“You have seen what DeepSeek has done — $5.5 million and a very very powerful model,” IT Minister Ashwini Vaishnaw said on Thursday, responding to criticism New Delhi has received for its own investment in AI, which has been much less than many other countries.

Since 2020, India has banned more than 300 apps and services linked to China, including TikTok and WeChat, citing national security concerns. The approval to allow DeepSeek to be hosted in India appears contingent on the platform storing and processing all Indian users’ data domestically, in line with India’s strict data localization requirements.

“Data privacy issues regarding DeepSeek can be addressed by hosting open-source models on Indian servers,” Vaishnaw said at an industry conference.

DeepSeek’s models will likely be hosted on India’s new AI Compute Facility. The facility is powered by 18,693 graphics processing units (GPUs), nearly double its initial target — almost 13,000 of those are Nvidia H100 GPUs, and about 1,500 are Nvidia H200 GPUs. Around 10,000 GPUs are ready to be deployed, and the facility is scheduled to begin operations “in the coming days,” according to the minister.

The facility will also offer computing services at steep discounts to firms in India. Vaishnaw said standard AI computing would be offered at a 42% discount to market rates, and high-precision computing would be discounted by 47%.

The minister’s remarks come a day after DeepSeek’s eponymous app was taken off Apple’s and Google’s app stores in Italy, after that country’s data protection regulator said it was asking how the Chinese firm was using and storing Italians’ personal data.

The release of DeepSeek’s R1 “reasoning” model, built on a purportedly modest budget, sent shockwaves through the tech industry this week, causing chip giant Nvidia’s market cap to decline by $600 billion. The model has quickly come under intense scrutiny, and has sparked heated debates around copyright issues, U.S. export controls, and how even more money needs to be poured into AI efforts.

Beyond hosting foreign AI models, India is also trying to drive development of AI models and related technology on its own turf. “Major chip designers are willing to work with India to develop indigenous GPUs,” Vaishnaw said.

Vaishnaw estimated that India would see investment of $30 billion in hyperscalers and data centres over the next two to three years. One of the country’s biggest conglomerates, Reliance, is planning to build what could become the world’s largest data center in the city of Jamnagar, with a capacity of 3 gigawatts, Bloomberg reported last week. 

“We believe there are at least six major developers who can develop AI models in six to eight months on the outer limit, and four to six months on a more optimistic estimate. A common compute facility is the most important component for creating a robust AI ecosystem,” Vaishnaw said.

The computing facility will also support India’s broader AI initiatives. Vaishnaw said 18 AI-driven applications focusing on agriculture, climate change, and learning disabilities have been selected for initial funding.

To oversee development of these AI initiatives, India will establish a regulatory body using what Vaishnaw described as a “hub-and-spoke model,” allowing multiple institutions to collaborate on safety frameworks. “We will be keeping our models open and application-focused,” he said.

Chinese AI lab DeepSeek broke into the mainstream consciousness this week after its chatbot app rose to the top of the Apple App Store charts. DeepSeek’s AI models, which were trained using compute-efficient techniques, have led Wall Street analysts — and technologists — to question whether the U.S. can maintain its lead in the AI race and whether the demand for AI chips will sustain.

But where did DeepSeek come from, and how did it rise to international fame so quickly?

DeepSeek’s trader origins

DeepSeek is backed by High-Flyer Capital Management, a Chinese quantitative hedge fund that uses AI to inform its trading decisions.

AI enthusiast Liang Wenfeng co-founded High-Flyer in 2015. Wenfeng, who reportedly began dabbling in trading while a student at Zhejiang University, launched High-Flyer Capital Management as a hedge fund in 2019 focused on developing and deploying AI algorithms.

In 2023, High-Flyer started DeepSeek as a lab dedicated to researching AI tools separate from its financial business. With High-Flyer as one of its investors, the lab spun off into its own company, also called DeepSeek.

From day one, DeepSeek built its own data center clusters for model training. But like other AI companies in China, DeepSeek has been affected by U.S. export bans on hardware. To train one of its more recent models, the company was forced to use Nvidia H800 chips, a less-powerful version of a chip, the H100, available to U.S. companies.

DeepSeek’s technical team is said to skew young. The company reportedly aggressively recruits doctorate AI researchers from top Chinese universities. DeepSeek also hires people without any computer science background to help its tech better understand a wide range of subjects, per The New York Times.

DeepSeek’s strong models

DeepSeek unveiled its first set of models — DeepSeek Coder, DeepSeek LLM, and DeepSeek Chat — in November 2023. But it wasn’t until last spring, when the startup released its next-gen DeepSeek-V2 family of models, that the AI industry started to take notice.

DeepSeek-V2, a general-purpose text- and image-analyzing system, performed well in various AI benchmarks — and was far cheaper to run than comparable models at the time. It forced DeepSeek’s domestic competition, including ByteDance and Alibaba, to cut the usage prices for some of their models, and make others completely free.

DeepSeek-V3, launched in December 2024, only added to DeepSeek’s notoriety.

According to DeepSeek’s internal benchmark testing, DeepSeek V3 outperforms both downloadable, openly available models like Meta’s Llama and “closed” models that can only be accessed through an API, like OpenAI’s GPT-4o.

Equally impressive is DeepSeek’s R1 “reasoning” model. Released in January, DeepSeek claims R1 performs as well as OpenAI’s o1 model on key benchmarks.

Being a reasoning model, R1 effectively fact-checks itself, which helps it to avoid some of the pitfalls that normally trip up models. Reasoning models take a little longer — usually seconds to minutes longer — to arrive at solutions compared to a typical non-reasoning model. The upside is that they tend to be more reliable in domains such as physics, science, and math.

There is a downside to R1, DeepSeek V3, and DeepSeek’s other models, however. Being Chinese-developed AI, they’re subject to benchmarking by China’s internet regulator to ensure that its responses “embody core socialist values.” In DeepSeek’s chatbot app, for example, R1 won’t answer questions about Tiananmen Square or Taiwan’s autonomy.

A disruptive approach

If DeepSeek has a business model, it’s not clear what that model is, exactly. The company prices its products and services well below market value — and gives others away for free.

The way DeepSeek tells it, efficiency breakthroughs have enabled it to maintain extreme cost competitiveness. Some experts dispute the figures the company has supplied, however.

Whatever the case may be, developers have taken to DeepSeek’s models, which aren’t open source as the phrase is commonly understood but are available under permissive licenses that allow for commercial use. According to Clem Delangue, the CEO of Hugging Face, one of the platforms hosting DeepSeek’s models, developers on Hugging Face have created over 500 “derivative” models of R1 that have racked up 2.5 million downloads combined.

DeepSeek’s success against larger and more established rivals has been described as “upending AI” and ushering in “a new era of AI brinkmanship.” The company’s success was at least in part responsible for causing Nvidia’s stock price to drop by 18% on Monday, and for eliciting a public response from OpenAI CEO Sam Altman.

As for what DeepSeek’s future might hold, it’s not clear. Improved models are a given. But the U.S. government appears to be growing wary of what it perceives as harmful foreign influence.

TechCrunch has an AI-focused newsletter! Sign up here to get it in your inbox every Wednesday.

A version of this story appeared in CNN Business’ Nightcap newsletter. To get it in your inbox, sign up for free here.

New York
CNN
 — 

The story of AI in the 2020s has gone something like this:

Sam Altman: Look, a toy that can write your book report.
VCs: This will fix everything!
Doomers: This will ruin everything.
Tech: We need money!
Everyone else: Could we maybe not destroy the environment over this?
Tech: Let’s restart Three Mile Island.
Tech: We need money!
Wall Street: Where’s our return?
Tech: (Chants) More power! More power! More power!

And finally, in the year 2025, here comes DeepSeek to blow up the industry’s whole narrative about AI’s bottomless appetite for power, and potentially break the spell that had kept Wall Street funneling money to anyone with the words “harnessing artificial intelligence” in their pitch deck.

ICYMI: DeepSeek dropped a bomb known as R1 that’s got all of Silicon Valley and much of Wall Street in a tizzy.

The Chinese company’s large language model is basically a cheaper, more efficient ChatGPT, built on a fraction of OpenAI’s budget and using far fewer chips than any other leading chatbot.

“That is a massive earthquake in the AI sector,” Gil Luria, head of tech research at investment group D.A. Davidson, told me. “Everybody is looking at it and saying, ‘We didn’t think this is possible. And since it is possible, we have to rethink everything that we have been planning.’”

Suddenly, all that money and computing power that the Sam Altmans, Mark Zuckerbergs and Elon Musks have been saying are crucial to their AI projects — and thus America’s continued leadership in the industry — may end up being wildly overblown.

DeepSeek, which on Monday climbed to No. 1 on the Apple app store, claims to have built its base model for less than $6 million (versus the more than $100 million Altman has said it cost to build GPT-4).

It also claims to have used just 2,000 Nvidia chips that it obtained before US export restrictions were put in place. (OpenAI says it used 25,000 of the more powerful Nvidia H100 chips to build GPT-4.)

It’s awkward timing for the Trump administration, which last week announced a half-trillion-dollar private-sector investment to build more data centers and keep the United States ahead of China in the AI race. (Oops!)

And it’s incredibly bad news for Nvidia, the American chip maker powering the AI gold rush. Nvidia shares sank 17% Monday, shedding $600 billion in market cap in a single session — the biggest one-day loss for a single stock in history. Alphabet, Microsoft, Oracle, TSMC and plenty of others sank, and because tech stocks are so dominant, that dragged the broader stock market down, too.

The tech-heavy Nasdaq plunged by 3% and the broader S&P 500 fell 1.5%. (The Dow, buoyed by health care and consumer companies, ended the day up less than 1%.)

Of course, one bad day on Wall Street does not an apocalypse make. (That’s for later, when one of these AI labs creates superintelligent murder bots. Kidding! Kind of.)

But DeepSeek is forcing investors to take a beat and question tech companies’ assumptions. By its own reasoning, the AI industry needed to keep increasing “compute” (or computational power), which meant buying tens of thousands of Nvidia’s state-of-the-art chips and building giant data centers.

“DeepSeek makes it very clear that that the current trajectory of scaling up of data centers is highly unlikely to be economic to Nvidia’s customers,” Luria said.

The AI industry, and OpenAI in particular, has been going down two paths at once.

There’s the business of designing AI models with better algorithms and sounder reasoning — the kind of stuff that requires “finesse, as opposed to brute force,” Luria says. And then there’s the Stargate path of giant energy investments.

The first task is still “valid and important,” while the second path looks “ridiculous,” Luria said. “DeepSeek makes it clear that that scale and that spend would be, at the very least, wasteful.”

In other words, AI isn’t dead. But the landscape is shifting faster than anyone, perhaps most of all Nvidia, expected.

Picks and shovels

Nvidia has become the ultimate “picks and shovels” play on Wall Street, transforming it into a $3 trillion company in the span of a couple of years. Up until now, the demand for Nvidia chips appeared boundless — tech companies were going to keep gobbling them up faster than Nvidia could produce them.

But if DeepSeek really did manage to build a ChatGPT competitor using a handful of old processors, then maybe Nvidia’s tech customers soon won’t need as many as they’d thought. Many on Wall Street seemed to think so Monday as the stock went into a tailspin. (For its part, Nvidia seemed to shrug at the selloff in a statement to Bloomberg, calling DeepSeek’s model an “excellent AI advancement” that “illustrates how new models can be created.”)

It’s also good to keep in mind that Wall Street is prone to tantrums, which is how some tech investors chalked up Monday’s selloff.

“At the end of the day, there is only one chip company in the world launching autonomous, robotics, and broader AI use cases, and that is Nvidia,” Wedbush analysts wrote in a letter to clients. “Launching a competitive LLM model for consumer use cases is one thing… launching broader AI infrastructure is a whole other ballgame, and nothing with DeepSeek makes us believe anything different.”