The artificial intelligence races have been heating up, with most of the Magnificent Seven and others spending huge amounts of money in a race to achieve the best artificial intelligence models the fastest.
But that strategy has raised doubts about whether these massive sums will eventually pay off.
Two weeks ago, a collection of major U.S. AI companies announced a $100 billion project called Stargate. Yet the following week, low-cost Chinese model DeepSeek R1 was introduced, showing that leading frontier models can be trained at much lower cost.
An ancillary conclusion from the DeepSeek introduction? It appeared to validate Apple's (NASDAQ: AAPL) AI strategy, which had come under criticism over the past couple years. As it turns out, Tim Cook & Co. may have the last laugh after all.
The implications of DeepSeek
While there is still some controversy as to the amount of money DeepSeek actually spent on its model, whether it "distilled" its model from OpenAI's o1, or whether it has access to more GPUs than it lets on, the optimizations published as part of its research paper revealed real, novel ideas on how to lower the cost of training frontier models.
There are a few consequences investors can anticipate from this. First, look for those ideas to get incorporated into all the major AI competitors quickly. This will further lower the cost of AI models across the industry.
Second, DeepSeek is partially open-source, which means its weights and technology optimizations are publicly available to developers to use and augment. Of the American tech giants, only Meta Platforms' (NASDAQ: META) Llama models are also open-source.
Having a major American and Chinese company disclosing their model optimizations publicly should lead to more open-source evangelism, which could even further speed up the proficiency of large, open models.
While OpenAI is still thought to have an overall technology lead, the time gap between it and its open-source competitors has closed to just a few months. For his part, Meta CEO Market Zuckerberg just said on last week's conference call with analysts he believes Meta's upcoming Llama 4 models will surpass OpenAI and lead the market this year.
Increased open-sourcing and further algorithm-based cost reductions should eventually lead to more commoditization. Commoditization means prices and margins for AI models should come down, benefiting the customers of those models at the expense of the builders.
Apple was prescient
As the AI races heated up over the past two years, basically all major technology companies dramatically increased their capital expenditures buying high-priced Nvidia(NASDAQ: NVDA) GPUs and data center infrastructure. But one major tech company with the deep pockets to participate in the spending spree chose not to do so: Apple.
All major tech companies grew their capital expenditures between 73% and 182% over the past three years, except for Amazon. But Amazon is a bit different, as it was coming off a major spending spree in its e-commerce infrastructure after the pandemic. So its data center spending for Amazon Web Services has probably climbed much more than its 9% overall growth.
That leaves Apple alone as having surprisingly decreased its capital expenditures over the past two years -- a sharp contrast to its rivals.
If a resource becomes commoditized and abundant, you would want to be a user of that resource, not a producer of it. This is why Apple's approach to focus on using AI for its core customers, all while eschewing the buildout of a massive Nvidia-based infrastructure, was smart.
Apple is still investing in AI
This is not to say Apple isn't investing in its own proprietary models. It is. And actually, its approach looks a lot like DeepSeek's in some ways.
At last June's Apple Intelligence event, Apple unveiled its own models, with two big caveats: One, each model was distilled and optimized for specific use cases on the iPhone or Mac. On its recent conference call with analysts, CEO Tim Cook highlighted some of the features most used on Apple devices since the rollout of Apple Intelligence last October.
[Apple Intelligence users] can use Writing Tools to help find just the right words, create fun and unique images with Image Playground and Genmoji, handle daily tasks and seek out information with a more natural and conversational Siri, create movies of their memories with a simple prompt, and touch up their photos with Clean Up. We introduce visual intelligence with camera control to help users instantly learn about their surroundings.
Aside from focusing on specific Apple customer use cases rather than generalized large language models, Apple's approach also seemed prescient in other ways. Apple Intelligence's base model was built on an open-source framework called Apple's AXLearn. And since it's open-source, Apple also eschewed Nvidia chips and the proprietary CUDA software stack wherever it could, as Apple knew from the start it didn't want to get locked into Nvidia's expensive ecosystem.
While the AI software framework Apple uses is open-source for its base model, Apple has a seemingly contradictory approach when it comes to data. Apple uses only licensed data, with publishers having the opportunity to opt out. Apple also scrapes the internet for publicly available data with its proprietary web crawler. The company turns proprietary again for post-training filtering algorithms and optimization, which are geared for specific tasks on Apple products.
By using open-source software that can run on non-Nvidia chips for the base model, including Apple's own proprietary data center chips, then using proprietary algorithms at the filtering and optimization level for known use cases, Apple can produce smaller yet highly useful models for specific tasks at a very reasonable cost.
Apple's discipline makes it a long-term winner
Following Nvidia's decline after the DeepSeek news, Apple has actually regained the crown as the largest company in the world.
Despite its being the biggest company in the world today, remember, Apple has never pioneered a new technology innovation on its own. It wasn't the first to invent the PC, portable music player, smartphone, or wireless headphones. What Apple did is take those inventions and then innovate on top of them, making them all especially intuitive and useful for users, while consistently optimizing for cost afterwards.
That discipline and lack of urgency to spend huge amounts on uncertain new inventions makes for a very derisked business model. Given that large language models are now set to become lower-cost and more accessible, Apple's discipline has showed why it was Warren Buffett's preferred technology stock over the past 10 years.
Don’t miss this second chance at a potentially lucrative opportunity
Ever feel like you missed the boat in buying the most successful stocks? Then you’ll want to hear this.
On rare occasions, our expert team of analysts issues a “Double Down” stock recommendation for companies that they think are about to pop. If you’re worried you’ve already missed your chance to invest, now is the best time to buy before it’s too late. And the numbers speak for themselves:
Nvidia:if you invested $1,000 when we doubled down in 2009,you’d have $311,343!*
Apple: if you invested $1,000 when we doubled down in 2008, you’d have $44,694!*
Netflix: if you invested $1,000 when we doubled down in 2004, you’d have $526,758!*
Right now, we’re issuing “Double Down” alerts for three incredible companies, and there may not be another chance like this anytime soon.
John Mackey, former CEO of Whole Foods Market, an Amazon subsidiary, is a member of The Motley Fool's board of directors. Randi Zuckerberg, a former director of market development and spokeswoman for Facebook and sister to Meta Platforms CEO Mark Zuckerberg, is a member of The Motley Fool's board of directors. Billy Duberstein and/or his clients have positions in Amazon, Apple, and Meta Platforms. The Motley Fool has positions in and recommends Amazon, Apple, Meta Platforms, and Nvidia. The Motley Fool has a disclosure policy.
