Understanding Economic Bubbles

I had a wide-ranging interview on CNBC’s “Squawk on the Street” with Carl Quintanilla, Sara Eisen, and David Faber. The full interview clip is attached. We first discussed the strong 3Q earnings season results so far. Then the conversation pivoted to the question of whether AI is in a bubble. I noted that there are two markets for AI: Price and Capital, and both those markets exist in Public and Private form. In the public market, Price levels and Capital availability do NOT appear to be in a bubble. ◾ Earnings drive stock prices over time. Consider that the NVDA share price has climbed by 13x since 2022 and the EPS has increased by a similar amount. ◾ Absolute equity valuations are not as stretched as in the past. The 5 largest stocks in the S&P 500 currently trade at a P/E of 30x, below the 40x in 2021 and 50x in 1999 at the Tech bubble peak. ◾ From a capital availability perspective, there have been roughly 50 IPOs YTD, well below the 261 flotations in 2021 (excluding SPACs) and the 388 IPOs that were completed in 1999. In the private market, however, elements of extended frothiness exist for AI investments where external capital raising and vendor financing are essential to maintaining the growth of the business. ◾ George Soros, in his book “The Alchemy of Finance” (1987), explained his theory of “Reflexivity” in which the valuation of an asset is not simply the discounted present value of future cash flows. Instead, he argued that the change in stock price is itself an important component of the valuation of the shares. ◾ Put simply, a self-reinforcing process exists that inflates equity valuations and accelerates the underlying growth, that in turn expands valuation leading to a further increase in anticipated growth. However, this process cannot continue indefinitely and at some point the results fail to meet expectations, and prices adjust accordingly. ◾ In the case of AI, private companies are being valued at a multiple of revenues, not earnings, and in many cases growth is being financed by selling stock. Developed nearly 40 years ago, Soros used his theory of reflexivity to explain trading patterns in the equity market (chapter 1) and the currency market (chapter 2). Investors today can use reflexivity to understand the interconnected nature of price and capital availability in the private market for AI assets.

“AI is in a bubble.” Is it really? Here are 5 objective gauges - not vibes. Everyone suddenly agrees we’re in an AI bubble. But conventional wisdom is often more conventional than wisdom. So let’s look beyond the mood and ask: Can we measure a bubble? Carlota Perez showed that big tech revolutions often bring big bubbles: railroads, telco, dot-com… and now AI. But booms and bubbles look identical - until fundamentals catch up (or don’t). Azeem Azhar devised 5 measurable signals that separate boom from bubble: 1️⃣ Economic Strain Is AI investment big enough to shake the economy? • In H1 2025, AI accounted for 92% of US GDP growth. • As % of GDP, AI capex is 0.9%–4%. Railroads hit 4%, telco 1.2%. We’re in that range. Unlike railroads or fiber, GPUs depreciate in 3–5 years. They break even in ~2 years, if revenue keeps up. 2️⃣ Industry Strain Are companies making money? AI today: $400B capex → $60B revenue (6×). Railroads were 2×, telco 4×. But unlike them, genAI revenue was zero five years ago and is exploding. 3️⃣ Revenue Momentum How fast is revenue catching up? Railways: doubled every 3 years. Telco: every 4. GenAI: doubling in <12 months - and accelerating. Data centers are filled the minute they power on. Demand > supply. AGI complicates the picture: if you think you're building “the last technology,” you invest every dollar you can raise (see Sam Altman’s $1.4T ambition). That optimism can create bubbles if reality lags. 4️⃣ Valuation Heat Are prices detached? Dot-com Nasdaq P/E: 72 Today: 32 (elevated but not insane) But there’s froth: • Palantir P/E 360 • CoreWeave: infinite • Private markets: Thinking Labs at $50B with no product/revenue. Many AI darlings still need to prove their growth is sticky and defensible. 5️⃣ Funding Quality Where’s the money coming from? Global data-center capex next 3 years: $3T. Big Tech can cover ~50%. The rest? SPVs, off-balance-sheet structures, and heavy debt. This funding pattern triggered 9 of the last 18 bubbles. So… bubble or boom? There are bubble signals. But fundamentals are also growing at historic speed. Conclusion: We’re in a high-risk, high-velocity boom, not a confirmed bubble. Yet. More details: https://lnkd.in/e3gNYhtG Where do you stand? Boom, bubble, or something new entirely?👇

$560 billion invested. $35 billion in revenue. 𝐀𝐈 𝐡𝐚𝐬 𝐛𝐞𝐜𝐨𝐦𝐞 𝐛𝐨𝐭𝐡 𝐭𝐡𝐞 𝐞𝐧𝐠𝐢𝐧𝐞 𝐚𝐧𝐝 𝐭𝐡𝐞 𝐞𝐜𝐡𝐨 𝐨𝐟 𝐭𝐨𝐝𝐚𝐲’𝐬 𝐦𝐚𝐫𝐤𝐞𝐭 𝐨𝐩𝐭𝐢𝐦𝐢𝐬𝐦. Valuations are soaring, infrastructure is multiplying and investors are leaning into FOMO much like they did in 1999. The parallels are hard to ignore: - 𝐄𝐱𝐮𝐛𝐞𝐫𝐚𝐧𝐭 𝐜𝐚𝐩𝐢𝐭𝐚𝐥 𝐝𝐞𝐩𝐥𝐨𝐲𝐦𝐞𝐧𝐭: In 1999, it was fiber optics. Today, it’s data centers. Tech giants are pouring record sums (over $300 billion in AI infrastructure this year alone) on the assumption that demand will catch up. - 𝐒𝐩𝐞𝐜𝐮𝐥𝐚𝐭𝐢𝐯𝐞 𝐯𝐚𝐥𝐮𝐚𝐭𝐢𝐨𝐧 𝐥𝐨𝐠𝐢𝐜: Then, it was website traffic. Now, it’s token counts, model parameters and user prompts. Abstract metrics standing in for profitability. - 𝐂𝐨𝐧𝐜𝐞𝐧𝐭𝐫𝐚𝐭𝐢𝐨𝐧 𝐫𝐢𝐬𝐤: The “Magnificent Seven” mirror the dominance of the early internet titans. Market gains hinge on a handful of names, amplifying volatility when sentiment turns. - 𝐔𝐧𝐝𝐞𝐫𝐞𝐬𝐭𝐢𝐦𝐚𝐭𝐞𝐝 𝐥𝐚𝐠 𝐭𝐢𝐦𝐞: The internet changed the world, but only after a painful reset. The same pattern is emerging: transformative technology, premature expectations. History reminds us: 𝐛𝐮𝐛𝐛𝐥𝐞𝐬 𝐝𝐨𝐧'𝐭 𝐛𝐮𝐫𝐬𝐭 𝐨𝐮𝐭 𝐨𝐟 𝐭𝐞𝐜𝐡𝐧𝐨𝐥𝐨𝐠𝐢𝐜𝐚𝐥 𝐟𝐚𝐢𝐥𝐮𝐫𝐞; 𝐭𝐡𝐞𝐲 𝐛𝐮𝐫𝐬𝐭 𝐰𝐡𝐞𝐧 𝐜𝐚𝐩𝐢𝐭𝐚𝐥 𝐟𝐨𝐫𝐠𝐞𝐭𝐬 𝐝𝐢𝐬𝐜𝐢𝐩𝐥𝐢𝐧𝐞. For boards, CXOs and investors, this moment calls for a few fundamental questions: What are we really valuing: technology, or potential? Are we investing in productivity, or in perception? When the dust settles, will the value created be broadly shared, or concentrated even further? AI will reshape our economies. 𝐓𝐡𝐞 𝐪𝐮𝐞𝐬𝐭𝐢𝐨𝐧 𝐢𝐬 𝐰𝐡𝐞𝐭𝐡𝐞𝐫 𝐰𝐞’𝐥𝐥 𝐥𝐞𝐭 𝐞𝐱𝐮𝐛𝐞𝐫𝐚𝐧𝐜𝐞 𝐬𝐡𝐚𝐩𝐞 𝐢𝐭𝐬 𝐩𝐚𝐭𝐡, 𝐨𝐫 𝐠𝐨𝐯𝐞𝐫𝐧𝐚𝐧𝐜𝐞 𝐠𝐮𝐢𝐝𝐞 𝐢𝐭𝐬 𝐜𝐨𝐮𝐫𝐬𝐞.

Everyone seems to be saying the same thing right now: AI is a bubble. And maybe it is. But calling something “a bubble” is incomplete unless you are clear about what you mean: not just for AI, but for EVs, Bitcoin, Indian startup IPOs, clean energy, and every theme markets will argue about in the next decade. Because there are two distinct kinds of bubbles, and they behave very differently. Here is an idea that helped me reconcile the confusion. George Soros called it reflexivity: markets don’t just reflect reality, they can shape it. Belief moves capital, capital changes outcomes, outcomes reinforce belief. Sometimes that loop is harmless. Sometimes it rewires the real economy. Mean reversion bubbles are the familiar kind. Prices detach from fundamentals, but nothing structural changes underneath. The feedback loop mostly stays in price: rising prices validate the story, and the story attracts more money. When the story breaks, prices snap back to the mean (often violently). What went up without changing the world comes down without mercy. Inflection bubbles are harder and much more interesting. They form when something genuinely transformative is happening. But capital arrives too early, too fast, and too indiscriminately. Here reflexivity spills into fundamentals: belief attracts capital, capital builds infrastructure, infrastructure accelerates capability, and capability makes the original belief look “right.” Think railroads, electricity, telecom, the internet: each reshaped society, and each burned a lot of capital along the way. Mean-reversion bubbles are built on delusion. Inflection bubbles are built on impatience. And crucially, inflection bubbles don’t destroy capital evenly. They reward the investors underwriting uncertainty early, when outcomes are fragile and prices still leave room for error. They punish capital that funds “inevitability” later, at any price, especially when leverage enters the system. So when someone says “X is a bubble,” try to understand which kind. That’s the uncomfortable power of reflexivity. Markets don’t just speculate on the future, they sometimes pull it forward. Some bubbles are about fantasy while others finance the future. The mistake is treating both as the same kind of excess. PS: read his classic ‘The Alchemy of Finance’ to understand the theory of reflexivity better.

Remember the “.com gold rush”? Now imagine its sequel—only this time the asset bubble isn’t websites, it’s intelligence itself. 👀 In the late 1990s, the Dot-com Bubble saw valuations soar across the tech sector while fundamentals were largely ignored. Between 1995 and 2000, the NASDAQ Composite rose from under 1,000 to over 5,000 — then collapsed to ~1,139 by late 2002, a drop of nearly 77%. Fast‑forward: we’re now in the middle of what many are calling the AI Bubble. Investment levels are sky‑high, but the real question is: are we building value — or inflating expectations? As a coach, I’m compelled to ask: when the hype fades, what remains? For clients, boards or leaders, this question isn’t optional — it’s existential. 🕰 Review: During the dot‑com era, many firms went public with little revenue or profit — valuations driven by stories, not cash flows. 📈 Observe: In today’s AI scene, the hype is similar — massive capital flows, lofty visions, and warnings from experts that many initiatives may deliver zero ROI. 🌱 Focus: But unlike then, there’s also a human element—the “made‑by‑humans” lens. If we ground AI’s advance in human purpose and sustainable value, we shift from risk to resilience. Actionable Takeaways 1. Test your fundamentals: Just as dot‑com firms were judged by traffic, not value, today’s AI bets need to be considered by real business output. Ask: “If this fails, will my organisation, role or skill still matter?” 2. Amplify the human edge: Automation is tempting. But the differentiator isn’t “we have AI” — it’s how humans deploy it, interpret it, and sustain its impact. 3. Prepare for the correction: Markets can shift suddenly. The Bank of England and other regulators are already signalling that AI valuations are stretched.  4. Boardroom & leadership lens: For aspiring board directors or executives, your value lies less in tech tools and more in governance of tech‑impact: ethics, accountability, business continuity when the ‘bubble’ pops. 5. Build human‑centric infrastructure: Tools will come and go. However, culture, decision rights, and human judgment frameworks tend to survive downturns. In your sphere of influence — team, board, portfolio or career — ask: “What happens if the AI boom bends, stalls or corrects?” Let’s talk: What’s your Plan B for value when hype fades? “When the crowd chases novelty, the wise leader builds meaning; when the bubble bursts, significance remains.” Follow me, Sudhakar Reddy G. , for more insights on Leadership and Board Governance. Coach. Mirror. Certified Corporate Board Director. Book Free 1:1 Coaching Call: http://bit.ly/49qhIHb “Like a silent conch in the storm — true coaching calms, awakens, and guides from within.”

Is AI a bubble — or the catalyst that finally ends 50 years of stagnation? For half a century, breakthrough innovation has slowed. The Manhattan Project and Apollo program produced civilization-shifting leaps. Since the 1970s, most “innovation” has centered on entertainment and attention capture. Productivity data tells the story clearly: • Total factor productivity averaged 1.4–2.2% from 1950–2000, now ~0.9%. • The 2010s had the slowest labor-productivity growth ever: 1.1%. • Postwar GDP growth of 4% has fallen to ~2–3%. Now we’re living through the biggest capex wave in history. Cloud providers spent ~$230B in 2024 and guide to $300–350B in 2025 — almost all going to GPUs and the data centers that house them. With 4–6-year depreciation schedules, many analysts doubt AI services can cover their own cost. On paper, it looks like classic bubble behavior. But this narrow financial lens ignores the larger dynamic. If innovation is measured by rising productivity, this surge may finally break stagnation. The binding constraint is energy. Economic growth and per-capita energy use still track with a 0.8–0.9 correlation globally. Emerging economies show nearly 1:1. AI is extraordinarily energy-hungry. Data centers consumed 415–500 TWh in 2024 (1.5–2% of global electricity). The IEA expects this to roughly double by 2030, with AI workloads quadrupling. In the U.S., data-center consumption is projected to jump from 183 TWh to 426 TWh — driving half of all national load growth. This is already transforming energy planning. In 18 months, Microsoft has agreed to restart Three Mile Island Unit 1. Amazon and Google have signed SMR (small modular reactor) deals. Utilities are advancing natural gas, nuclear, and renewables at a scale and speed not seen in decades. Goldman Sachs estimates $720B of grid upgrades required by 2030. In effect, AI is functioning as the world’s strongest industrial policy for energy abundance. The compute being built today will matter — but the power capacity built to support it may be the real payload. History rhymes here. The railroad mania of the 1800s, the electrification boom of the early 1900s, and the dot-com overbuild of the 1990s all produced bubbles — yet each left behind physical infrastructure that powered the next century of growth. Some bubbles leave toys. Others leave railroads, fiber, and nuclear reactors. Foundational-model competition adds another twist: durable advantage remains elusive. New entrants with comparable capability appear every few months, often open-source and cheaper. Long-term winners will be those who solve concrete customer problems at scale. Today’s AI capex may look excessive. In ten years, it may be remembered as the moment the world finally built the energy + compute foundation for the next era of progress.

AI Euphoria and Historical Bubbles The chart from Bank of America highlights how periods of technological innovation have historically driven extreme market enthusiasm — often followed by steep corrections. • The Railway Mania of the 1840s saw returns of 135% before collapsing • The Telecom Bubble of the 1990s peaked at over 400% returns before the crash in the early 2000s • The Shale Revolution delivered gains of 206% but eventually faded • Today, AI euphoria stands apart, with returns already at nearly 600%, dwarfing past innovation-driven rallies This surge reflects both the transformative potential of artificial intelligence and the frothy market sentiment surrounding it. Like past episodes, the enthusiasm may well drive further upside in the near term, but history suggests that parabolic growth phases often end with sharp corrections once expectations outpace reality. For investors, this means two things: there are still opportunities to ride the momentum, but risk management is crucial, as the “AI trade” increasingly resembles prior bubbles in both scale and speed.

Can bubbles burst in a world of abundant Capital? Not a day passes without debates on the likelihood of AI being a financial bubble and when and how it might burst. In most cases, discussion draws on past experiences. However, in a world where everything real turns fake and fake turns real, is past any longer a guide to the future? First, bubbles are created at the intersection of real economies and financial capital. Liquidity, relationship between labor and capital as well as changing nature of capital underpins creation and burst of “good bubbles” that change the world Second, through most of human existence, capital has been very scarce. However, there were times when this scarce capital was channelled on a massive scale to fund major technological breakthroughs that are usually described as GPTs or general purpose technologies (e.g. canals, railways, electricity, internet) Third, rising investment tended to outstrip likely revenues, with excitement resembling a casino. Investors invested and borrowed beyond rational limits while valuations turned irrational. It then usually took a failure of just one company to raise capital, spreads widening or contagion from another sector for the house of cards to collapse. At that point, investors lost money and governments stepped in to reconcile benefits of technology with realities of capital markets But, this conventional description misses vital differences with today’s environment First, we live in a world of abundant not constrained capital. FSB estimates that value of financial instruments is at least 5-6x larger than underlying economies. Even these are conservative, as FSB measures derivatives on a net basis, have difficulty integrating private equity and debt or fully reflecting basis point and parity trades as well as leveraged treasuries positions or unfunded liabilities. The cloud of finance might be 10x+ GDP. Even during recent QTs, liquidity had to continue to grow, as any return to sound money is no longer feasible (I.e. that train left the station in 1990s) Second, unlike prior bubbles, AI is driven by intangibles, and these have very different qualities vs tangible capital (e,g. higher synergistic benefits and scalability, unprecedented spillover effects). A key side-effect of abundant capital and intangibles is a massive acceleration of both creation and destruction. Something that might have taken a decade, could be now accomplished in a year or two, enabling us to blow new bubbles almost daily Is AI a bubble? Yes. While it is hard to predict what will cause its collapse, in a world characterized by abundant capital, intangibles and compressed business cycles, any value drawdown should be softer than in the past and might not last as long. Do we pay a price for blowing bubbles? Absolutely, but it is found in politics, polarization, inequities, wars, climate and healthcare, not in asset prices, with gold acting as a defacto insurance policy in case things get out of control.

Financial bubbles don’t just come from over-optimism. They’re often fueled by feedback loops where one dollar of real demand gets recycled, reinterpreted, and amplified across multiple companies. Each echo looks like growth, valuations soar, and when the cycle breaks, the fall is brutal. We’ve seen this before. In the 1990s, Cisco was hailed as the backbone of the Internet. Startups bought routers, raised capital on the back of “Internet growth,” and service providers touted adoption as proof of demand. The same router sale was echoed three, four, even five times across financial statements and valuations. When demand faltered, Cisco lost nearly 80% of its value. Fast forward to today and NVIDIA’s GPUs are the routers of the AI era. 💡 Primary sale: NVIDIA sells GPUs and books billions in revenue. 💡 Secondary monetization: Startups and cloud firms lease those GPUs, booking their own revenue. 💡 Tertiary echo: Microsoft integrates OpenAI services and reports AI-driven cloud growth. Each layer capitalizes the same dollar of demand as if it were brand new. As a result, one $1,000 GPU can justify more than $150,000 in market capitalization spread across NVIDIA, a startup, and Microsoft. This system creates the illusion of exponential growth and ensures that when real demand slows, the collapse is sharper than fundamentals alone would dictate. History tells us bubbles inflated this way don’t deflate gently - they burst. Right now, markets still cheer the cycle. But it’s worth asking: Are these truly distinct pillars of demand, or just different echoes of the same dollar?

Bezos Just Called AI a “Good Bubble.” Here’s Why That’s a Big Deal. “Investors don’t usually give a team of six people a couple of billion dollars with no product, and that’s happening today.” He painted two types of bubbles:  • Financial bubbles (think 2008): they ravage market value and leave widespread damage.  • Industrial bubbles (like AI, says Bezos): inflated, yes but potentially productive. When the frenzy dies down, the breakthroughs last. He pointed out a strange pattern: firms with tiny teams and no product are now pulling in billions in funding: something that rarely happens in more rational markets. Why? Because in this phase, every idea gets backed, good and bad alike. The challenge becomes figuring out which ones will survive. Bezos argues that this kind of chaos isn’t just noise: it’s creative destruction. The hype will fade, the valuations will fall, but the infrastructure built (chips, data centers, algorithms) will remain. They form the backbone for what comes next. He compared it to past technological bubbles: → The dot-com crash annihilated many companies, yet we ended up with fiber-optic infrastructure, e-commerce, and new standards of global connectivity. → The biotech boom collapsed many firms but delivered life-saving drugs, diagnostic tools, and new biology platforms. → Today, AI is going through its boom. Billions flow into ventures that will never reach escape velocity. But the hardware, the software, the systems: those will stick. Bezos sums it up: this is a “productive kind of chaos.”  In moments like these, every wild experiment gets its chance. And from that chaos, real progress emerges. That’s how innovation looks before it becomes obvious.