DeepSeek Took a Bite Out of the Mag 7. Here's What Happens Next
Dow Jones
$Meta Platforms, Inc.(META)$
The so-called Magnificent Seven group of tech stocks have spent the past two years looking invincible, so it's ironic that the companies worth $18 trillion in total market value were undercut this past week by a Chinese start-up no one on Wall Street had heard of just days earlier.
Such is the power -- and vulnerability -- of being on the cutting edge.
The new R1 artificial-intelligence model from China's DeepSeek has comparable performance to some of the world's best models, and seemingly comes at a lower price to train and run in the cloud.
There are many reasons to be skeptical of R1, beginning with the fact that the Chinese government may be subsidizing R1's creation and deployment. But DeepSeek has made the model open source and produced two white papers detailing its methods, and AI labs in the U.S. and allied countries are busy trying to replicate what DeepSeek did.
If they can, it may signal the beginning of a new, much lower cost structure for AI, and this means more risk for the companies that are spending hundreds of billions on capital expenditures for new AI data centers.
Amazon.com, Microsoft, Google parent Alphabet, and Meta Platforms have spent $212 billion on capex over the past 12 months, up 45% from the year before.
Microsoft has projected $80 billion this year, and Meta, $60 billion to $65 billion.
Investors spent most of the week wondering what happens if all that spending goes for naught.
Here's our breakdown of what DeepSeek means for the Mag 7:
Apple
Apple has the most to win from a low-cost AI world. The company has been frequently criticized in recent years for spending less on AI infrastructure and cloud data centers than its large rivals and lagging in model development. The company's reticence to spend could now prove prescient.
"I think innovation that drives efficiency is a good thing," Apple CEO Tim Cook told investors about DeepSeek on the company's earnings call this past week. "That's what you see in that model. Our tight integration of silicon and software will continue to serve us very well."
Cook defended its more parsimonious AI strategy, saying Apple has taken a "prudent" approach to capital expenditures for a reason.
If DeepSeek's optimization techniques prove to be replicable by U.S. AI vendors, the rise of smaller, more efficient models could be positive for Apple and other hardware companies. It means the timeline for more powerful models that can run on local devices like smartphones and laptops may have been pulled forward. If large-language models can run entirely on an iPhone, it's a win for Apple and a loss for the larger cloud platforms.
Amazon.com
It's the cloud, in fact, that looks most vulnerable from the DeepSeek development. The ability to run models on local servers may reduce the profit margins of the cloud. That could be particularly problematic for Amazon, the pioneering public cloud firm, whose Amazon Web Services still leads the market.
Amazon will benefit from cheaper AI across its retail e-commerce business, allowing for more personalization and better recommendations. But investors traditionally buy Amazon for its high-growth and high-margin cloud business. Cheaper AI run outside of the cloud represents a challenge. Amazon reports earnings next week, and DeepSeek is sure to be a popular topic on the company's earnings call.
NVIDIA
No one has benefited from the AI trade more than Nvidia, so it isn't a surprise that the stock took the biggest hit of the Mag Seven group this past week. The stock tumbled 17% on Monday, before regaining some of the losses as the week went on. DeepSeek's suddenly efficient AI raised concerns that Nvidia's big advantage -- its powerful Blackwell graphics processing units for AI training and inference -- could become its main vulnerability if an AI chip shortage turns into a glut.
Nvidia recently started shipping the GB200 NVL72 AI server system, which combines 72 Blackwell GPUs for unrivaled AI computing power.
But investors are missing the big picture when it comes to Nvidia. In fact, greater efficiency in driving down the cost of AI is exactly what Nvidia wants, as it will encourage adoption and enable innovative applications. Over the past week, many technology executives referenced Jevons paradox, which says that increases in a technology's efficiency tend to increase consumption rather than lower it.
Tesla Motors
DeepSeek could also be positive for Tesla. As the company's disappointing earnings demonstrated this past week, Tesla is no longer seen as a car company. Rather, it's viewed as an AI beneficiary, particularly as artificial intelligence moves out of the digital world and into the physical one, writes Morgan Stanley analyst Adam Jonas.
"While the journey may be volatile and nonlinear, we believe 2025 will be a year where investors will continue to appreciate and value these existing and nascent industries of embodied AI where we believe Tesla has established a material competitive advantage," Jonas wrote.
Tesla is mainly about applying AI, not building it, says Greg Martin, managing director at Rainmaker Securities, a platform that facilitates transactions in privately held companies, including Elon Musk's xAI. As costs to develop AI drop, there is a shift in value from those developing the foundational models to those building applications. Tesla is essentially a vertically integrated application developer. It won't mind cheaper AI.
The company said in its fourth-quarter earnings report that its AI computing power more than quintupled in 2024. That didn't come cheap. Tesla spent $11.3 billion on plants and equipment and $4.5 billion on research and development. DeepSeek has the potential to drive those costs lower.
Microsoft, Alphabet, and Meta Platforms, Inc.
That leaves us with Microsoft, Google, and Meta -- the largest spenders in the AI arms race after Amazon. Not to fret -- they could overspend and still wind up winners.
Microsoft's cloud segment has been growing quickly, with revenue up 19% last quarter, but its business software unit is still the star, accounting for 42% of revenue and 53% of operating profit. The centerpiece of Microsoft's AI strategy for business software is Microsoft 365 Copilot, the $360 a year add-on to Office apps like Word and Excel. Until now, the cost of those apps has been prohibitive for many. With a lower AI cost structure, Microsoft could reduce the price significantly, speed up adoption, and get to AI profit far more quickly
Revenue from Alphabet's Google's cloud unit grew by 35% last quarter, and R1 brings new risk for that unit. But ad-supported services are still 75% of its revenue, and these stand to benefit. Search can be enhanced with natural-language queries, more relevant results, better AI summaries, and improved ad targeting. YouTube can serve up better video feeds in Shorts, its TikTok competitor, and target ads more personally. Cheap AI is a benefit across all of Google services.
Meta is in a similar place, with most of its revenue from ads. It's unique within the Mag Seven in that it's spending a lot on AI data centers, but just for its own use, not to rent out in the cloud. It's a risk not having rental income to offset a potential overspend. But regardless of spending levels, Meta will be able to use less expensive AI to increase engagement and enhance ad targeting, and the balance is likely to work out well for the company.
Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.
#nvda #Magnificent7 #ai #usmarket #spy #qqq #Netflix #Google
Comments