Ed Zitron was recently on Bloomberg discussing the AI bubble. The interview starts off strong with the interviewer asking “What are the AI bulls getting wrong?”. Zitron says, “People are conflating a semiconductor rally with an underlying successful business, which doesn’t really exist.”

Let’s first align on terminology. I have no idea what people mean when they say “AI bubble”. Does AI mean semiconductors, CPUs, RAM, neoclouds, energy providers, photonics companies, hyperscalers, networking companies, private companies like Anthropic, etc?

In fairness to Zitron, he primarily talks about Anthropic, OpenAI, and hyperscalers. He references Apple as a good example of a company that isn’t spending on AI. The general sentiment of the interview is that “AI”, whatever that means, is a bubble. I couldn’t disagree more. If his point is more nuanced than this, and I missed it in the interview, please correct the record in the comments.

In short, his point are the following:

• the ROI on AI can’t be measured and it might be negative
  • Uber's COO says it's getting harder to justify the money spent on AI tokenmaxxing
• Frontier labs are burning cash
• Some AI infrastructure companies are overlevered
  • “There will be a lot of losers in this because we don’t know if GPU compute is margin positive” (minute 4:50)
    • Coreweave is burning billions in capex and their customers are Nvidia, OpenAI, Microsoft

Where we agree

• ROI on AI is hard to measure
• Anthropic and OpenAI are losing money and they don’t have a clear path to profitability
• There will be winners and losers
  • Couldn’t agree more. Some companies are over-leveraged.
    • “As of December 2025, CoreWeave's gross EBITDA leverage, excluding leases, was 7.0x with lease-adjusted gross leverage at 7.7x.” (Fitch)
  • Some companies have the right ideas but are simply too early
    • Poet, POET -2.33%↓ has almost a $3 bln market cap and AI data centers are still relying on pluggables, active electrical cables, and conventional optics while the industry has not yet standardized around deeper photonic integration like optical interposers, co-packaged optics, or optical I/O.

Where our opinions diverge

An enormous amount of money is still going to be made by investors

AI is still going to transform the way we live, and some companies are going to make incredible sums of money. If you’re a passive investor and you want to stop the analysis at the 30,000 feet view, then that’s okay, and I’d agree with everything Zitron said.

Let’s dig into these points in greater detail. I’ll start each section with a quote from Zitron and then we’ll work through it.

ROI on AI

“Uber's COO Andrew McDonald said that they're having trouble justifying the AI spend based on the actual return . So you've got a thing where you can't measure the costs and you can't measure the return on investment. What do you call that? You call it a thing without an ROI”. (minute 0:40)

“We don’t know if GPU is margin positive, which is crazy considering the amount of money going into it.” (minute 4:49).

One of Zitron’s main points is that ROI can’t be measured. Uber is being used as a main talking point by the AI Bears. This point is more nuanced than the headlines lead you to believe.

1. First, it’s not like Uber’s engineers are going back to handwriting code. The days of handwriting code are done. Gone forever.
2. Second, tokens cost money. For companies that wanted to let Claude Code run wild, let engineers tokenmaxx, who cares about managing context windows, etc., it was very predictable that they’d burn through insane amounts of money. Any engineer can tell you that AI needs very specific prompts, it needs very clear direction from a good engineer. There’s a reason software companies like Salesforce still exist. You can’t just sit down for a few days, prompt your favorite bot and duplicate their platform. Hell, you probably can’t even build a good TODO application in one day.

Before any company was going to tokenmaxx, executive should have sat down in a working session with a few engineers and built a toy application. They could have saved tens of millions by having a one day working session. And if they were working diligently, they probably could have realized by lunchtime that their toy application wasn’t going to be built by 5pm.

This is a analogous to the Laffer Curve in action. We know two things to be true: 1) extreme AI use (i.e. tokenmaxxing) has negative ROI and 2) we’re never going back to handwriting code.

Now we have our boundary limits. We’re using AI, but we’re also not going to be extreme with our use. This is starting to sound more reasonable. So AI usage will be greater than 0 but less than infinite. Is this really shocking to anyone?

Examples of positive ROI from AI

• Amazon uses robots that sort, lift, and carry packages
• Lemonade to cut insurance rates for Tesla drivers in endorsement of EV maker’s software technology
• OpenAI has solved one of the Erdos problems
• Waymo's Rider-Only fleet achieved a statistically significant lower crashed vehicle rate across all combined categories compared to human benchmarks

In a nutshell, as Warren Buffett famously quipped, “We would rather be approximately right than precisely wrong.” (2010 letter to shareholders).

Winners and losers

Interviewer question, “at some point there will be some winners, some losers, and we'll see somethings filter their way out. There will be some ROI for larger players when it comes to AI. Do you not see some winners in this?”. Ed Zitron says, “I think the winner is Jensen Huang and the winner is the construction firms who have got prepaid for all the data center construction, maybe Sam Altman and Dario Amodei, and and they have become billionaires through this. But, when it comes to the actual businesses you can’t find anyone who can measure the ROI because you can’t do it. So when it comes down to the model companies themselves, they’re horrifyingly, horrifyingly unprofitable. I think OpenAI lost like 20 something billion dollars last year. Anthropic is probably not far behind that.” (minute 1:40)

For active investors, this point is way too high level. I completely agree, Jensen Huang and Nvidia are winners. Altman and Amodei are billionaires. That’s indisputable. But, why does it matter? It’s irrelevant. None of those things affect me as an investor.

There are clear winners and opportunities to invest in the AI boom. Can you blindly throw money at the technology and expect it to be positive? No.

Here are some winners for investors.

• Self driving cars
  • “Waymo began serving over 1 million fully autonomous rides every month; a number that we’re on a path to hit every week by the end of 2026. In total, we served over 14 million trips so far in 2025 alone – more than tripling our public rides from last year” (source)
  • Their stock, GOOG 0.09%↑ is up 200% in the past 5 years and 110% in the past year
• Healthcare
  • Guardant Health’s stock up after FDA approves its blood test for colon-cancer screening
    • Guardant is providing critical insights into what drives disease through its advanced blood and tissue tests using real-world data and AI analytics
    • Their stock, GH -2.41%↓ is up 165% in the last year
• Energy efficiency
  • Meta
    • Meta will deploy standalone Nvidia Grace CPUs in production, with Vera to follow — company sees perf-per-watt improvements of up to 2X in some CPU workloads
  • Vertiv
    • Liquid cooling leverages the higher thermal transfer properties of water or other fluids to support efficient and cost-effective cooling of high-density racks and can be up to 3000 times more effective than using air
      • VRT -0.27%↓ is up 193% in the past year

This list goes on forever. I could name countless other AI plays, but I want to avoid any names that could fall in the “AI is a bubble” category. Those use cases are real-world examples where companies used AI to deliver a meaningful benefit to you as an individual and an investor via better healthcare, self-driving cars, and energy efficiency.

Anthropic going public

“When we see these S1s, I think it's going to be kind of a massacre, because I think that people have this view that these companies are becoming more profitable or even have a path to profitability, and they don't have one.”

“OpenAI and Anthropic should not be allowed to go public” (minute 6:40).

For anyone following the industry, none of this is news. The CEOs themselves have said the path to profitability is going to be challenging.

For example, Dario has said this is an incredibly challenging environment. He said they can easily get it wrong and lose money. Watch his interview with Dwarkesh Patel.

“With the way we buy these data centers, if you’re off by a couple years that can be ruinous. “There’s a real dilemma here…When you go to buy data centers, the curve I’m looking at is okay we had a 10x increase every year. The beginning of this year, we had $10 billion of annualized revenue. We have to decide how much compute to buy. It takes a year or two to build out the data center. In 2027 how much compute do I get? I could assume the revenue will continue growing 10x. So revenue will be$100 billion at the end of 2026, and $1 trillion at the end of 2027. It would actually be buying$5 trillion of compute because it would be $1 trillion a year for five years.

If my revenue is not $1 trillion, if it’s$800 billion, there’s no force on earth, there’s no hedge on earth that could stop me from going bankrupt.

A part of my brain wonders if it’s going to keep growing by 10x. I can’t buy $1 trillion of compute. If I’m off by one year or if the growth rate is 5x instead of 10x then you go bankrupt. So you end up in a world where you’re supporting hundreds of billions but not trillions [of compute]. And there’s a risk that you can’t serve all customers, and there’s a risk that you still overestimated growth. Have we been thoughtful about it or are we YOLO’ing it?” (minute 50:44)” - Dario Amodei

Who has said this is going to be a slam dunk and it’s easy? The CEO of Anthropic clearly stated that this could go wrong and any company can go bankrupt. He should be given credit for transparency and saying things this openly ahead of an IPO. Find another CEO who speaks this plainly.

No one. Absolutely no one has said these companies have a guaranteed path to predictability. Including Altman and Amodei. All they’re guilty of is being optimistic and trying to make it work. Do we want to criticize optimism and transparency?

Takeaways

There are plenty of companies today that are managing their balance sheets properly and poised for tremendous growth going forward. They’re not easy to find. Investing is hard. In today’s markets you basically have to fall in one of three camps.

1. I’m a passive investor; I don’t research the markets a. Congrats - you’re an AI bull. The Mag 7 are approximately 37% of your S&P 500 index fund.
2. I’m bearish on AI and actively not materially investing in those companies a. Fair enough. It’s always possible you’re right. My question is how long will you sit out until you capitulate? What if Nvidia becomes a $10 trillion company? What if self-driving cars completely take over your neighborhood? At what point do you join the party?
3. I’m bullish on AI and recognize that markets are frothy a. This is my camp. Let’s look at how we manage investing and the downside if we’re wrong. i. How do we know if this wrong:
   1. Smarter models: Models could get exponentially smarter and cheaper (e.g. smarter tokens) and therefore we could run them locally to solve the majority of personal tasks, rendering future demand for memory, compute, etc. significantly lower than expected.
   2. Supply vs demand: If supply significantly outstrips demand then prices will drop and company’s profits will not offset increased valuations ii. How do we know if we’re right:
   3. Growth: OpenAI and Anthropic are currently growing at 10x, and I expect 3x or better growth for the foreseeable future.
   4. Centralized models: Maybe we use local models for simple voice queries like “what’s the weather today” or “make a dinner reservation at Dorsia”. But, real work like engineering, scientific research, etc will continue to be dominated by massive centralized models that require API calls
   5. Agentic: agentic continues to explode in demand and use cases
   6. Video: video is consuming the internet, “more than 200 billion reels are viewed across Instagram and Facebook” (source). Once models get good enough to generate and edit videos, demand will explode.

A final takeaway for the passive investors.

A great aspect of investing is that not making a decision is the same thing as making a decision. I hope you enjoy buying the SpaceX IPO when the indexes change the rules and fast track it into your 401k. The indexes modifying the rules to win the IPO by jamming it into your 401k is high-finance beauty.

I need to do a separate post on financial beauty. I touch on it in Fannie Mae: a 10x opportunity? by summarizing Ackman’s covid trade where he warned ‘hell is coming’ because of virus and then he pocketed $2B in bets against markets.