DeepSeek shows just how fragile the AI market is

DeepSeek DeepSeek DeepSeek! That word has taken over the internet over the past week. Though things have cooled off a bit, it’s still making some decent headlines. What was once the most promising chatbot of the year has turned into a lingering cybersecurity threat. More importantly, it really turned the AI industry on its ear. Because of that, DeepSeek pretty much showed just how fragile the AI market is.

What is DeepSeek?

In case you don’t know what DeepSeek is, here’s a quick refresher. DeepSeek is an AI chatbot just like ChatGPT or Gemini. It burst onto the scene just over a week ago, and it started making waves; huge waves! Though it is a chatbot just like the others we use today, DeepSeek’s model R1, was able to surpass many of the top models on the market, including some of OpenAI’s powerful o1 reasoning models. This was a bit of a shocker, seeing as it basically came out of nowhere.

Not only that, but the news came out that it only took DeepSeek a few million dollars to train its model. This is contrary to the hundreds of billions of dollars that other AI companies are paying to train theirs. Training a model that outperforms many top models- while costing a fraction of the cost- was also one of the most notable aspects of R1. The company claims that it had a software-based approach to training its model rather than relying on hardware-based data centers.

Controversies

Not long after DeepSeek came out, things started to hit the fan. For starters, people found out that DeepSeek gathers a ton of information from users, which isn’t unheard of for an AI company, and stores it on servers located in China. That’s not the sort of news you’d want to hear during the drama currently going on with TikTok.

What makes this worse is the fact that DeepSeek has been hit by some pretty severe cyberattacks since its launch. People found out that DeepSeek hasn’t been using the best security measures to protect people’s data.

On top of that, we also found out that the company’s claims to train R1 on just a few million dollars might be false. There’s no telling just how much money the company spent, but it could be much higher. Additionally, DeepSeek might have been trained on data stolen from OpenAI. Some of the stories are still developing at the time of writing this.

DeepSeek shows just how fragile the AI market is

With all of the innovations and AI services that came out over the past two years, you’d think that the AI market has pretty much solidified. Companies have an idea of the kinds of AI tools people typically use, and new companies are finding new ways to shove AI down our throats. There exists a fleet of flagship AI models and chatbots that lead the march of progress from the likes of Google, OpenAI, Meta, Microsoft, Anthropic, Perplexity, and (eventually) Apple.

Things seemed stable… then, along came a little blue whale. Saying that DeepSeek made a splash in the industry would be an understatement- a huge understatement. DeepSeek took a sledgehammer to everything we knew about the AI industry.

Stock market shake-up

Let’s talk about the biggest example of that, the stock market. We’ve seen a record fall in stocks from Nvidia. This is the company that supplies AI chips to all of the companies training their models. Once people realized that its chips are the best for this kind of thing, its stock prices soared, and it became a $2 trillion company in a relatively short amount of time.

When DeepSeek came along, it shaved $597 billion off of the company’s value. Not $597 million; we’re talking nine zeros! When investors found out that they were paying to support a company supplying hardware when a company came along with a software-based solution, they backed out quickly. Not only that, but other companies suffered immensely in the market like Microsoft. All of the top 10 companies to lose value over DeepSeek lost billions upon billions of dollars, but none of them came close to what Nvidia lost.

The thing is that people still don’t know what’s going to bring humanity into the AGI (Artificial General Intelligence) era. This is the term that people, mostly OpenAI CEO Sam Altman, talk about. This is an AI that’s as smart as a human being in several ways. We have no idea if creating AI that powerful will require more data, more energy, or more money.

However, this unknown company comes out of the woodwork saying that its model cost pocket change to train, and the AI stock market loses a trillion dollars. That’s right; the accrued losses from all the top companies add up to around a trillion dollars. It dipped and then fluctuated for a couple of days.

The progress of AI depends on several factors like data, the companies training the models, energy, user-facing services, and so on. One of those factors is money, and a fair bit of that money comes from investors. Wealthy and hopeful investors are pumping money into these companies in the hopes that AI technology will take off. However, they want to know that their money is going to good use. So, the minute it seems like the company they’re investing in won’t give them a return investment, they’ll pull out.

The fear of non-U.S.-based AI companies

As stated earlier, the U.S. government is still grappling with TikTok and how it handles the data of U.S. citizens. This data was being hosted on servers in China, which means that it could be accessed by the Chinese government. This doesn’t sit right with American politicians.

Well, guess what, DeepSeek is a whale swimming in a sea of your data. In order to properly use the platform, you have to give the platform access to data like:

• Username
• Email address
• Phone number
• Passwords
• Payment information
• Age
• Proof of identity
• Uploaded files
• Text Prompts
• Chat history
• “Other content you provide”, as per the company’s privacy page
• Audio recordings
• Browser cookies
• IP addresses
• The model of the device you’re using
• The operating system of the device you’re using
• Keystroke patterns
• System language
• Crash reports
• Service logs
• Information about you that it may obtain from advertisers
• Access tokens from third-party services that you log in with (i.e Google or Apple)

Does that scare you? Well, don’t use Meta AI, ChatGPT, Gemini, Claude, Perplexity, or any of the other AI platforms out there, as they scoop up similar information. So, what makes DeepSeek so problematic?

The majority of the top AI companies chomping on your data like Pac-Man are based in the U.S. So, it’s less of an issue. However, we can’t doubt that such a large amount of data being stored in China comes with its problems. In that country, if the government requests data from a company, it’s more or less compelled to give up that data, no questions asked. So, there’s that issue.

How does this point to the instability of the AI market? Well, no one expected something like this to happen. Again, all of the top AI companies are in the U.S., so we’ve been pretty comfortable with most of the AI data being stored in the States. The AI landscape hasn’t diversified enough so that major companies from other countries have access to data. Also, there aren’t really any laws in place that address this.

We’re still living in a world where most of the AI-related data is stored in the U.S. While storing data in China is one story, we’re sure that data being stored in Europe, the UK, Russia, Australia, or most other countries would also cause some sort of concern.

What would happen if a new rising star AI model were to come out of France (let’s just say it’s called Tête-AI), and we also got the news that the information listed above lives in a French data center? Would people feel the same? Would the government react the same way as with China? Also, would the world be happy knowing that their data is simultaneously being stored in Europe (for Tête-AI), China (for DeepSeek), and the U.S. (for all the others)? What if one comes out of South Africa? That’s another country to store data.

The world just doesn’t know where everyone’s data is going to live, so when new countries launch AI models, we’ll go through the same drama.

People fell for DeepSeek’s charming words

It should be common knowledge that maintaining an AI company is EXPENSIVE. This is why the major models in the industry are being trained primarily by billion-dollar/trillion-dollar corporations or companies being funded by those corporations. Google, Amazon, Meta, Microsoft, and Apple are training their own models, and they’re all worth trillions. OpenAI, LinkedIn, Anthropic, and Perplexity are being funded by one or several of those companies. The list goes on.

However, it took one company to (most likely falsely) claim to train a model on just a few million dollars to redefine what everyone knew about AI. Suddenly, the paradigm shifted. “Hear ye, hear ye! Training AI doesn’t require billions of dollars! New startup does it with less than $7 million!!” And everyone bought it! Suddenly, people started to question what it takes to make a good AI model, and that’s pretty dangerous.

DeepSeek came from the ether and took a shot at one of the biggest things holding AI back, and that’s money. The company basically said “You rely so much on large hardware-based AI models to do your AI computations. What if we could do it at a fraction of the cost?” With just a few simple words, and a little white lie (according to sources, DeepSeek spent about $1.6 billion just on hardware), the AI industry was shaken. It didn’t take much.

This shows that we really don’t know what to believe at this point. We don’t know what we should rely on or what companies to invest our money in just yet. If DeepSeek can come by and do this with just a few sentences, it could very well happen again.

DeepSeek’s data security issues

Adding to DeepSeek’s growing list of problems is the company’s lackluster approach to security. Cybersecurity experts found a massive data leak within the company where millions of lines of code were made publicly available with no encryption. Not only that, but the company also had to restrict signups due to a cybersecurity attack shortly after launching to the public. Adding to that, DeepSeek was proven to be vulnerable to jailbreak attacks.

All of this points to a company with a ton of data and flimsy locks to keep it safe. We all know that any AI company that comes along will need to extract a ton of data in order to operate. However, this issue doesn’t show how DeepSeek is a data security risk, it’s a reminder that any company that takes as much data from us can be hacked. If you’re afraid that your data is being stolen by a company with weak security, then you shouldn’t use any AI tools at all.

Other AI companies also need to access your data, and they’ve also been the target of cyberattacks. For example, Microsoft recently went through some drama with being hacked by Russia, and we can’t forget about the cybersecurity issues that OpenAI went through last year.

Large companies are always the targets of hackers, and we’ve seen many cases (just look at T-Mobile) where hackers were able to make off with tons of data. Right now, we’re just not ready for the kind of world where a company can just come in and eat up a ton of data from several countries around the globe.

No company is safe from data breaches. Granted, this doesn’t only apply to AI. However, AI is such a data-heavy industry. If someone launches a music streaming service, they’ll need to extract certain types of data from you, and the same would go for a navigation app. However, AI requires that the company extract every bit of information about you- short of the day you had your first kiss. There’s not much of an infrastructure keeping new AI companies from sucking up data and holding it in leaky data servers.

Wake-up call

DeepSeek seems to be more trouble than it’s worth, but that may not be the case. Sometimes, it takes a company like that to come along in order to show us that the AI industry is still solidifying. While the products we use are smart, the industry itself still has a ways to go. We weren’t ready for such a shake-up in the AI space, and it shows.

The thing is that we’ll need to address these issues with the industry now before we get the next company like DeepSeek; the next company to take the AI world and treat it like a snow globe. The genie is out of the bottle at this point. We know what things to say in order to get people to notice a new AI company. Other companies may follow in DeepSeek’s footsteps in how they train their models.

More importantly, other companies will also market their models much the same way. How many other companies will launch with lies about how they train their models for just a few million dollars? How many more companies will cause massive waves in the stock market?

It’s just a matter of time, so the best thing that we can do is learn from this experience. It could greatly benefit the AI industry in the long run.