DeepSeek is shaking up the AI industry with cost-efficient large language models it claims can perform just as well as rivals from giants like OpenAI and Meta. The Chinese startup says its flagship R1 reasoning model is capable of achieving “performance comparable” to OpenAI’s o1 equivalent, while the newly released Janus Pro multimodal AI model can supposedly outperform Stable Diffusion and DALL-E 3.
DeepSeek’s ChatGPT competitor quickly soared to the top of the App Store, and the company is disrupting financial markets, with shares of Nvidia dipping 17 percent to cut nearly $600 billion from its market cap on January 27th, which CNBC said is the biggest single-day drop in US history.
The AI assistant is powered by the startup’s “state-of-the-art” DeepSeek-V3 model, allowing users to ask questions, plan trips, generate text, and more. As downloads of DeepSeek’s app spiked, the startup began restricting signups due to “malicious attacks.”
Launched in 2023 by Liang Wenfeng, DeepSeek has garnered attention for building open-source AI models using less cash and fewer GPUs when compared to the billions spent by OpenAI, Meta, Google, Microsoft, and others. If DeepSeek’s performance claims are true, it could prove that the startup managed to build powerful AI models despite strict US export controls preventing chipmakers like Nvidia from selling high-performance graphics cards in China.
Here’s all the latest on DeepSeek.
Apple is reportedly working with Alibaba to launch AI features in China.
While Apple Intelligence has reached the EU -- and, according to some, devices where it had already been declined -- the company hasn’t launched its AI features in China yet. A report by The Information on Tuesday indicates it could be getting closer, saying that after evaluating models from Tencent, ByteDance, Alibaba, and DeepSeek, Apple has submitted some features co-developed with Alibaba for approval by Chinese regulators.
DeepSeek gets the TikTok treatment.
A new bipartisan bill seeks to ban Chinese AI chatbot DeepSeek from US government-owned devices to “prevent our enemy from getting information from our government.” A similar ban on TikTok was proposed in 2020, one of the first steps on the path to its recent brief shutdown and forced sale.
Australia, Italy, and South Korea have already enacted similar bans, as has Texas, while the US Navy and NASA have blocked the app internally.
Nvidia says its new GPUs are the fastest for DeepSeek AI, which kind of misses the point
Image: Cath Virginia / The Verge
Nvidia is touting the performance of DeepSeek’s open source AI models on its just-launched RTX 50-series GPUs, claiming that they can “run the DeepSeek family of distilled models faster than anything on the PC market.” But this announcement from Nvidia might be somewhat missing the point.
This week, Nvidia’s market cap suffered the single biggest one-day market cap loss for a US company ever, a loss widely attributed to DeepSeek. DeepSeek said that its new R1 reasoning model didn’t require powerful Nvidia hardware to achieve comparable performance to OpenAI’s o1 model, letting the Chinese company train it at a significantly lower cost. What DeepSeek accomplished with R1 appears to show that Nvidia’s best chips may not be strictly needed to make strides in AI, which could affect the company’s fortunes in the future.
AI is ‘an energy hog,’ but DeepSeek could change that
DeepSeek startled everyone last month with the claim that its AI model uses roughly one-tenth the amount of computing power as Meta’s Llama 3.1 model, upending an entire worldview of how much energy and resources it’ll take to develop artificial intelligence.
Taken at face value, that claim could have tremendous implications for the environmental impact of AI. Tech giants are rushing to build out massive AI data centers, with plans for some to use as much electricity as small cities. Generating that much electricity creates pollution, raising fears about how the physical infrastructure undergirding new generative AI tools could exacerbate climate change and worsen air quality.
How DeepSeek crashed the AI party
Image: Alex Parkin / The Verge
The DeepSeek story contains multitudes. It’s a story about the stock market, whether there’s an AI bubble, and how important Nvidia has become to so many people’s financial future. It’s also a story about China, export controls, and American AI dominance. And then, somewhere in there, there’s a story about technology: about how a startup managed to build cheaper, more efficient AI models with few of the capital and technological advantages its competitors have.
On this episode of The Vergecast, we talk about all these angles and a few more, because DeepSeek is the story of the moment on so many levels. Nilay and David discuss whether companies like OpenAI and Anthropic should be nervous, why reasoning models are such a big deal, and whether all this extra training and advancement actually adds up to much of anything at all. (Nilay has a long comparison to Bluetooth, in case that helps you guess where we land.)
The too-online finance dorks are at it again.
Incredible kicker from FT Alphaville, on top of some truly bizarre memes from Deutsche Bank. We love our blogging brethren.
DeepSeek database left user data, chat histories exposed for anyone to see
Image: Cath Virginia / The Verge
DeepSeek has secured a “completely open” database that exposed user chat histories, API authentication keys, system logs, and other sensitive information, according to cloud security firm Wiz. The security researchers said they found the Chinese AI startup’s publicly accessible database in “minutes,” with no authentication required.
The exposed information was housed within an open-source data management system called ClickHouse and consisted of more than 1 million log lines. As noted by Wiz, the exposure “allowed for full database control and potential privilege escalation within the DeepSeek environment,” which could’ve given bad actors access to the startup’s internal systems. These findings were first reported by Wired.
Inside Microsoft’s quick embrace of DeepSeek
Image: Cath Virginia / The Verge
The Chinese startup DeepSeek shook up the world of AI last week after showing its supercheap R1 model could compete directly with OpenAI’s o1. While it wiped nearly $600 billion off Nvidia’s market value, Microsoft engineers were quietly working at pace to embrace the partially open- source R1 model and get it ready for Azure customers. It was a decision that came from the very top of Microsoft.
Sources familiar with Microsoft’s DeepSeek R1 deployment tell me that the company’s senior leadership team and CEO Satya Nadella moved with haste to get engineers to test and deploy R1 on Azure AI Foundry and GitHub over the past 10 days. For a corporation the size of Microsoft, it was an unusually quick turnaround, but there are plenty of signs that Nadella was ready and waiting for this exact moment.
DeepSeek, Stargate, and the new AI arms race
Image: Alex Parkin / The Verge; Getty Images
On today’s episode of Decoder, we’re talking about the only thing the AI industry — and pretty much the entire tech world — has been able to talk about for the last week: that is, of course, DeepSeek, and how the open-source AI model built by a Chinese startup has completely upended the conventional wisdom around chatbots, what they can do, and how much they should cost to develop.
DeepSeek, for those unaware, is a lot like ChatGPT — there’s a website and a mobile app, and you can type into a little text box and have it talk back to you. What makes it special is how it was built. On January 20th, the startup’s most recent major release, a reasoning model called R1, dropped just weeks after the company’s last model V3, both of which began showing some very impressive AI benchmark performance. It quickly became clear that DeepSeek’s models perform at the same level, or in some cases even better, as competing ones from OpenAI, Meta, and Google. Also: they’re totally free to use.
Mark Zuckerberg says Meta isn’t worried about DeepSeek
Image: Cath Virginia / The Verge; Getty Images
Nearly everyone seems to be suddenly freaking out about the rise of DeepSeek. Meta isn’t worried, though.
That was CEO Mark Zuckerberg’s message to investors during his company’s fourth-quarter earnings call on Wednesday. During the Q&A portion of the call with Wall Street analysts, Zuckerberg fielded multiple questions about DeepSeek’s impressive AI models and what the implications are for Meta’s AI strategy. He said that what DeepSeek was able to accomplish with relatively little money has “only strengthened our conviction that this is the right thing to be focused on.”
Someone might be squatting on DeepSeek’s trademark.
Just days before DeepSeek filed an application with the US Patent and Trademark Office for its name, a company called Delson Group swooped in and filed one before it, as reported by TechCrunch. The outlet found that Delson Group’s owner has a “history of trademark squatting,” which could prove inconvenient for DeepSeek.
But like my colleague Sarah Jeong writes, just because someone files for a trademark doesn’t mean they’ll actually get it.
Microsoft makes DeepSeek’s R1 model available on Azure AI and GitHub
Image: Cath Virginia / The Verge
Microsoft is bringing Chinese AI company DeepSeek’s R1 model to its Azure AI Foundry platform and GitHub today. The R1 model, which has rocked US financial markets this week because it can be trained at a fraction of the cost of leading models from OpenAI, is now part of a model catalog on Azure AI Foundry and GitHub — allowing Microsoft’s customers to integrate it into their AI applications.
“One of the key advantages of using DeepSeek R1 or any other model on Azure AI Foundry is the speed at which developers can experiment, iterate, and integrate AI into their workflows,” says Asha Sharma, Microsoft’s corporate vice president of AI platform. “DeepSeek R1 has undergone rigorous red teaming and safety evaluations, including automated assessments of model behavior and extensive security reviews to mitigate potential risks.”
OpenAI has evidence that its models helped train China’s DeepSeek
Chinese artificial intelligence company DeepSeek disrupted Silicon Valley with the release of cheaply developed AI models that compete with flagship offerings from OpenAI — but the ChatGPT maker suspects they were built upon OpenAI data.
OpenAI and Microsoft are investigating whether the Chinese rival used OpenAI’s API to integrate OpenAI’s AI models into DeepSeek’s own models, according to Bloomberg. The outlet’s sources said Microsoft security researchers detected that large amounts of data were being exfiltrated through OpenAI developer accounts in late 2024, which the company believes are affiliated with DeepSeek.
Why everyone is freaking out about DeepSeek
Cath Virginia / The Verge, Getty Images
It took about a month for the finance world to start freaking out about DeepSeek, but when it did, it took more than half a trillion dollars — or one entire Stargate — off Nvidia’s market cap. It wasn’t just Nvidia, either: Tesla, Google, Amazon, and Microsoft tanked.
DeepSeek’s two AI models, released in quick succession, put it on par with the best available from American labs, according to Alexandr Wang, Scale AI CEO. And DeepSeek seems to be working within constraints that mean it trained much more cheaply than its American peers. One of its recent models is said to cost just $5.6 million in the final training run, which is about the salary an American AI expert can command. Last year, Anthropic CEO Dario Amodei said the cost of training models ranged from $100 million to $1 billion. OpenAI’s GPT-4 cost more than $100 million, according to CEO Sam Altman. DeepSeek seems to have just upended our idea of how much AI costs, with potentially enormous implications across the industry.
OpenAI CEO Sam Altman on DeepSeek R1: “an impressive model.”
The ChatGPT boss says of his company, “we will obviously deliver much better models and also it’s legit invigorating to have a new competitor,” then, naturally, turns the conversation to AGI.
Screenshot: @sama (X)
Trump says he’ll put tariffs on imported chips ‘in the near future’
Image: Cath Virginia / The Verge, Getty Images
Without going into detail about what might happen to the $52 billion in subsidies from the CHIPS Act under his administration, Donald Trump said tariffs on foreign computer chips, semiconductors, and pharmaceuticals are coming “in the near future.” He also namechecked DeepSeek’s AI releases, saying, “...coming up with a faster method of AI and less expensive, that’s good. I view that as a positive if it is fact and it is true, and nobody knows, but I view that as a positive.”
In the speech at the House GOP Issues Conference held at the Trump National Doral Resort in Miami Monday afternoon, he said that to return the production of these goods to the US, “we don’t want to give them billions of dollars like this ridiculous program Biden has.” Instead the incentive for manufacturers will be “they will not want to pay a tax.”
Nvidia’s market cap drops by almost $600 billion amid DeepSeek R1 hype.
As Chinese AI startup DeepSeek draws attention for open-source AI models that it says are cheaper than the competition while providing similar or better performance, AI chip king Nvidia’s stock price dropped today.
CNBC said that after closing at $118.58, down 17 percent, this was “the biggest drop ever for a U.S. company.”
Nvidia responds to the DeepSeek hype.
In a statement to Bloomberg, an Nvidia spokesperson said DeepSeek is an “excellent AI advancement” and shows how a company can create new AI models using the test-time scaling method, while “leveraging widely-available models and compute that is fully export control compliant.”
DeepSeek says its newest AI model, Janus-Pro can outperform Stable Diffusion and DALL-E 3.
Input image analysis is limited to 384x384 resolution, but the company says the largest version, Janus-Pro-7b, beat comparable models on two AI benchmark tests.
Correction: As TechCrunch notes, Janus-Pro image input is listed as limited to low resolution, not its output.
DeepSeek’s top-ranked AI app is restricting sign-ups due to ‘malicious attacks’
Image: Cath Virginia / The Verge
After surging to the top of Apple’s App Store charts in the US, DeepSeek’s AI Assistant is now restricting new user sign-ups. According to an incident report page, registrations are being temporarily limited “due to large-scale malicious attacks on DeepSeek’s services,” though it’s unclear how these limitations are being applied.
“Existing users can log in as usual,” DeepSeek said in its update. “Thanks for your understanding and support.” An alert banner on the DeepSeek web sign-up page says that “registration may be busy,” rather than entirely restricted, however, and encourages users to wait and “try again” if their application is unsuccessful.
China’s DeepSeek AI is hitting Nvidia where it hurts
Image: Cath Virginia / The Verge
A chatbot made by Chinese artificial intelligence startup DeepSeek has rocketed to the top of Apple’s App Store charts in the US this week, dethroning OpenAI’s ChatGPT as the most downloaded free app. The eponymous AI assistant is powered by DeepSeek’s open-source models, which the company says can be trained at a fraction of the cost using far fewer chips than the world’s leading models. The claim has riled financial markets, with Nvidia’s share price dropping over 12 percent in pre-market trading.
Downloads for the app exploded shortly after DeepSeek released its new R1 reasoning model on January 20th, which is designed for solving complex problems and reportedly performs as well as OpenAI’s o1 on certain benchmarks.