Issue #10, Volume #2
Chinese Startup Rattles Markets… And What’s Next For AI
This is Porter & Co.’s free e-letter, the Daily Journal. Paid-up members can access their subscriber materials, including our latest recommendations and our “3 Best Buys” for our different portfolios, by going here.
| China’s AI startup DeepSeek rattles markets… what it means – and doesn’t mean… what you should do with your money now… Plus, Fed meeting on tap… and consumer credit card delinquencies rise… |
China just fired a trillion-dollar bazooka at America’s big tech monopolies.
Late last week, Chinese artificial intelligence (“AI”) startup DeepSeek shocked the tech world when it announced details of its groundbreaking large language model (“LLM”). DeepSeek achieved the same output that America’s leading AI giants have produced – for example, OpenAI’s ChatGPT and Anthropic’s Claude – with their LLMs, but at a fraction of the cost.
This new, lower-cost LLM threatens to disrupt hundreds of billions in capital investment and trillions of dollars in market value among America’s leading technology firms.
(Confused about what we call the parallel processing revolution… which was launched in a Denny’s restaurant in 1993, when three computer engineers drew up plans for a startup that eventually became Nvidia? We explain it all… see Porter’s presentation here.)
After digesting the news over the weekend, investors sold shares of companies that stand to lose the most if DeepSeek is all that it appears to be. As of 2 pm ET, the tech-heavy Nasdaq Composite was down 3.4%, with even larger losses among individual companies like Nvidia (NVDA), which fell 17.4% to wipe out around $600 billion in market value – the largest single day loss of value of any U.S. stock in history.
So, what exactly happened… and what does it mean for your money?
LLMs are the workhorses underpinning today’s AI revolution, which take user input (via prompts) and generate an answer on virtually any topic imaginable. ChatGPT and its competitors can interpret text, write essays, answer questions, execute complex instructions, create pictures, and generate entire videos. These high-powered algorithms require an exhaustive training process that analyzes virtually all of the written text on the entire internet.
OpenAI, America’s leading LLM provider, analyzed 100 trillion data points (known as parameters) for its latest ChatGPT4.0 model. Crunching through that much data is time consuming and expensive – requiring six months of scanning through vast troves of text and converting human language into numerical representations, and $100 million in computing costs. And that’s just the beginning of the investment involved.
The ongoing computing power needed to run these models requires thousands of specialized computer chips known as GPUs (graphic processing units), which cost around $40,000 each.
This explains why OpenAI has raised a staggering $18 billion (a fundraising round in October valued the company at $157 billion) to fuel the development and ongoing operating costs of its ChatGPT LLM. It’s also why America’s largest technology giants like Alphabet (GOOG), Amazon (AMZN), Microsoft (MSFT), and Meta (META) have been collectively pouring more than $200 billion each year into capital expenditures for new AI-capable data centers.
That spending – and the anticipation of far more, as highlighted by last week’s announcement by President Donald Trump of Stargate AI, a $500 billion AI infrastructure investment plan – has inflated the valuations of the companies driving the parallel processing revolution. Among the biggest beneficiaries has been Nvidia, which (until today) had a market capitalization of around $3.5 trillion.
But every bubble – and there’s no question that U.S. markets, thanks to AI hype, have been in the midst of an all-time bubble – eventually finds its pin. And while it’s too early to say, DeepSeek could be just that, at least for the exploding AI sector of the market. If a small, agile competitor can do the same thing as multi-trillion-dollar companies – at a tiny fraction of the cost, using cheaper, older GPUs – it’s bad news for the giants.
How much cheaper, exactly? In contrast with the $100 million training budget for ChatGPT, DeepSeek spent just $5 million training its current LLM model (or so it claims… there’s an active debate raging about the accuracy of these reported figures). DeepSeek also reduced the computational power required to run its model down to just 2,000 low-tech (i.e., cheaper) Nvidia GPUs – versus the 100,000+ cutting-edge GPUs used to run ChatGPT.
In line with this lower horsepower, DeepSeek is able to run its LLM using 75% less memory than ChatGPT, measured by the number of parameters used (or individual data points used to train and run its model). It does this by breaking down its LLM process into a series of separate computational steps that run in isolation, rather than all at once… and by focusing on producing the correct answer, rather than detailing every logical step, which requires vastly more processing power.
While there is some skepticism regarding DeepSeek’s cost structure, its output is comparable to that of ChatGPT and other leading LLMs. DeepSeek’s LLM model is open source, meaning anyone can peer into the inner workings of the model and verify the accuracy of its outputs. DeepSeek’s open-source model means anyone around the world can replicate DeepSeek’s approach (unlike the proprietary algorithm behind ChatGPT and other top platforms), threatening to level the AI playing field overnight. Applying an open-source approach may be a game changer: famed technology investor Marc Andreessen said “Deepseek R1 is AI’s Sputnik moment,” in a post to X on Sunday.
America’s concerns over China’s AI efforts – DeepSeek is in some ways a nightmare come true – was the motivation of the U.S. imposing export controls on high-end chips. These bans were supposed to cripple China’s access to essential inputs to develop LLMs. The Nvidia chips available to the Chinese market were capped at half the speed of the company’s top products.
It turns out that convoluted export bans didn’t do the trick, and in fact didn’t even matter… and that those constraints may have actually helped DeepSeek – by forcing it to make due with less. The company had to figure out how to train its LLM in a way that wasn’t as taxing on its GPUs.
In other words… U.S. export controls forced Chinese companies to be far more efficient, with what were severely limited computing resources. In a way it was a triumph of capital efficiency… potentially at the expense of investors in AI technology.
Here’s the core of the problem DeepSeek presents for America’s tech giants, and the U.S. stock market. The Chinese company’s pocket-change LLM throws into doubt the hundreds of billions of dollars in capital spending surrounding the current LLM models. If it’s possible to achieve the same LLM output as ChatGPT and others, but at a fraction of the cost, it could decimate the future order book for the suppliers of these projects, like Nvidia – which has powered much of the gains in the U.S. stock market over the past several years. It also means that the hundreds of billions of investment into super-powered data centers could turn into significant excess capacity, and potentially result in untold billions of dollars in asset impairments for the largest U.S. tech companies.
When markets move so fast, and so abruptly, it’s easy to get caught up in the moment. This may be the market bubble’s pin… or it may be the pin for just a select portfolio of AI stocks… or it may just be a pause (or, perhaps, a buying opportunity). Markets anticipate – and sometimes they’re wrong.
And they also overreact. Often the first take (in markets… or anywhere in life…) is incorrect and the early interpretation of winners and losers is just plain wrong. (Concerns over the weekend that DeepSeek would trigger a market-wide selloff, for example, turned out to be only partially true… while the tech-heavy Nasdaq is down substantially, the more analogue Dow Jones Industrial Average was up 0.3% as of 2 pm ET today.)
And also… when investors sell… the baby is often tossed out with the bathwater. When investors sell, they often don’t discriminate between companies that could be decimated by the concern du jour… and those that won’t be touched. And that’s where opportunity lies.
Despite its enormous losses today, Nvidia shares are where they were as recently as October. The company’s market capitalization has nearly doubled over the past year… and is up almost 500% over the past two years. If DeepSeek is the real thing, Nvidia and its AI cohorts could fall a lot more.
For now? Hold on to the cash that you’ve raised (which we’ve been recommending for a while), including the gains we locked in from our basket of Parallel Processing Revolution stocks. Across all six recommendations made for Big Secret on Wall Street paid subscribers last year, we generated an average total return of 11.7% over an average holding period of just over two months, or a 126% annualized average return.
In The Big Secret on Wall Street, we focus heavily on “forever companies.” These are businesses that have dominant competitive positions, and enduring business models generally free from threats of technological obsolescence or other unpredictable shifts in the economy… and which have been around for decades – if not centuries. Think about world-famous brands of popular snack foods that millions of people eat every day – products that don’t stop flying off the shelf, regardless of which AI technology comes and goes.
In this week’s upcoming Big Secret portfolio review, we will dive into three of these companies in particular that have experienced recent declines in their stock prices of 30% to 50% due to temporary hiccups that we believe will soon pass.
For each company, we’ll make the case for why these businesses will recover from their recent setbacks and ultimately return to what they’ve always done historically: dominating their industries for years to come.
In the meantime, Mr. Market’s mistake of selling these shares due to temporary setbacks has provided a tremendous value opportunity. As a result, we’re putting each of these three names into our updated Top 3 “Best Buy” list – making them our three highest conviction ideas in today’s market. If you do not already subscribe to The Big Secret on Wall Street, click here to get access.
Three Things You Need to Know Before We Go…
1. All eyes on the Fed. The U.S. central bank is scheduled to announce its latest interest rate decision at its monthly meeting on Wednesday. Despite growing uncertainty about inflation, burgeoning market turmoil (see above), and pressure from President Trump that “interest rates drop immediately,” the Fed is expected to hold rates steady at 4.50%. The CME’s FedWatch tool puts the odds of another 25 basis point cut at Wednesday’s meeting at just 2.7%.
2. Growing consumer credit stress. The percentage of active credit card account holders making just the minimum payment rose to 10.8% in the third quarter of 2024, according to the Federal Reserve – a 38% increase over the past four years, to the highest level of those making just the minimum payment since at least 2012. Credit card 30-plus day delinquencies have also doubled over the past four years to 3.5%.
3. Coming this week: GDP and inflation numbers… and a special new Spotlight. Thursday brings Q4 GDP numbers – a snapshot of the health of the U.S. economy – and the economy is projected to have grown 2.0%, down from 3.1% growth in Q4 2023. On Friday, we’ll see December’s PCE (personal consumption expenditures) data, which tracks inflation based on prices for consumer goods and services. Headline PCE, which includes food and gas, is expected to increase by 2.6% year over year (compared to a 2.4% increase last December), while core PCE (excluding food and gas) should remain steady at 2.4%.
And, on Wednesday, expect something special in your inbox… Our Spotlight, where we highlight an analyst or service outside of Porter & Co. that we think is worthy of your attention – and we provide research that’s usually reserved for that analyst’s subscribers – normally comes out on Tuesday. (See here for our most recent Spotlight.) But this week we have a special Spotlight release… and it has to wait until Wednesday. So stay tuned…
And one more thing…
Poll Results… How Aggressive Will Trump Be With Tariffs?
In Friday’s Daily Journal, Porter was critical of President Trump’s plans to impose tariffs on America’s trading partners. Our survey on Friday asked readers: “How aggressive do you think the Trump administration will be on tariffs?” The results were nearly unanimous: 81% of survey takers selected the option “Strategic and targeted and used mostly as a negotiation tool,” while 16% voted for “Batten down the hatches… it’s Fordney-McCumber all over again,” referring to the 1922 tariff law, mentioned in Friday’s essay, that decimated the U.S. economy. Just 4% feel that “It’s just a lot of hot air…”
Mailbag
Today’s first letter comes from Bob L., who disagrees with what I wrote in Friday’s Daily Journal. I said that if President Donald Trump follows through on his plans to enact stiff tariffs against our trading partners, we will experience, once again, the disastrous effects tariffs had on our country 100 years ago… Bob writes…
So I guess it’s okay to have other countries put tariffs on us but bad when we reciprocate? Ordinarily, I think you have good advice, but you’ve got this dead wrong! Let’s go back to stock and bond picking and avoid the pontifications! Regards, Bob L.
Porter’s comment: There’s no doubt that many countries have unfair trade policies with the U.S. But dealing with those countries isn’t what Trump is talking about.
Trump has said that on February 1, he will impose a 25% across-the-board tariff with our two largest trading partners, Canada and Mexico.
These countries do not have tariffs on our products because of NAFTA (The North American Free Trade Agreement) that President Bill Clinton first negotiated in the late 1990s.
These tariffs will not raise much revenue, but they will certainly screw up a lot of supply chains, resulting in more inflation.
Perfect example: we import a lot of cheap Canadian crude oil for refining into gasoline. I guess you want to pay 25% more for gasoline?
These tariffs will also lead to retaliation against American exports, which will hurt a lot of high-quality American businesses, like software and media.
And, as I was trying to point out, we’ve tried this before – in the 1920s. This approach leads to big declines in trade, growth in radical politics, and, eventually, war.
So maybe we should just reduce our government’s spending, and negotiate better trade agreements with the countries that are trading fairly… How’s that sound?
Warm regards,
Porter
The next letter comes from J.W., who writes…
Thank you for Saturday’s Daily Journal, “The Modified Munger Screen,” and 52-week highs and lows. I’m a small-business owner but manage a bit of money. This makes my investing much much better. It especially helps keep my eye on property & casualty (P&C) insurance companies. My subscription to Porter & Co. is well worth the cost. Thanks again and love this kind of work you do. J.W.
Porter’s comment: I’m glad you like it, J.W. And to those who you didn’t see it in your inbox (at 10 am ET on Saturday)… J.W. is talking about our latest product… Porter & Co.’s Saturday Stock Screen, where we pull back the curtains to show you some of the tools we use to find the best investment opportunities in the market. One of these is a series of stock screens to comb through the thousands of publicly listed companies, which we use to sift through a range of financial, accounting, and performance parameters to find the investment opportunities that are the diamonds in the coal mine.
We often use the results from these screens as a starting point for more in-depth analysis for possible inclusion in the The Big Secret on Wall Street portfolio. In Saturday Stock Screen, we used a screen that’s a modification of one inspired by Warren Buffett’s late partner Charlie Munger. And we also highlighted (for Parter Pass members) one opportunity that’s particularly compelling. To see the first Saturday Stock Screen, go here.
Let me know what you think by sending comments to [email protected]
Good investing,
Porter Stansberry
Stevenson, MD
P.S. When should you sell a stock that’s falling… and just how risky is your portfolio as a whole?
One of the biggest challenges facing investors – and one of the most frequently asked questions in our mailbag – is when to sell a stock. If we could see the future, we’d know the answer.
But barring Nostradamus and a crystal ball… there’s data. TradeStops is an emotion-free, data-driven way to determine when to sell a security. It crunches billions of bits of information, and gives you an emotions-free answer to the hardest part of investing.
Here at Porter & Co. we use TradeStops to monitor our portfolio performance… and we use it as an important input for when to sell (and most of the Porter & Co. team uses TradeStops to help guide their own investment decisions).
Whether or not you sell Nvidia because of DeepSeek… that’s up to you. But TradeStops can help. Learn more about it here.
