The AI Wildfire, Unemployment, and Wealth

Hey Everyone,

Hope you’ve had a wonderful holiday season so far. It’s been a busy semester at HBS, but now that classes are wrapped up, I will be sharing more on the newsletter.

Today, some thoughts on why AI isn’t a bubble, but the implications could be just as dire.

First, my publisher is offering a holiday discount on The Experimentation Machine for the month of December. Get copies for yourself, your colleagues, or your team.

We have sold nearly 10,000 copies since the book came out in February. I was worried a book on AI would become outdated quickly, but it's more relevant than ever.

And if you’ve already read the book, please leave a review on Amazon. It helps so much!

Order directly from Damn Gravity to get the discount (link above)

Without further adieu:

The AI Wildfire, Unemployment, and Wealth

One of the most common questions asked today is, "Are we in an AI bubble?"

My short answer is unequivocally no. But I also think it's the wrong question.

Instead, the most urgent question is, "What happens when unemployment crosses 10%?" And the corollary: "What happens when wealth disparity increases dramatically?" Sitting here in the midst of Startupland, I am seeing the very early signs of a job destruction tsunami and a wealth creation tsunami. I don't think we are talking enough about the societal implications.

Many Dark Servers vs. No Dark GPUs

To help understand why, we need to abandon the tired "bubble burst" framework and adopt a better metaphor, as articulated by Dion Lim: the AI Wildfire. As Lim writes, any coming correction will clear out the underbrush. Bad, undifferentiated ideas led by tourist entrepreneurs will burn out. Talent (the absolutely most precious resource in the market, particularly in an era of AI-native 10x founders and 10x joiners) will flock to durable, companies and this reallocation will be useful and a boon to the ecosystem.

I was an entrepreneur at e-commerce pioneer Open Market during the Internet 1.0 era, joining the early team and helping lead it through its IPO in 1996. That period was indeed a real bubble. The "Burn of the 1990s" was a speculative incineration where easy capital flowed into infrastructure overbuild and largely profitless growth stories. Remember Webvan? eToys? Kozmo? Probably not. But these flimsy, early models destroyed billions.

We sold our Internet commerce infrastructure software to all the big players at the time -- AT&T, AOL, Time Warner, WSJ. After spending millions with us to set up these capabilities, no one showed up. There were hardly any users or transactions. The Internet 1.0 crash was marked by vast amounts of unused capacity—what's commonly referred to as "dark fiber" (after all the fiberoptic cable laid) but I remember it as being an era of many "dark servers" and "dark stores".

The fundamental difference between then and now? The AI companies today aren't built purely on hype; they are generating real revenue and margins. NVIDIA is generating $100 billion in annual operating income. OpenAI is already at $13 billion in revenue, projecting $100 billion by 2027, and 800 million weekly active users. The cloud providers are absolutely ripping, growing an astounding 20-30% per year on a massive base. As has been famously said by many, there is "no such thing as a dark GPU" today (referring to NVIDIA's chips that underly the age of AI). One of our AI infrastructure portfolio company execs was asked at our recent Flybridge annual meeting about where the supply-demand equilibrium sits. His quote: "We are expanding capacity 10x next year over this year. And then 5x the year after. And we are still way, way behind it fulfilling the available demand."

This unusual supply-demand mismatch in part because the AI capabilities are so compelling and in part because we are in a unique moment in distribution. How is it possible that adoption for the first two "killer AI apps" -- intelligent chatbots and automated software development -- has happened so fast? Because as soon as new models and capabilities come out, they are instantly available to:

• 8 billion smart phone users

• 2 billion cloud-connected laptops

• 1 billion cloud-connected automobiles

This kind of instant, global distribution for new capabilities is unprecedented. And pretty darn magical.

Au contraire - Hype and Valuations

I acknowledge the serious counterarguments, often centered on the current cost structures. Critics rightly point out the risks posed by circular revenue schemes and ballooning AI-related debt issuance among hyperscalers. Concerns exist that high GPU utilization is a function of subsidized supply and unprofitable workloads, citing examples like Anthropic’s compute spending potentially outpacing revenue today and OpenAI's precarious operating model (as highlighted in a withering analysis by HSBC). And private company valuations? Fuhgeddaboudit. All of us in the VC community roll our eyes at least once a week, if not daily, when we see some of the flawed companies raising money at eye watering prices.

So, yes, these are real headwinds, and the market is showing early warning signs that the "Overgrown Forest" of easy capital is ending. Existing franchises are vulnerable to profit erosion thanks to these VC-subsidized upstarts. It's going to get messy out there.

But we are not mid-cycle; we are nowhere mid-cycle. We are still in the very, very early innings of innovation, growth, and transformation. Across the Flybridge portfolio companies and their customers, we are seeing early pilots, real ROI, and business process transformation across industries. But only in the last few quarters have companies begun rolling out their AI initiatives enterprise wide. It will take time, but the writing is on the wall. Software development is the first business process impacted and adoption has been insanely fast. Customer care is currently in the midst of transformation, with other knowledge work like legal, consulting, and banking following suit. And the result? Companies will employ millions of AI agents. And fewer humans.

Unemployment at 10%?

The magnificent seven have been the most AI forward companies in terms of internal process improvement and so represent a window into the future. They are each consistently growing 20-30% year over year with flat or down headcount. Amazon and Meta are great examples of the two ends of the spectrum on this point. Amazon's business is very operationally intensive and so they carry a huge headcount of low-paid workers. Meta's is very knowledge intensive and so they have a headcount full of high-paid workers. Here's what each of their trends look like over the last few years on headcount and revenue:

• Amazon 2021: 1.6M employees, $470B in revenue.

• Amazon 2025: 1.6M employees (flat), $670B in revenue (43% higher)

• Meta 2022: 86K employees, $117B in revenue

• Meta 2025: 78K employees (down 10%), $190B in revenue (62% higher)

AI is not the only story here, but it's a big one. These two titans are able to keep headcount relatively flat and repurpose people, and/or continue to hire new talent while dropping low performers, because they are each seeing a surge in demand for their services (as evidenced by achieving growth rates of 40-60% over 3-4 years). In other industries, that will not be the case. Instead, companies in industries that are not growing so rapidly that follow in the operational and adoption path of the AI leaders will cut headcount as a result of internal, AI-driven efficiencies. Dramatically.

We did an internal analysis at Flybridge that suggested that roughly half of all North American company payroll ($4.1T out of $8.6T) will eventually be automatable with AI. My best guess is that roughly 20% of that payroll will actually be reduced based on what I am seeing in terms of productivity improvements across the customers of our portfolio companies. The implication is that we will see incremental unemployment of 8-10%. On top of today's 4.6% unemployment (which has creeped up from 3.5% in the fall of 2019), that implies unemployment of roughly 13-15%. Humans are adaptable and so perhaps a surge in entrepreneurship will mitigate this dynamic. And I am probably overestimating the speed of adoption and impact, so let's call it approximately 10%. My prediction is that we hit that level of unemployment by the end of 2030 (i.e., five years from now). And, to be clear, this will be a structural unemployment level not cyclical. Meanwhile, GDP and the stock market are likely to continue to soar as this AI-driven automation drives efficiency. And that leads to exacerbating wealth disparity.

Wealth Creation, Wealth Concentration

Maor Shlomo is a remarkable solopreneur. He raised zero capital and built a product franchise, Base44, that resulted in a sale for $80 million. I am writing a case for my upcoming HBS class on his journey, detailing how he used dozens of AI agents to bootstrap his way to success. That pattern is going to repeat itself over and over again. "Tiny teams" is the mantra for many talented founders who are using the modern AI tools to become 10x founders -- leveraging AI agents for software development, go to market, customer success, and operations.

That means more entrepreneurs are going to generate massive success, and thus wealth, with smaller teams. And my above analysis of Amazon and Meta suggests that we will see a surge in market valuations as a result of growing revenue with flat headcount (i.e., growing profitability). More wealth for fewer entrepreneurs. More wealth for fewer executives. More wealth for shareholders.

The top 1% of the US population had a net worth of $20 trillion in 2007. Today, that figure is $60 trillion. That means $40 trillion in wealth has been created for the top 1% of the US population. In the next five years, with the increased productivity and potential of AI, could that number double or triple again to $80-120 trillion (a gain of $40-80 trillion)?

All of that leads me to the conclusion that we will see a dramatic rise in wealth concentration alongside the dramatic rise in unemployment.

I'd really like to see more conversation on this topic versus pointless talk about bubbles. Because if we don't get ahead of the unemployment cum wealth disparity dynamic as a country, we are going to be in a world of trouble.