By Pam Martens and Russ Martens: January 29, 2025 ~
We’ll get to Google’s AI Search model shortly, but first some necessary background.
Like many of you, we spent a good part of our day yesterday reading up on DeepSeek, a Chinese startup that purports to have built an AI model that rivals U.S. industry leaders like OpenAI and Google – for a mere fraction of the cost.
On Monday, the news that DeepSeek’s AI model might have rendered most of those sophisticated and expensive chips from Nvidia obsolete shaved $600 billion off the market value of Nvidia – the largest one-day dollar loss in a stock in U.S. market history.
Adding to the nervousness, DeepSeek has made its AI model open-source, meaning its base code is publicly available for tech geeks in the U.S. and around the world to examine and build on. Also bringing out the worry beads in Silicon Valley, DeepSeek has been around for less than two years and is the brainchild of 39-year old Liang Wenfeng, a computer wizard who started a quant hedge fund at age 25 which had garnered a $39 billion portfolio 11 years later, according to Reuters.
By yesterday, a lot of inquiring minds in the media were asking if U.S. investors had been snookered in a fashion similar to the dot.com bubble; if all those giant data centers built or in progress across the U.S. landscape were a big waste of money; and if Big Tech investors were about to witness a rapid repricing of their multi-trillion-dollar tech stocks.
On June 20th of last year, we wrote the following:
On Tuesday, a stock most Americans had never heard of four years ago – Nvidia – closed with a market cap of $3.34 trillion. That makes it the most valuable company in the world, overtaking Microsoft’s heady $3.32 trillion market cap.
Nvidia’s share price (ticker NVDA) has soared 174 percent year-to-date while the S&P 500 is up just 15 percent. The much broader index, the Russell 2000, has flat-lined this year. Without the gains from Nvidia, the S&P 500 would be reporting one-third less percentage increase year-to-date.
Nvidia trades on the Nasdaq stock market. Its share price has been riding the artificial intelligence (AI) hype in a manner reminiscent of how the Nasdaq skyrocketed in value on the tech and dot.com mania of the late 1990s.
That era did not end well, to put it mildly. The Nasdaq reached a closing high of 5,048.62 on March 10, 2000. The Nasdaq then proceeded to lose 78 percent of its value over the next 2-1/2 years, reaching a closing low of 1,114.11 on October 9, 2002.
As late as February 2000, there was little recognition in mainstream media that the Nasdaq was on the cusp of entering one of the bloodiest selloffs in stock market history. CNNMoney reported as follows on February 29, 2000:
“U.S. stocks rallied broadly Tuesday, sending every major market gauge higher and the Nasdaq composite index to its 12th record close of the year as investors snapped up technology shares expected to lead the economy’s growth.”
The same news report quoted Legg Mason’s Chief Market Strategist at the time, Richard Cripps, as follows: “People want to own these (technology) stocks, and that’s what limits any significant drop on these stocks and it’s what puts pressure on the remainder of the market.”
Less than two weeks later, investors began the stampede out of the market darlings.
In April of last year, we had written about the dodgy history of Nasdaq and cautioned about similarities in the market’s misallocation of capital today. We said this:
“Today we have airplane components falling off commercial passenger planes in the sky and unsafe bridges, while a Donald Trump startup, Trump Media & Technology Group, (owner of a social media platform whose primary use seems to be for Trump to slander sitting judges and elected officials), has a market cap of $5.5 billion and trades at 1800 times revenues. According to an SEC filing on Monday, the company lost $58.19 million last year on revenues of a meager $4.13 million.”
Which brings us to the much-hyped Google AI Search Engine model, Gemini.
Yesterday, we decided to find out what Google’s AI actually knows about the critical core of the U.S. financial system: that is, the interaction of the Federal Reserve with the handful of megabanks on Wall Street.
First, we typed the following into the Google Search box: “how much did the federal reserve provide to banks in cumulative loans as a result of the 2007-2010 financial crisis according to the GAO?”
We specifically asked for GAO data because that is the Government Accountability Office, the government audit arm that works for Congress. In July 2011, the GAO released the definitive report and chart showing that the Fed’s cumulative emergency loans to bail out the hubris of the megabanks on Wall Street came to a tally of $16.1 trillion and lasted from December 2007 to at least July 2010. (See the chart on page 131 of the GAO report here.)
Google’s Gemini gave us a decidedly wrong answer. Per the screenshot below, it missed the correct answer by $13.6 trillion. It also referred to TARP, which was a U.S. Treasury program not a Federal Reserve program.
Next, we asked Google’s Gemini “what banks are the largest owners of the federal reserve bank of new york.” The first bank listed was the Bank of Millbrook, which has 4 branches and $298 million in assets according to the FDIC. The first bank listed should have been JPMorgan Chase, which has 5,195 branches and $3.6 trillion in assets. The Google AI model was, for unknown reasons, incapable of quickly going to our 2019 authoritative article which was headlined These Are the Banks that Own the New York Fed and Its Money Button.
We asked numerous other related questions about the Fed and the megabanks and received similarly wrong answers from the Google Search AI model.
