China’s DeepSeek AI Models Raise Doubts

The Wall Street Journal reports China’s DeepSeek AI Models Raise Doubts Over U.S. Tech Dominance

Global chip stocks slumped Monday after Chinese artificial-intelligence company DeepSeek said it had developed AI models that nearly matched American rivals despite using inferior chips, raising questions about the need to spend huge sums on advanced gear provided by Nvidia and other tech giants to train AI models.

DeepSeek said last week that the performance of its latest R1 model was on par with OpenAI’s o1-mini model that the ChatGPT maker released in September. The announcement came after DeepSeek said in a late-December report that it used a cluster of more than 2,000 Nvidia chips to train its other V3 model, compared with the tens of thousands of chips that are normally used for training models of a similar size.

The company said training one of its latest models cost $5.6 million, compared with the $100 million to $1 billion range cited last year by Dario Amodei, chief executive of AI company Anthropic.

“There is a new AI challenger in town and investors are spooked at what they’ve discovered,” AJ Bell investment director Russ Mould wrote in a note to clients. “Its assistant is free to use and runs off lower-cost chips and less data—implying a major challenger to the established AI names in the West.”

DeepSeek’s perceived success risks intensifying the AI war between the U.S. and China, Quilter Cheviot’s Ben Barringer said in a note to investors, particularly after the recent announcement of Stargate—a joint venture between OpenAI, SoftBank Group, Oracle and MGX to build data centers in the U.S. for OpenAI.

News Corp, owner of Dow Jones Newswires and The Wall Street Journal, has a content-licensing partnership with OpenAI.

Is AI Spending Justified?

AI spending is just one of the concerns that Sends U.S. Tech Stocks Reeling.

The sudden popularity of a Chinese artificial intelligence model called DeepSeek pummeled stocks Monday, with the tech-focused Nasdaq composite index down 2.7 percent in morning trading.

Several tech companies that have banked on a surge of AI interest sold off, with U.S. chipmaker Nvidia down more than 12 percent at the opening bell. Microsoft lost around 3.5 percent, Meta Platforms lost 1.6 percent, and Oracle dropped more than 7 percent.

Analysts said the sell-off underscores anxieties about whether the massive spending on artificial intelligence ― and the specialized chips, data centers and related power infrastructure ― are justified. Nvidia has exploded in value in recent years because it dominates the market for the chips at the center of the global AI race. It’s now one of three companies with a market capitalization above $3 trillion.

But some analysts have raised questions about how soon those investments will translate into straightforward moneymaking use cases for U.S. companies. The emergence of a low-cost Chinese competitor adds to skepticism over those investments.

The emergence of the model from China has escalated concerns about the U.S. adversary’s tech prowess amid a global race to develop advanced artificial intelligence. China’s advancements in the technology are viewed as a national security threat to the United States, and the success of DeepSeek has only exacerbated those concerns.

DeepSeek is a China-based start-up that last week launched a free AI assistant that it says can operate at a lower cost than American AI models like ChatGPT. The company was founded in 2023 by Liang Wenfeng, co-founder of the hedge fund High-Flyer. By Monday, it had rocketed to the top of downloads in the Apple Store.It also reported outages and limited registrations due to “large-scale malicious attacks” amid the uptick in interest.

President Donald Trump has said that the United States needs to remain competitive with China in developing artificial intelligence, and he appeared focused on the technology during his first week in office. He told reporters Saturday that he was considering using emergency powers to provide the “tremendous energy” that U.S. companies need to develop AI models.

We’re already leading,” Trump said on Air Force One. “Very shortly, we’re going to be leading by a lot.”

Longtime technology investor Marc Andreessen, a Trump ally, called DeepSeek’s AI model “one of the most amazing and impressive breakthroughs I’ve ever seen” and “a profound gift to the world” in a post on X.

Leading tech firms have spent billions building out artificial intelligence technology for sale to large businesses. Meta chief executive Mark Zuckerberg said in a Facebook post last week that his company plans to invest between $60 billion and $65 billion on AI and build a massive data center. OpenAI, Oracle and SoftBank recently announced a Trump-supported joint venture, called Stargate, that seeks to spend up to $500 billion building out data centers to support AI projects.

China’s DeepSeek Turns AI on its Head

Sherwood reports China’s DeepSeek turns AI on its Head

DeepTrouble
The weekend buzz around the large language model — the fact that it “thinks” before it speaks, shows its workings, and matches OpenAI’s most powerful model, the o1, on a range of metrics — seems to have left much of Silicon Valley wowed and worried, in almost equal measure.

Per DeepSeek’s own figures, the R1 model outperforms the OpenAI o1 on a variety of key tests, shining particularly brightly in math, where it beats the latest model from Sam Altman’s company on three different tests. While it’s less consistent on coding and language tests — it fared particularly badly on the “SimpleQA” (not shown in chart above), a test evaluating the simple factual accuracy of the info that LLMs spit out — the differences are fairly slim, making the cost-effective R1 look impressive.

The Chinese company’s slimmed-down training costs, use of cheaper chips, API, and open-source model have hauled the endless drive for more chips and compute that’s driven much of the market for the last 18 months into question. MetaMETA $649.93 (0.38%), for example, is planning to spend more than $60 billion on capital expenditures just this year.

At a time when people are wondering if we can trust TikTok due to Chinese government ties, many have similar questions about DeepSeek. Tech evangelist Marc Andreessen was among those singing R1’s praises over the last few days — though he may not have asked it about Tiananmen Square yet.

Another Sanction Failure

Biden placed numerous exports bans on chip technology to prevent this from happening.

Instead DeepSeek said in a late-December report that it used a cluster of more than 2,000 Nvidia chips to train its AI.

No one should be surprised by this.

To Those Hard of Learning, Here’s a Repeat Lesson on Why Sanctions Fail

On September 26, 2024, I commented To Those Hard of Learning, Here’s a Repeat Lesson on Why Sanctions Fail

Let’s discuss a claim that sanction failures are due to a lack of political will.

Robin Brooks on X: “When someone tells you that sanctions can’t and won’t work, that’s basically pro-Russian propaganda. Are we seriously to believe that nothing can be done to stop the shameful flood of transshipments to Russia via Central Asia? Come on. This is just about a lack of political will.“

I am pretty sure that “someone” is me because we have gone round and round on this.

When someone tells you that sanctions do work. Ask them for evidence.

The above post was on oil-related sanctions. The next article is how and why chip sanctions failed.

On August 26, 2024, I commented China Gains Secret Access to Nvdia Microchips by Renting Computers

The US has blocked export of Nvdia chips to China. But where there’s profit, there’s a way.

Know Your Customer’s Customer’s Customer

China sets up an AI company in Singapore. AI developers buy cloud time through a subsidiary that further masks the operation by paying in Bitcoin.

In turn, the subsidiary buys time from a company Dubai or Singapore that hosts the servers.

US politicians are outraged. But some of us are amused knowing full well that sanctions don’t work. So instead of cloud profits going to US corporations, the profits go to Saudi Arabia, Singapore, Dubai, and South Korea.

Only Amazon is forced to “know your customer”.

Where Art Thou EU?

Trump secured pledged to spend $500 billion on data centers.

However, If the benchmarks are accurate, China caught up on AI development by renting NVDIA chips.

Meanwhile, the EU is doing nothing but attempting to regulate AI to death.

For discussion, please see The EU Will Suffer Because it Is Hostile Territory for the AI industry

More By This Author: