Tom Yeung here with today’s Smart Money.
Since OpenAI launched ChatGPT in November 2022, everything in tech has been about more.
More computing power…
More chips…
More electricity…
More resources…
This quest for “more” has slowed the progress of AI development. New models take months to train. Cloud computing firms wait around for deliveries of next-generation AI servers, which can cost in the millions. And many AI data centers wait as long as seven years for hookups to the electrical grid.
When you have a hungry, hungry AI architecture, it can take a long time to feed.
That was until Americans got their hands on DeepSeek R1, a Chinese-developed large language model (LLM), over the weekend.
The open-source software performs almost as well as OpenAI’s latest model, o1, and was supposedly trained with a tiny $6 million budget – a fraction of what American companies have spent. What’s more, some online commentators believe DeepSeek’s model costs one-sixth of OpenAI’s to run. (OpenAI keeps its costs a secret.)
And perhaps most shocking of all, the Chinese startup built R1 without the latest chips or hardware due to chip sanctions from America.
In other words, DeepSeek is doing more with less.
As such, the major U.S. AI players were treated to a rude awakening this week. On Monday, shares of Nvidia Corp. (NVDA), the world’s leading AI chipmaker, sank 17%, while suppliers like Astera Labs Inc. (ALAB) dove more than 30%.
Suddenly, the leading edge of AI architecture has gone on a diet.
Tech giants previously planned to spend $50 billion to $150 billion on AI infrastructure. These figures now seem comically large after OpenAI’s best models were almost beaten by a Chinese startup with – as Andrej Karpathy, the former head of AI at Tesla Inc. (TSLA), put it – “a joke of a budget.”
So, in today’s Smart Money, let’s discuss how DeepSeek achieved this feat.
And since massive AI investments may now be unnecessary, we’re going to start seeing models improve faster than ever before. This will have major implications for our investments… and our livelihoods. So, I’ll also share how you can best prepare for the incredible technological changes to come.
Let’s take a look…
Going Deeper on DeepSeek
DeepSeek likely built R1 by turning to a “Mixture of Experts” (MoE) model, a more efficient form of machine learning.
Imagine an LLM as a human brain (which is incidentally how the technology was conceived). These AI models have billions of “neurons” and adjust the strength of their connections to learn. It’s much like how human brains work.
In dense LLM models, every neuron is connected. These connections link the LLM to an external data source, which allows the AI system to process information from that source. These connections are essentially a bridge to the LLMs knowledge. However, while every neuron is connected, not every connection is used. This creates a problem of having too many connections.
This is where DeepSeek differs. Its MoE uses a “gate network” (or router) to determine which connections must be made. This significantly reduces the connections required, because only a small subset of neuron sections is used for each prompt.
That means fewer connections for chips to calculate, and slashing the costs involved.
This is what turned DeepSeek into an overnight success. On Monday, the firm’s AI Assistant app surged to the top of Apple’s App Store, leaving OpenAI’s ChatGPT in second place.
However, the tech giants will not be content to fall behind. Over the next several months, American AI developers will scrutinize DeepSeek’s open-sourced software and copy everything they can. It only took three months for Google’s Gemini model to incorporate the “thinking” concept used by OpenAI’s o1 model to solve logic problems.
So, this “Sputnik moment” will cause tech firms to work even harder now that their dominance is on the line.
In any case, we know that DeepSeek R1 is very good. We are treating this new competitor as a wake-up call to American LLM developers… and a warning bell that we could arrive at artificial general intelligence (AGI) faster than previously expected.
AGI Is Coming… and Coming for Your Jobs
AGI refers to AI technology that has reached human-like intelligence. Most importantly, it has the ability to solve problems through reasoning, just as we do.
Last September, Eric began a 1,000-day countdown to AGI following OpenAI’s release of a series of AI models designed to reason instead of recognize patterns.
The launch of DeepSeek signals that AGI remains on track to reach that 1,000 day target. We’re soon going to see super-intelligent AI begin making better versions of itself, which will make even better versions, and so on.
Here’s why this is so important…
By now, you’ve heard us warn that AI is coming for people’s careers. Accounting, legal research, medicine, even high-powered decision-makers near the C-suite could eventually be outsourced to superintelligent AIs.
In the workplace, AGI is like the equivalent of having a human you could hire, or compete against, as a coworker. Once AI hits this tipping point, there’s no reason a model couldn’t develop even better versions of itself to replace humans in almost any role.
That’s going to be disruptive, to say the least.
The world will suddenly see a split between the “haves” on the right side of this technological divide who control this technology… and the “have-nots” being replaced by it.
Entire sections of the population will become multimillionaires, thanks to timely investments in AI firms. Others will be employed by companies that know how to use tech.
Meanwhile, millions more will lose their jobs overnight – replaced by the very same AI that was once seen as a novelty.
Many experts initially thought we had time to prepare. DeepSeek changes that.
That’s why I believe investors only have a tiny window to prepare during this “pre-AGI” moment before it’s too late. It’s a period where we know certain things will happen before the curtain of AGI comes down.
Luckily, Eric has identified several companies that are strategically positioned to capitalize on AGI’s imminent arrival, and he wants to share them with you.
You can learn more about these companies in Eric’s free 1,000 Days to AGI special broadcast.
Click here to access the special presentation.
Regards,
Thomas Yeung
Markets Analyst, InvestorPlace
