I’ve been doing this for a decade now, and I want to open with something that’s going to sound strange coming from someone writing an AI bubble article: I actually don’t think “bubble or not bubble” is the right binary question. It’s the question everyone asks because it fits neatly into a headline, but the real answer is messier, more interesting, and honestly more useful for your portfolio than a simple yes or no. So let’s go through this properly, module by module, the way I’d actually want a research note handed to me before I sized a position.
The Number That Should Make You Sit Up Straight
Let’s start with the scale, because I don’t think people have fully internalized how large this has gotten. Google, Amazon, Microsoft, and Meta are collectively planning to spend roughly $725 billion on capital expenditures in 2026 alone, up a genuinely staggering 77% from last year’s already record-breaking $410 billion. Break that down by company: Amazon is leading the pack at around $200 billion, Alphabet is guiding to $175-185 billion (recently nudged up toward $190 billion), Microsoft is tracking near $190 billion, and Meta is sitting in the $115-135 billion range. Analysts are already penciling in combined spend north of $1 trillion for 2027.
To put that in perspective the way I do when I’m trying to shock a dinner party into paying attention: that’s more than the annual GDP of most mid-sized countries, being spent by four companies, in a single calendar year, mostly on GPUs and the buildings to house them. Microsoft’s CFO Amy Hood attributed $25 billion of the company’s capex increase specifically to rising memory chip prices, which tells you the input costs themselves are inflating even before you get to the buildout scale. Alphabet’s cloud contract backlog reached roughly $460 billion, nearly double what it was a year prior, which is one of the genuinely bullish data points in this whole mess, because backlog isn’t hopeful storytelling, it’s signed paper.
The Bull Case, Stated Fairly
I want to steelman this properly before I tear into the scary parts, because I’ve noticed a lot of bubble commentary skips straight to the fear porn without giving the bulls their due. Jefferies analyst Brent Thill put it bluntly: the AI economy is healthy, and the bear thesis is garbage, his words, not mine, though I’ll admit I chuckled when I read it. His reasoning isn’t crazy. Google Cloud revenue jumped 63% year over year to $20 billion in a single quarter. AWS added $8.3 billion year over year to hit $37.6 billion. Azure added $7.9 billion to reach $34.7 billion. This isn’t spending into a void, revenue is genuinely accelerating alongside the capex, which is exactly the pattern you’d want to see if this were a real infrastructure buildout rather than a speculative frenzy.
Nvidia’s own data center revenue tells a similar story, clocking in at $75.2 billion in a single quarter, up 92% year over year. When the picks-and-shovels layer is showing that kind of growth backed by actual GPU shipments rather than vaporware, it’s hard to dismiss the whole thing as pure fantasy. And here’s a comparison that genuinely reframes how I think about this: Bernstein analysts once called Big Tech “the new Big Oil,” noting that Big Oil’s combined capex peaked around $166 billion back in 2013. Big Tech blew past that years ago and hasn’t looked back. Massive capital cycles aren’t new, they’re how railroads, telecom, and electrification got built too, and the companies that survived those cycles came out the other side with genuine, durable moats.
The Bear Case, and This Is Where It Gets Genuinely Uncomfortable
Now for the part that keeps me up at night more than the headline capex number itself: circular financing. Here’s the mechanism in plain English, because the jargon obscures how simple and slightly absurd it actually is. A chip maker or cloud provider invests billions into an AI startup. The startup then turns around and spends that exact money buying chips or cloud capacity from the same investor. The investor books it as revenue. Everyone claps. Nobody outside the loop has actually paid for anything yet.
The poster child here is the Nvidia-OpenAI relationship, and the timeline alone should make you raise an eyebrow. It started in September 2025 with a letter of intent for Nvidia to commit up to $100 billion to OpenAI’s infrastructure buildout. By February 2026, Jensen Huang had walked that figure back, calling it “never a commitment,” and what actually materialized instead was a $30 billion equity stake tied to a $122 billion funding round valuing OpenAI at $852 billion. Then came a newly reported $250 billion guarantee tied to a 10-gigawatt Ohio data center, structured as debt support rather than equity, meaning Nvidia has essentially agreed to backstop its own customer’s ability to keep buying its own chips. UBS estimates the OpenAI-Nvidia arrangement alone could represent up to 13% of Nvidia’s projected 2026 revenue. Sit with that for a second. Thirteen percent of one company’s revenue, tied to a single circular relationship with a customer that’s projected to lose roughly $14 billion in 2026, nearly triple its 2025 loss, while chasing a stated goal of $100 billion in revenue by 2029.
Michael Burry, the short-seller who became a household name for calling the 2008 housing collapse, responded to the Nvidia-OpenAI news with a one-liner that’s stuck with me: history is repeating itself. Bernstein’s Stacy Rasgon flagged the circular concern all the way back in September 2025, noting Nvidia had already participated in more than 50 venture deals for AI companies in a single year, several of which turned around and used that capital to buy Nvidia’s own hardware. The dot-com parallel that keeps getting invoked isn’t lazy commentary, it’s structurally similar to the telecom vendor-financing collapse of the late 1990s, when companies like Nortel financed their own customers’ purchases, booked the resulting sales as organic growth, and then watched the entire arrangement implode when real-world usage never caught up to the inflated revenue figures.
Not Everyone Agrees This Is 1999, and Their Argument Has Teeth
I don’t want to be the guy who cherry-picks only the scary data, because that’s intellectually lazy and it’s also just bad analysis. There’s a genuinely thoughtful counterargument from Acadian’s research team worth sitting with: they’d only find circular financing alarming if it involved selling equity to outside shareholders, hidden leverage, or distorted fundamentals, and for most of the recently announced deals, including the AMD-OpenAI arrangement, they’re not seeing those specific red flags. Their framework, which I’ve started stealing for my own analysis, calls net equity issuance one of the “Four Horsemen of the Bubble Apocalypse,” alongside a wave of AI IPOs, acquisition binges paid for with inflated stock, and megacap companies halting buybacks. All four were present in 1999-2000. Right now, by their read, none of the four are showing up in force.
That’s a genuinely fair point, and I think it’s the crux of where reasonable people actually disagree. The counter-counter-argument, though, and this is where economist Noah Smith’s writing has stuck with me, is that Nvidia’s trade receivables have been growing faster than its revenue, which is a subtler warning sign than headline equity issuance. Vendor financing through receivables doesn’t show up as flashy IPO mania, it shows up quietly on a balance sheet until the day it doesn’t.
The Depreciation Problem Nobody Wants to Talk About at Parties
Here’s a module most bubble commentary skips entirely, and it’s the one I find most quietly important. When you spend $150 billion or more on assets that depreciate over roughly five to six years, you’re booking somewhere in the neighborhood of $17-20 billion in annual depreciation per company before you’ve earned a dollar back on that specific tranche of spending. That’s not speculation, that’s accounting arithmetic. Free cash flow has compressed accordingly. Meta’s FCF has tightened as capex has outrun operating cash generation. Microsoft’s capex-to-operating-cash-flow ratio has climbed past 50%. The market has tolerated this so far because cloud revenue growth has kept accelerating, and because the perceived alternative, under-investing and ceding the AI infrastructure race to a competitor, looks strategically worse to every CEO in the room. But tolerance is conditional, not permanent. The moment AI cloud growth decelerates while capex keeps climbing at the current pace, multiple compression across the entire sector could get genuinely brutal, and it would happen fast, because markets don’t gently reprice hundred-billion-dollar assumptions, they reprice them all at once.
My Actual Read, After Ten Years of Watching Capital Cycles Play Out
Here’s my honestly unpopular take, the one that usually gets me a mixed bag of replies whenever I say it out loud: I don’t think this is a clean binary between “genuine productive buildout” and “1999 rerun.” I think it’s both, simultaneously, in the same market, at the same time, which is a genuinely uncomfortable place to invest from. The hyperscaler-to-hyperscaler spending, backed by real cloud revenue growth like Google Cloud’s 63% jump and Azure’s continued acceleration, looks structurally sound to me. The Nvidia-OpenAI circular financing loop, layered with a company projecting a $14 billion loss while receiving hundred-billion-dollar guarantees from its primary chip supplier, looks like the part of this cycle that could genuinely unwind ugly if enterprise AI adoption disappoints even modestly.
What I’m actually doing with my own money reflects that split view. I’m comfortable holding the picks-and-shovels layer, the Nvidias and the hyperscalers themselves, because their revenue is diversified across real paying customers beyond just the circular deals, and their balance sheets, receivables growth aside, aren’t yet showing the equity-issuance red flags Acadian’s framework flags as the actual bubble tell. I’m deliberately more cautious about anything whose valuation depends heavily on a single circular financing arrangement holding together, because that’s the layer where a single stalled negotiation, and we’ve already seen headlines about exactly that kind of stall between Nvidia, OpenAI, and Oracle, can trigger panic across three mega-cap names simultaneously.
The Uncomfortable Truth I’ll Leave You With
Big capital cycles have built genuinely durable industries before, railroads, telecom, electrification, and they’ve also produced spectacular, portfolio-destroying wreckage when the underlying demand didn’t materialize on the timeline the spending assumed. The honest answer to “is this a bubble” isn’t yes or no, it’s “watch the receivables, watch whether equity issuance starts creeping in, and watch whether cloud revenue growth keeps decelerating or reaccelerating.” Those are the tells that matter far more than any single headline number, no matter how many zeros it has. I’ve been burned before betting on a clean narrative in a messy market, and this is about as messy and consequential a capital cycle as I’ve watched in ten years of doing this. Position accordingly, and don’t let anyone, bull or bear, sell you the version of this story that fits too neatly into one sentence.
