China's DeepSeek Tops Apple's App Store As It Sends Shivers Down Silicon Valley

UPDATE:

As per latest reports, DeepSeek has caused havoc in the US markets, where the Nasdaq fell by over 3%, dropping over $1 trillion off the index of technology companies at one point.

NVIDIA’s stocks fell by over 17%, resulting in a wipe out of $600 billion (eventually $520 billion) from its market capitalization at its peak, making it the biggest market value drop in U.S. stock market history. Yet, its value is more than 480% of what it was over the last two years. 

Resultantly, even the shares of Broadcom (17%), Microsoft (2.5%), Alphabet (4.2%), and Tesla (3.5%) among others. On the other hand, Apple picked up 3.7%. 

The drop is a result of 2023-founded DeepSeek building an AI assistant for a fraction of the cost of what rival leaders OpenAI, Microsoft, Google, and others used. Additionally, DeepSeek’s chatbot was created in around two months, faster than its competitors have been able to manage.  

DeepSeek is available to users for free, which begs the question; do AI companies really need expensive hardware, large amounts of money, and copious amounts of time to build such tools? 

— — — — 

Just over a year ago, a product named r1 was launched by artificial intelligence (AI) startup Rabbit (styled as rabbit).

That product was a ChatGPT-powered device that could help users perform tasks such as hailing a taxi, ordering food, translating speeches, and more, along with getting answers to questions.

Once touted to be the next big thing in the world of AI-powered devices, unfortunately, the rabbit r1 didn’t live up to its hype.

Now, a year on, another R1 is making waves in the generative artificial intelligence (GenAI) world—one that comes from Chinese AI company DeepSeek.

(Not related to the GenAI world, but Yamaha’s R1—formally YZF-R1—is also a popular product with the same name. Interestingly though, Yamaha’s R1 and VMAX motorcycles played the inspiration for an AI-themed concept superbike called Yamaha Y/AI which featured in Netflix’s anime series Tokyo Override.)

Just over a week ago, DeepSeek released its first-generation reasoning models, DeepSeek-R1 and DeepSeek-R1-Zero. The models are based on Alibaba Cloud’s Qwen and Meta’s Llama.

This release also included six dense models distilled from R1—1.5B, 7B, 8B, 14B, 32B, and 70B. Here, “B” stands for billion, which is the adjustable parameter count of the model and represents the parameters used during training to learn the relationship between words and phrases.

Reasoning models are AI models that “think” or reason out their outputs and consider context before generating answers.

DeepSeek’s R1 is open source and is currently available to users on their web app, mobile app, and API.

As per the company, the R1 rivals OpenAI’s o1-1217 model on reasoning tasks and even outperforms it in various benchmarks, including AIME, MATH-500, and SWE-bench Verified. Of course, the model isn’t far away from the other compared benchmarks—Codeforces, GPQA Diamond, and MMLU.


This announcement follows another one made by the company in late December, which saw the release of its DeepSeek-V3, an AI model that was trained using reduced-capability NVIDIA H800 chips, and cost only $5.6 million to build. This comes in stark contrast to the $100 million to $1 billion used to train models made by Anthropic, OpenAI, Meta, Amazon, and other American companies.

DeepSeek’s R1 also uses fewer tokens than the others—$0.14 per million input tokens (cache hit), $0.55 per million input tokens (cache miss), and $2.19 per million tokens (Output API). In comparison, OpenAI’s o1 model requires $7.5, $15, and $60 respectively.


DeepSeek’s powerful capabilities have seen the company’s AI assistant top the Apple App Store, overtaking OpenAI’s ChatGPT, to become the top-rated free application. This assistant is powered by the DeepSeek-V3 model.

As DeepSeek’s models have been built using less-powerful chips and costs less and can still outperform America’s best models, the company’s products have garnered the attention of not just users but also the Silicon Valley stalwarts leading the United States’ AI industry charge.

Speaking at the World Economic Forum in Davos, Switzerland, Microsoft CEO Satya Nadella said, “To see the DeepSeek new model, it’s super impressive in terms of both how they have really effectively done an open-source model that does this inference-time compute, and is super-compute efficient.”

“We should take the developments out of China very, very seriously.”

This refers to the ongoing technology battle between the two countries, which includes the chip export restrictions that left DeepSeek having to use the less-powerful NVIDIA H800 chips, in addition to bans on apps such as TikTok.

“DeepSeek R1 is one of the most amazing and impressive breakthroughs I’ve ever seen — and as open source, a profound gift to the world,” said Marc Andreessen, venture capitalist and General partner of Andreessen Horowitz.

The news of DeepSeek’s capabilities also hit the share market in the U.S., Europe, and Japan, as Nasdaq 100 futures were down 2.6% and S&P 500 futures fell 1.4%, while the shares of NVIDIA’s supplier Advantest fell 8.5% in Tokyo. Even SoftBank saw an 8.3% drop. Ahead of this, NVIDIA shares in Frankfurt dropped around 7%, and Tesla, Amazon, and Meta’s European shares dipped by 2%. Other technology companies also saw slumps.

This development comes just days after the U.S. announced one of the most heavily funded projects—Stargate—which will see a collaboration between OpenAI, Oracle, SoftBank, and HGX, with the participation of Arm, Microsoft, and NVIDIA.

Do you think DeepSeek will replace OpenAI’s ChatGPT as the world’s most popular GenAI tool or do you think OpenAI’s Stargate project will help it remain on top?

Let us know in the comments below!