The Nvidia stock may fall as Deepseeks ‘Amazing’ AI Model disturbs Openai

America’s policy of limiting Chinese access to Nvidia’s most advanced AI chips has inadvertently helped a Chinese AI developer jumping American rivals who have full access to the company’s latest chips.

This proves a fundamental reason why startups are often more successful than large companies: Button creates innovation.

An example of this is the Chinese AI model Deepseek R1 – a complex problem solving model competing with Openai’s O1 – which “zoomed in to the global top 10 in performance” – yet was built much faster with fewer, less powerful AI chips , at a much lower price, according to Wall Street Journal.

The success of R1 should benefit companies. That’s because companies see no reason to pay more for an effective AI model when a cheaper model is available – and is likely to improve faster.

“Openai’s model is the best in performance, but we also don’t want to pay for capabilities we don’t need,” Anthony Poo said, co-founder of a Silicon Valley-based startup that uses generative AI to predict financial returns. Journal.

Last September, Poo’s company switched from Anthropics Claude to Deepseek after testing showed that Deepseek “performed in the same way for about a quarter of the cost,” noted Journal.

When my book, Brain RushWas published last summer, I was concerned that the future of generative AI in the United States was too dependent on the largest technology companies. I contrasted this with the creativity of American Startups during the Dot-Com boom-which brought about 2,888 first public offers (compared to zero IPOs for American Generative AI-Startups).

Deepseek’s success could encourage new rivals to US-based developers of large language models. If these startups build powerful AI models with fewer chips and get improvements to the market faster, NVIDIA revenue can grow slower as LLM developers copy Deepseeks strategy of using fewer, less advanced AI chips.

“We refuse to comment,” a nvidia spokesman wrote in an e-mail on January 26th.

Deepseeks R1: Excellent performance, lower cost, shorter development time

Deepseek has impressed a leading American venture capitalist. “Deepseek R1 is one of the most amazing and impressive breakthroughs I’ve ever seen,” wrote Silicon Valley Venture Capitalist Marc Andreessen in an X-post on January 24.

To be righteous, Deepseeks technology lags after US rivals such as Openai and Google. However, the company’s R1 model – launched on January 20 – is “a close rival despite using fewer and less advanced chips and in some cases skipping steps that American developers considered significant,” noted. Journal.

Due to the high cost of implementing generative artificial intelligence, companies are increasingly wondering whether it is possible to achieve a positive return on the investment. As I wrote last AprilCould more than one trillion be invested in technology and a killing app for AI-Chatbots has not yet appeared.

Therefore, companies are excited about the prospect of lowering the necessary investment. Since the R1’s Open Source model works so well and is so much cheaper than those from Openai and Google, companies are very interested.

How then? R1 is the top trend model downloaded on HuggingFace – 109,000, according to VenturebeatAt and matches “Openai’s O1 to only 3% -5% of the price.” R1 also provides a search function that users consider to be superior to Openai and Perplexity “and is only competing with Google’s Gemini Deep Research,” noted Venturebeat.

Deepseek developed R1 faster and at a much lower price. Deepseek said it trained one of its latest models for $ 5.6 million in about two months, noted CNBC – Far less than the $ 100 million to $ 1 billion range, as anthropic CEO Dario Amodei mentioned in 2024 as the cost of training its models, Journal reported.

To train his V3 model, Deepseek used a cluster of more than 2,000 nvidia chips “compared to tens of thousands of chips for similar size training models,” noted Journal.

Independent analysts from the Chatbot Arena, a platform hosted by UC Berkeley scientists, rated V3 and R1 models in the top 10 for chatbot performance on January 25. Journal wrote.

The CEO of Deepseek is Liang Wenfeng, which manages a $ 8 billion hedge fund. His hedge fund, named High-Flyer, used AI chips to build algorithms to identify “patterns that could affect stock prices,” noted. Financial Times.

Liang’s outsider status helped him succeed. By 2023, he launched Deepseek to develop AI at human level. “Liang built a unique infrastructure team that really understands how the chips worked,” a founder of a rival LLM company said. Financial Times. “He took his best people with him from the hedge fund to Deepseek.”

Deepseek gained advantage when Washington banned Nvidia from exporting H100s – Nvidia’s most powerful chips – to China. It forced local AI companies to construct about the scarcity of the limited computing power of less powerful local chips – nvidia H800S, according to CNBC.

The H800 chip transfers data between chips with half of the H100’s 600 gigabit per day. second and is generally cheaper, according to one Medium Post of NScale Chief Commercial Officer, Karl Havard. Liang’s team “already knew how to solve this problem,” noted Financial Times.

Microsoft is very impressed with DEEPSEEK’s results. “Watching the new Deepseek model, it’s super impressive in terms of how they really effectively made an open source model that makes this inference time calculation and is super-calculating effective,” said CEO Satya Nadella on January 22 at. World Economic Forum according to one CNBC report. “We should take the development out of China very, very seriously.”

Will Deepseeks breakthrough slow down the growth in demand for NVIDIA chips?

Deepseek’s success should spur changes to US AI policy while making the NVIDIA investors more cautious.

US export restrictions to Nvidia put pressure on startups like Deepseek to prioritize efficiency, resource pooling and collaboration. To create R1, Deepseek reconstructed his training process to use NVIDIA H800S ‘lower treatment speed, former Deepseek employee and current PhD student told in computer science at Northwestern University Zihan Wang My Technology Review.

An Nvidia scientist was excited about Deepseek’s results. Deepseek’s paper reporting the results brought back memories of pioneering AI programs that mastered board games such as chess built “from the bottom, without first imitating human major masters,” said senior nvidia scientist Jim Fan at X as mentioned by Journal.

Will Deepseek’s success inhibit Nvidia’s growth rate? I don’t know. But based on my research, companies clearly want powerful generative AI models that pay off. As companies seek generative AI applications with high dividends, they will be able to perform more experiments if the cost and time to build these applications is lower.

This is why R1’s lower costs and shorter time to perform well to continue to attract more commercial interest. A key to Deepseek’s ability to deliver what companies want is its ability to optimize less powerful GPUs – which cost less than the latest chips.

If more startups can copy what Deepseek has achieved, there may be less demand for Nvidia’s most expensive chips.

I don’t know how Nvidia will react if this happen. But in the short term, this could mean less revenue growth, as startups after Deepseeks strategy build models with fewer chips at lower prices.