Every so often, a startup story stops feeling like a startup story and starts feeling like a weather system. That is where AI is right now. It is not just a category, not just a trend, and definitely not just a bunch of founders using the word “agent” like it is garlic powder. It is a full-blown climate pattern sweeping through venture capital, public markets, infrastructure spending, enterprise software, and probably your group chat.

That is why the latest 20VC+SaaStr discussion landed so hard. On one side, Databricks crossed the magical $100 billion valuation line and looked less like a private software company and more like a sovereign emotional-support asset for the AI era. On the other side, CoreWeave embodied the less glamorous truth of the boom: none of this works without debt, data centers, GPUs, contracts, and enough capital intensity to make an old-school telecom executive spill their coffee.

Put those two stories together and you get the real plot of the AI economy. Databricks is the polished promise. CoreWeave is the industrial plumbing. One sells the software layer enterprises want to buy. The other finances the compute layer that makes the whole dream possible. If you want to understand why everybody keeps shouting about an AI bubble, start there. The bubble is not just hype. It is hype stapled to very real demand, very real spending, and very real risk.

Why This 20VC+SaaStr Conversation Hit a Nerve

The brilliance of the episode was not that it picked winners and losers. It was that it treated AI like what it has become: a giant stack of interlocking bets. The Databricks question was simple on the surface: is a $100 billion valuation bold, absurd, or merely the cover charge for the next generation of enterprise software? The CoreWeave question was more visceral: if you borrow mountains of money to build the compute backbone of AI, are you a genius early to a platform shift, or the canary in the capital-markets coal mine?

Those are not separate debates. They are the same debate wearing different shoes. Databricks represents the thesis that enterprise AI will create huge, durable software winners with strong revenue quality, expanding product surfaces, and enough cash flow discipline to justify elite multiples. CoreWeave represents the thesis that AI demand is so explosive that even a debt-heavy business can win big if it secures scarce GPU supply and long-term customer commitments before everyone else.

And in between sits the rest of us, watching capital pour into every layer of the AI stack like someone left the funding faucet on overnight.

Databricks at $100 Billion: Expensive, Yes. Irrational, Not Necessarily.

Databricks hitting $100 billion sounded, to some people, like peak mania. That reaction is understandable. A 12-figure private valuation has a way of making even seasoned investors suddenly rediscover the phrase “discipline.” But the Databricks case is stronger than the headline suggests.

Why Investors Keep Paying Up

At the time the debate heated up, Databricks had moved beyond the “promising AI company” bucket and into the much rarer “large, fast-growing, enterprise-grade platform with actual traction” bucket. That is the bucket investors will fight over with sharpened spreadsheets. The company projected a roughly $4 billion annualized revenue run rate, said its AI products had already reached $1 billion, served around 15,000 customers, and had more than 650 customers spending over $1 million a year. That is not a science-fair demo. That is a real business with real gravity.

Even better for the bull case, Databricks was not pitching itself as a one-trick pony tied to a single AI feature or fashionable workflow. It was expanding the platform. Agent Bricks pushed deeper into enterprise AI agents. Lakebase signaled ambition around operational databases for AI-native applications. The broader strategy was clear: do not just help customers analyze data; become the place where data becomes production AI.

That matters because the best enterprise software companies do not win by selling one tool. They win by becoming harder to leave every quarter. Databricks is trying to become infrastructure in the psychological sense, not just the technical one. Once a company stores, governs, analyzes, and builds AI systems around your data, switching becomes less like changing apps and more like moving houses with your fridge still full.

Why the Skeptics Still Have a Point

Now for the cold shower. A great company can still be a too-expensive stock. Or in this case, a too-expensive private mark. The argument against Databricks is not that it lacks momentum. It is that the market may be pricing years of continued excellence as if they are already sitting safely in the bank.

That is where the bubble talk sneaks in. At these levels, investors are no longer just underwriting product quality. They are underwriting sustained hypergrowth, durable AI demand, healthy margins, category expansion, and a future IPO market willing to bless the math. That is a lot of assumptions stacked on top of each other like a Jenga tower built by optimists.

Still, Databricks is exactly the kind of company that can make a scary valuation look surprisingly normal in hindsight. If AI becomes the default interface for enterprise data, governance, analytics, and applications, then the company may not look overpriced at all. It may look early. The problem is that “early” and “too expensive” often wear the same jacket for a while.

CoreWeave’s $11 Billion Debt Gamble: The AI Boom With Hard Hats On

If Databricks is the elegant software story, CoreWeave is the story that smells faintly of metal, transformers, cooling systems, and lender term sheets. It is the reminder that AI does not run on vibes. It runs on chips, racks, power, and financing structures large enough to make ordinary startup funding look like couch-cushion change.

Why CoreWeave Became the Perfect Bubble Test

CoreWeave’s rise has been astonishing. The company’s revenue jumped from $16 million in 2022 to $229 million in 2023 and then to $1.9 billion in 2024. That kind of growth is not normal. It is the kind of chart that makes investors grin, critics squint, and historians start muttering about parallels.

But the growth came with real financial weight. A major 2024 debt raise brought in $7.5 billion. By the end of 2024, CoreWeave’s filing showed roughly $7.9 billion in net debt plus about $2.7 billion in operating and finance lease liabilities. Add those together and you can see why the 20VC framing of an roughly $11 billion debt bet resonated. This was not a casual balance-sheet flex. It was a giant wager that AI demand would remain hot enough, long enough, to outrun the cost of building the rails.

And then there is customer concentration, the two-word phrase that can ruin any investor’s lunch. CoreWeave disclosed that its top two customers accounted for roughly 77% of revenue in 2024, with Microsoft alone representing 62%. That is not ideal diversification. That is “please do not let one procurement meeting go sideways” diversification.

Why Bulls Love the Story Anyway

The bull case is simple: scarcity creates pricing power, and speed matters more than elegance in a land grab. If frontier-model companies and hyperscalers need enormous GPU capacity now, not three years from now, then the provider that can deliver compute at scale becomes incredibly valuable. In that world, debt is not recklessness; it is acceleration.

There is evidence that this logic is not fantasy. Even after the market worried about CoreWeave’s capital intensity, the company kept landing large strategic relationships. In 2026, Meta expanded its relationship with CoreWeave through a fresh multiyear AI cloud deal worth about $21 billion, adding to an earlier agreement. That does not erase risk, but it does show why lenders and equity investors were willing to believe the demand curve might justify the buildout.

In other words, CoreWeave is what happens when Wall Street meets GPU hunger and neither side wants to blink first.

Why Bears Keep Reaching for the Fire Extinguisher

The bearish case is just as intuitive. Debt works beautifully until timing stops cooperating. A business with huge financing needs, concentrated customers, expensive infrastructure, and a fast-moving hardware cycle does not have much room for narrative drift. If a major customer pulls back, if pricing weakens, if chip transitions get awkward, or if capital markets become less generous, the model can go from “brilliantly leveraged” to “alarmingly exposed” in a hurry.

That is why CoreWeave became such a useful symbol in the AI bubble conversation. It turns the abstract question of “is spending excessive?” into something concrete: what happens when the entire ecosystem assumes future demand will be so large that today’s debt looks conservative?

Why We’re All Living in the AI Bubble

Now to the spicy part. Are we in an AI bubble? Yes, but not in the lazy, dismissive way people often mean it. We are living in an environment where capital, expectations, and narratives are all running ahead at a heroic pace. That is bubble behavior. But it is happening on top of genuine adoption and real economic utility. That is what makes this era both exciting and confusing.

The Money Is Ridiculous

Stanford HAI reported that U.S. private AI investment reached $109.1 billion in 2024, while global private investment in generative AI hit $33.9 billion. Then the NVCA and PitchBook data for 2025 showed AI and machine learning deals capturing 65.6% of all U.S. venture deal value. That is not normal sector enthusiasm. That is capital behaving like it has decided every other category can wait in the car.

This concentration changes behavior everywhere. Founders start optimizing for AI adjacency. Investors start treating “not AI” like a mild character flaw. Public markets begin pricing future infrastructure demand before the revenue stack underneath it has fully settled. Bubble conditions do not require the technology to be fake. They only require belief, money, and extrapolation to start feeding each other in a loop.

The Adoption Is Real, Too

Here is the inconvenient detail for the bubble skeptics: companies are not spending all this money solely because everyone lost their minds. McKinsey found that 92% of companies plan to increase AI investment over the next three years. Stanford also found that 78% of organizations reported using AI in 2024, up sharply from the year before. So yes, there is frenzy. There is also broad-based implementation.

The gap between investment and maturity is the real tell. McKinsey’s research found that only 1% of leaders describe their organizations as truly mature in AI deployment. That means companies are pouring money into the category while still figuring out how to translate pilots into repeatable operational value. This is exactly how major platform shifts look in the messy middle. Spending rushes in first. Efficiency shows up later, if at all.

The Weird Circularity Problem

Goldman Sachs put its finger on one of the strangest features of this market: the growing circularity of the AI ecosystem. Capital moves through model companies, cloud providers, infrastructure vendors, and enterprise customers in ways that can make demand look larger, cleaner, and faster than it really is. Revenue at one layer can depend heavily on funded spending at the layer below or above it.

That does not mean the revenue is fake. It means the ecosystem can start validating itself in a circle. Investors fund model builders. Model builders buy compute. Compute providers finance more capacity. Software platforms build new AI layers. Enterprises invest because they fear being left behind. Everybody points to everybody else as proof. It is a little like a conga line made of term sheets.

So Is This a Bubble or a Buildout?

The honest answer is both. Calling AI “a bubble” as if that ends the conversation is too shallow. Calling it “the future” as if valuation no longer matters is equally lazy. The more useful framework is this: we are in a buildout phase being financed with bubble energy.

That is why Databricks and CoreWeave are such perfect companion stories. Databricks shows why investors believe enormous enterprise value will accrue to the software winners of the AI stack. CoreWeave shows what it costs, literally, to make that software future physically possible. One story is about revenue quality and product expansion. The other is about leverage, infrastructure scarcity, and execution under pressure.

If AI adoption keeps compounding, both companies can look smart in retrospect. If spending slows or value accrues more slowly than expected, the same market that celebrated them can turn on a dime and act like it never liked them in the first place. Financial markets are funny that way. They will throw you a parade on Monday and ask for margin discipline by Thursday.

What Living Inside the AI Bubble Actually Feels Like

Here is the part that numbers alone do not capture: living through this moment feels strangely normal and completely absurd at the same time. That is probably the clearest sign we are in a bubble-shaped era. On Tuesday, someone calmly tells you a company needs tens of billions of dollars of compute. On Wednesday, another person says a software platform tied to enterprise data might be “cheap” at $100 billion. On Thursday, everyone goes back to scheduling meetings and pretending this is just ordinary business.

For founders, the experience is disorienting. If you have an AI angle, the market wants to meet you yesterday. If you do not, people squint at your deck like you forgot to put pants on. Teams feel pressure to reposition, repackage, or outright reinvent themselves around agents, copilots, automation, and whatever the phrase of the week happens to be. Sometimes that leads to genuine product evolution. Sometimes it leads to old software wearing a shiny AI mustache.

For operators inside companies, the experience is more complicated. There is real excitement because the tools are improving fast and some of them genuinely save time, reduce friction, and unlock work that used to feel tedious. At the same time, there is constant anxiety. Leaders worry they are moving too slowly. Employees worry the strategy changes every month. Buyers worry they are overpaying for software that still needs a human babysitter. Nobody wants to miss the shift, but nobody wants to be the fool who bought the decorative shovel in the middle of a gold rush.

For investors, the mood is equal parts fear and greed, which is a very efficient way to make a market weird. The fear is missing the next generational winner. The greed is obvious. That combination creates a strange emotional logic where paying a high price can feel safer than waiting, because the scariest thing in a breakout cycle is not overpaying. It is being right too late.

And for the rest of the tech world, the AI bubble feels like living next to a construction site that promises to become a palace. The noise is constant. The spending is enormous. The cranes never seem to stop moving. Some of what gets built will absolutely matter for the next decade. Some of it will be torn down, refinanced, written off, or quietly rebranded as enterprise workflow optimization. History tends to keep both outcomes.

That is why the phrase “we’re all living in the AI bubble” rings true. Even if you are not trading the stocks, funding the rounds, or borrowing against GPU contracts, the bubble changes behavior. It changes expectations. It changes hiring plans, product roadmaps, capex decisions, and dinner-table conversations about which companies are geniuses and which ones are setting money on fire with premium branding.

In the end, the lived experience of an AI bubble is not simply mania. It is acceleration. It is the unsettling feeling that the future is arriving too fast for clean accounting, neat categories, or emotionally reasonable valuation models. Some of that future will absolutely stick. Some of it will look hilariously overpriced in hindsight. The annoying part is that both things can be true at once.

Conclusion

The latest 20VC+SaaStr debate worked because it captured the AI economy in one snapshot: breathtaking upside, aggressive pricing, giant infrastructure risk, and a market trying to decide whether it is funding the next industrial revolution or just the world’s most expensive group project. Databricks at $100 billion makes sense if enterprise AI becomes deeply embedded in how companies operate. CoreWeave’s debt gamble makes sense if demand for compute stays ravenous and contractual. The AI bubble is real, but so is the underlying transformation.

That is the trickiest part of all. Bubbles are often built around technologies that matter. Railroads mattered. The internet mattered. Mobile mattered. AI clearly matters, too. The only question is how much of today’s price, optimism, and financing structure survives the part where the future has to start producing boring, repeatable returns. Until then, expect more giant rounds, more giant headlines, and more people saying “this is totally normal” while standing ankle-deep in rocket fuel.