The End of Elon? American Atlas Review: Is Marc Chaikin’s Power Gauge Report Worth It?

Marc Chaikin has a habit of building his presentations around a real piece of technology that most retail investors have never heard of, and then explaining it well enough that you walk away smarter whether or not you buy anything. His August pitch did that with Anthropic’s frontier models. His new one, titled “The End of Elon?”, does it with something even more obscure: the difference between 16-bit and 64-bit floating point math, and why the Department of Energy just paid for a supercomputer at Oak Ridge National Laboratory that runs on the latter.

The headline is provocative on purpose. Elon Musk’s Colossus cluster in Memphis is the largest AI training facility on Earth, and Chaikin’s argument is that a machine “less than 1% the size” and drawing “less than 1% of the electricity” will beat it at the thing that actually matters, which is scientific discovery. He calls the coming network of these machines “American Atlas,” puts a $248 trillion number on the disruption, and names his first pick for free: Advanced Micro Devices (AMD).

We read all 93,000 characters of the transcript. Here is what we found.

The Presenter

Marc Chaikin entered the markets in 1965 and became a licensed broker in 1966. He spent the following decades as a Wall Street quantitative analyst and trader, and the presentation is candid about who his clients were: “His former clients have included George Soros, Michael Steinhardt, Paul Tudor Jones, and Steve Cohen.” That is not a list you can fake. Chaikin’s institutional work was real and well known in the industry long before he started selling research to individuals.

His most durable contribution is the Chaikin Money Flow indicator, which measures institutional buying and selling pressure by weighting volume against where a stock closes within its daily range. The promo describes it without naming it: “Marc invented one of Wall Street’s favorite stock indicators – found in every Bloomberg and Reuters terminal on the planet today.” That is accurate. Chaikin Money Flow and the Chaikin Oscillator ship as standard studies on Bloomberg, Reuters, and virtually every retail charting platform. Nasdaq later hired him to build three indices for the exchange, and he rang the opening bell.

The Power Gauge is his second act. Launched in 2011 through Chaikin Analytics, it rates stocks Bullish, Neutral, or Bearish on 20 factors spanning financials, earnings, technicals, and expert opinion. The presentation cites a backtest showing it “would’ve accurately flashed bullish on at least 7 out of the top 10 stocks of the year since 2016,” and points to individual wins including Micron before a 970% run, Celestica before a 6,600% run, Nvidia in 2014, Sterling Infrastructure, Super Micro, Vistra, and Alpha Metallurgical Resources. These are maximum peak-to-trough gains cherry-picked from a large universe, and the promo says so: “These are long term, maximum gains from stocks flagged by Marc’s system. Past performance is no guarantee of future results.”

On the macro side, host Molly Hendrickson recounts a run of market calls: the COVID crash warning in January 2020, the bottom call 24 hours before the rebound, the 2022 bear warning, the early 2023 rebound call, the early 2025 Liberation Day warning, and an early 2026 correction warning that hit “two days” later. We have covered Chaikin’s track record in detail in our Frontier AI review and the MAGI review, and the pattern holds: he has been directionally right on the big turns more often than most, and he is upfront that his system is probabilistic rather than prophetic.

The Big Idea

The pitch has three coined pieces of vocabulary, and each one maps to something real.

The Floating Point Problem. Chaikin’s core argument is that today’s generative AI runs on low-precision hardware. “FP” stands for floating point, and the number after it is how many bits the chip uses to store each value. “FP16 chips calculate stuff extremely fast,” he says, “Which is why they power almost every major AI model out there. But they can’t execute precise calculations.” He goes further on Musk’s cluster: “Colossus runs mostly on FP8 or even FP4.”

He demonstrates the point by asking ChatGPT how many C’s are in the word broccoli. It gets it wrong. He calls this “the Broccoli Problem” and argues that it “isn’t a software problem. It’s a hardware problem.” His cleaner illustration is arithmetic: start with 1 and add 0.00001 five times. “After five steps, the FP64 computer would give you the right answer: 1.00005.” An FP16 machine “would just spit out your original input: the number one,” because it cannot represent the small increment.

That second example is genuinely correct. FP16 has about three decimal digits of precision, and 1.00001 rounds to exactly 1.0. The broccoli demonstration is a tokenization quirk rather than a precision failure, but the underlying claim that scientific simulation requires FP64 while chatbots do not is standard knowledge in high-performance computing.

Lux and the AI micro cluster. The solution, Chaikin says, is “a completely new kind of AI data center” being built at Oak Ridge. “The official name for this new breakthrough device is Lux. It’ll be the world’s first FP64-powered AI supercomputer.” Its designers call it “the first dedicated U.S. AI factory for science.” He estimates it will draw “as little as 24 megawatts” versus Colossus’s “about 2 gigawatts,” with a footprint “about the area of a 7-Eleven convenience store.” He labels this category “the world’s first AI micro cluster.”

American Atlas. Lux is “just the first node.” Under what the promo calls Executive Order #14363, the DOE is building nine new AI supercomputers (Lux, Discovery, Minerva, Janus, Tara, and others) and converting its existing fleet of “more than 50” FP64 machines (Frontier, El Capitan, Aurora, Perlmutter, Polaris, Venado, Tuolumne, Crossroads) into one networked system across 17 national labs. “It’s so new, it doesn’t even have an official name yet, Molly. So I’m calling this AI supercluster American Atlas.” Its footprint, he says, “will be bigger than the state of Texas.”

The payoff is what he calls “AI super agents”: agentic AI engineers, doctors, and physicists running on FP64 hardware, working in DOE “autonomous labs” on 26 initial targets across fusion, cancer, quantum computing, and grid optimization. We unpack the coined terms further in What Is the American Atlas Portfolio? and The AI White Swan of 2027.

The Key Claims

The promo stacks its numbers carefully, so it is worth laying them out.

The $248 trillion. This is the sum of three third-party figures: McKinsey’s $23 trillion AI market estimate, Bloomberg’s $40 trillion nuclear fusion opportunity, and a $185 trillion figure from Unleash Prosperity on the value of eliminating cancer. Chaikin frames it as “about eight times the size of the whole U.S. economy… And more than 50 times the size of the current AI market.” He then contrasts it with McKinsey’s $4.4 trillion generative AI estimate to get “56 times bigger than the first boom.” The arithmetic is honest; what it measures is total economic value across four industries over decades, not a stock market opportunity.

The power multiples. “FP16 chips can process 65,000 unique values at once. FP64 chips can handle 18 quintillion calculations at once. That makes them nearly 283 trillion times more powerful than FP16.” Those numbers are the representable value ranges of the two formats (2^16 and 2^64), and the ratio is correct as arithmetic. It is not a measure of useful compute; a chip does not become 283 trillion times more capable by widening its registers. The “more than 1 trillion times more powerful than Colossus” line has the same origin. Treat these as illustration, not benchmark.

The acceleration. “360X acceleration in AI breakthroughs… from 5 years… down to 5 days.” Later stated as “36,000%” and “breakthroughs that were going to happen in 2036 would happen this year.” This is attributed to “project leaders” and is the least verifiable claim in the deck.

The disruption. “It will create a complete collapse of the legacy AI industry. What’s coming next will tear the AI industry down to the studs.” And: “if you have a dime invested in Tesla… Or Nvidia… Or Microsoft… or Alphabet… or Meta… You need to understand what’s coming next.”

Chaikin does qualify. “Of course, these are extraordinary examples,” he says, “I can’t promise we will see gains this high.” And near the close: “if you’re already positioned in today’s top AI plays, that’s a good thing. What I’m showing you today is a different layer of the same opportunity. I want you to have both.” That last line quietly walks back the “collapse” framing, and it is the more defensible position.

The energy data. This section is the strongest in the presentation. Data centers “will double their electricity use between now and 2030… to around 945 terawatts” (IEA). “By 2028, data centers in America will consume an estimated 150 gigawatts.” Electric bills up “an average of 267% in areas with high data-center concentrations, according to a Bloomberg study.” Colossus at completion “will cover 1,156 acres,” which he converts to “about 1,049 football fields.” We go deeper on this in The AI Data Center Energy Problem.

The Free Ticker

Chaikin gives the pick away outright: “It’s the first company on my American Atlas ‘buy’ list, Molly. I’m talking about Advanced Micro Devices — ticker symbol AMD.”

His rationale is specific. Lux “starts with this next-generation AI chip… This beast is called the MI430X Instinct Accelerator,” which AMD describes as “engineered specifically for sovereign AI and scientific computing.” He adds that “The company just released its brand-new MI455X chips” and that “At least $1 billion dollars are about to hit this company’s account as a result.” The stock is “rated bullish in my Power Gauge right now.”

Here is what we can confirm. In October 2025, the DOE, Oak Ridge, AMD, HPE, and Oracle announced two new systems: Lux, an AMD Instinct MI355X-based AI cluster slated for early 2026, and Discovery, the Frontier successor built on AMD’s next-generation MI430X. AMD publicly characterized the partnership as roughly a $1 billion program. The promo blends the two machines somewhat, attributing MI430X to Lux, but AMD silicon is at the center of both, so the investment thesis survives the detail.

The FP64 angle is also fair to AMD. Its Instinct line has led Nvidia in double-precision throughput for several generations, which is exactly why Frontier and El Capitan run on AMD rather than Nvidia. If the DOE is going to network its national lab fleet, a large share of that fleet is already AMD.

The obvious caveat is that AMD is a $300 billion-plus company whose share price is driven overwhelmingly by hyperscaler demand for its MI300 and MI400 GPUs, not by a $1 billion government contract. Owning AMD is a bet on the whole AI accelerator market with a national lab bonus attached. That is a perfectly reasonable bet. It is not a hidden gem. Chaikin’s Power Gauge has favored AMD before; it showed up as the large-cap pick in the MAGI presentation as well.

The teased second pick is more interesting. “The biggest challenge will be connecting all of these AI supercomputers together into one massive device, seamlessly. That’s where our first American Atlas recommendation comes in.” It is a company with “cutting-edge networking solutions… The unsexy stuff,” and “The market hasn’t priced in the firm’s massive potential.” We take a run at the candidates in our AI supercluster stocks article.

What You Get

The offer is Chaikin Analytics’ standard Power Gauge Report package, which we have reviewed twice before.

  • Power Gauge Report, the monthly newsletter, “where I publish all of my newest research and recommendations each month.”
  • The Power Gauge system, described as “a $1,000 value on its own,” with “24/7 free access… for one full year.” You can type in any ticker and get the 20-factor Bullish/Neutral/Bearish rating.
  • Bonus report: The American Atlas Portfolio: Top Stocks to Profit from the Dawn of AI Superclusters. Chaikin says his team “pinpointed 17 companies that have signed deals” and that “some of them are absolutely tiny compared to AMD.”
  • Bonus report: The Worst AI “Landmine Stocks” to Avoid Right Now, the Power Gauge’s bearish AI names.
  • Model portfolio, position updates, and the special report library.

List price is stated as “$499” per year. The promo offers “a steeply discounted rate” without naming it in the transcript; the actual figure is on the order page, and in prior Chaikin promos it has been $49 for the first year with renewal at list. Check the renewal terms before you buy.

The guarantee is a real 30-day money-back: “Take a full 30 days to enjoy your new benefits. And if you’re not 100% satisfied, simply let Marc’s member services team know… And you’ll receive a full refund on your subscription, instantly, right away.” That is a cash refund, not a credit, which is better than many publishers offer. For a fuller treatment of the service itself, see Is the Power Gauge Report Worth It?

Our Take

We liked this presentation more than we expected to from the headline.

What Chaikin gets right. The DOE Genesis Mission is real. Lux and Discovery at Oak Ridge are real and are built on AMD. The distinction between FP64 for science and FP8/FP4 for chatbots is real and important, and Chaikin explains it more clearly than most technology journalists do. The data center energy figures are drawn from IEA and Bloomberg and are broadly accurate. The roster of DOE supercomputers is correct down to the lab locations. A retail investor who watches this will come away understanding something about high-performance computing that most of Wall Street does not.

What is stretched. The “trillion times more powerful” and “283 trillion times” figures are arithmetic on register widths, not performance. The “360X faster breakthroughs” number is unsourced. The “collapse of the legacy AI industry” framing is at odds with reality: Lux is a 24-megawatt science machine, Colossus is a multi-gigawatt training machine, and they do different jobs. Chaikin himself concedes this at the end when he says he wants you to “have both.” And the September 29 date in the headline never appears in the body of the presentation.

On the pick. AMD is a sound company that we would not argue against owning. It is also not a stock that goes up because of a DOE contract. If you buy it, you are buying the AI accelerator duopoly, and that is a fine thing to buy, but calibrate your expectations to a mega-cap.

Who this is for. If you want a systematic stock-rating tool with a real methodology and 15 years of history, the Power Gauge at the promo price is one of the better values in the newsletter business, and the 30-day refund means you can test it for free. If you are buying because you think Elon Musk’s empire is about to collapse, you have misread the pitch, and so, to some extent, has the headline writer. Chaikin’s actual thesis is narrower and more sensible: the government is spending billions on precision compute, AMD is the vendor, and a handful of smaller networking and infrastructure companies will ride along. That thesis is worth $49 to explore.

Where to Learn More

NewsletterVetter is an independent publication. We receive compensation from some of the services we review through affiliate links. Nothing on this site is investment advice. Always do your own research.