall spending actually goes up as things get cheaper, not down.
Or what about Moore's Law, which basically says that a given microchip gets roughly twice as cheap every two years, but spending on chips is at an all-time high? If you add up all of our cell phones, laptops, desktops, tablets, and server infrastructure, either way, this observation that when something gets cheaper, people adopt it faster and use it way more is why Deep Seek might actually cause AI hardware spend to rise, not fall.
And if you believe that, then the Taiwan Semiconductor Manufacturing Company is another no-brainer since they sit at the heart of the entire AI revolution. TSMC is a pure-play foundry that makes around 90% of all advanced microchips on Earth, from Apple's A-series and M-series processors for iPhones and MacBooks to Tesla's full self-driving chips and Nvidia's data center GPUs.
The reason they can make all these advanced AI chips is they only focus on chip fabrication, not design, and that comes with two huge benefits. First, TSMC benefits from economies of scale to the point where they can even stand up custom processes for their clients.
For example, TSMC has a custom 4-nanometer process called 4NP, which they use to make Nvidia's Blackwell GPUs. Blackwell is around four times better at AI training and 30 times better at AI inference than Nvidia's Hopper chips, which is why there's so much demand for Blackwell.
That demand flows down the supply chain to companies like TSMC, which leads to them building more custom processes and that creates a big positive feedback loop that's very hard for competitors like Intel to disrupt.
The other big benefit is the network effect. Since TSMC makes many different kinds of chips, they can learn something once and then apply it everywhere—think best practices for specific chip designs or fabrication machines.
For example, Nvidia's Blackwell chip actually takes two separate dies and connects them with a 10 TB per second link so they act like one GPU. But this high-speed chip-to-chip connection needs to be placed with incredible precision. If the placement is off, even by a tiny bit, the whole chip can fail as the metal layers and materials expand when they heat up during normal operations.
So if another company wants to make a chip with two dies that are connected by a high-speed link, they're probably going to hire the one company that's already solved these kinds of challenges to build it.
The global AI chip market is expected to almost 9x in size over the next eight years, which would be a compound annual growth rate of over 30% through 2033. TSM stock is up by over 80% in the last year alone.
But will this growth continue after Deep Seek, or will chip demand collapse, taking companies like TSMC and Nvidia with it?
And now we've come full circle. At the start of this video, I said there were two paths before us: one where AI spend shifts from hardware to software and one where it doesn't. But the truth is both of these things will happen because when costs go down, the entire pie gets bigger.
Let me be more specific: somebody already got Deep Seek working on a Raspberry Pi and a single desktop GPU. That means large language models are no longer only for companies that can afford tens of thousands of Nvidia's most expensive chips.
The barrier to entry for hardware costs to run large language models just went way down, which means way more schools, research labs, small businesses, and new markets will start to use generative AI way more, which will more than offset these lower training costs.
On the flip side, I expect the hyperscalers and other tech giants to fill huge data centers to max capacity since those same innovations in memory just made every dollar spent on hardware go that much further than before, which lets them support more workloads at higher margins.
That means I expect all four stocks in this video to do well—not just hardware or just software like everyone else is arguing about right now.
And it's not just these four stocks. If you want to know what other stocks I'm buying to get rich without getting lucky in 2025, even after Deep Seek, make sure to watch this video next.
Either way, thanks for watching to the end, even though I gave you everything up front. Until next time, this is Ticker Symbol U. My name is Alex, reminding you that the best investment you can make is in you.