Is AI a Bubble? The Trillion-Dollar Question for Builders

Amazon, Google, Microsoft, and Meta are collectively spending $725 billion on AI infrastructure in 2026. As a percentage of US GDP, that exceeds the Apollo space program, the interstate highway system, and every major American capital project except the Louisiana Purchase. OpenAI is filing for an IPO at over $1 trillion while projecting $14 billion in losses this year. Anthropic just closed a $65 billion funding round at a $965 billion valuation. Global AI spending is projected to surpass $2.5 trillion in 2026, up 44% year-over-year.

Meanwhile, a National Bureau of Economic Research study found that 90% of firms reported no measurable impact of AI on workplace productivity. Chinese hedge funds are warning that the "super bubble" is ready to burst. The South Korean KOSPI halted trading in June to prevent a crash after AI-related stocks triggered a sharp sell-off. The Shiller price-to-earnings ratio has exceeded 40, a level reached only once before in history, immediately preceding the dot-com crash.

So: is AI a bubble?

The honest answer is that it can be both transformative and overvalued at the same time. The internet was genuinely transformative. The dot-com bubble was genuinely a bubble. Both things were true simultaneously. The internet did not become less real when Pets.com went to zero. It became more real, because the surviving companies built durable businesses on the infrastructure the bubble paid for.

That framing matters more than the binary question, especially if you are someone building a product right now. Whether AI is a bubble or not changes what tools you should bet on, how you should think about AI dependencies in your product, and whether the platforms you are building on will exist in five years.

The Numbers That Worry People

The scale of investment in AI infrastructure is historically unprecedented. The core data points driving the bubble narrative are hard to dismiss:

The gap between spending and revenue is the core vulnerability. Tech companies are collectively investing hundreds of billions of dollars in infrastructure to serve an enterprise AI market that generates a fraction of that in actual revenue. That gap either closes as adoption accelerates, or it becomes the defining characteristic of the crash when investor patience runs out.

The Numbers That Reassure People

The bull case is not fantasy. Today's AI companies differ from dot-com era startups in important ways:

Nvidia reported $215.9 billion in revenue for fiscal 2026 with confirmed purchase orders totaling $1 trillion through 2027. These are real revenues and real orders, not projections. Anthropic's revenue exploded from roughly $1 billion in early 2025 to a $47 billion annualized run rate by May 2026, with approximately 85% coming from enterprise customers. The company expects its first profitable quarter in Q2 2026. Unlike the dot-com era, where companies with zero revenue commanded multi-billion dollar valuations, today's AI leaders are generating substantial real income.

The counterargument is that revenue alone does not prove sustainability. Anthropic plans to spend approximately $19 billion on compute in 2026, roughly matching its full-year revenue. Gross margins reportedly compressed to around 40% after inference costs ran over projections. The revenue is real, but the profits are not yet proven at the scale the valuations demand.

What a Bubble Bursting Would Actually Look Like

If the AI bubble pops, it will not mean AI disappears. The internet did not disappear after the dot-com crash. What happened was a violent repricing: companies without sustainable economics failed, infrastructure spending contracted, and the surviving companies — Google, Amazon, eBay — built enduring businesses on cheaper infrastructure that the bubble had paid for.

An AI correction would likely follow the same pattern. The infrastructure gets built regardless. The chips are already ordered. The data centers are already under construction. What changes is who profits from them. Companies burning cash faster than they generate revenue — including potentially one or both of the current AI leaders — face existential risk. Companies with sustainable unit economics and real customer retention survive and inherit the infrastructure.

For builders, the practical implication is about dependency risk. If your product is built on top of a platform that might not survive a correction, you have a problem. If your product is built on a platform with sustainable economics that does not depend on continued VC subsidies, you are in a stronger position.

What This Means for People Building Products

This is where the bubble question stops being abstract and starts being practical. If you are building a product right now, the AI bubble debate affects three specific decisions:

1. Platform Dependency Risk

AI app builders like Lovable and Bolt depend on AI model providers (primarily Anthropic and OpenAI) for their core functionality. If those providers raise prices dramatically, cut API access, or — in a worst case — face financial distress during a correction, every product built on top of them is affected. This is not hypothetical. Companies are already switching providers to cut costs: Lindy, an AI startup, recently moved 100% of its traffic from Anthropic's Claude to DeepSeek to survive rising token costs.

Visual no-code platforms like Bubble.io and FlutterFlow have a different risk profile. Their core functionality does not depend on AI model providers. They run on their own infrastructure with their own economics. AI features are additive, not foundational. If AI model costs spike or providers consolidate, Bubble and FlutterFlow continue to function exactly as they did before AI entered the conversation.

2. The Cost Trajectory Question

The AI bull case assumes model costs will continue falling, making AI-powered tools cheaper over time. That has been true historically, but the recent data is less encouraging. Leading model developers have been slower to slash prices in recent months, and enterprise customers are reporting that falling per-token costs are offset by dramatically higher usage volumes as AI workflows expand. The net effect for many companies is that AI costs are rising, not falling.

Tools with predictable, non-AI-dependent cost structures become more attractive in that environment. Bubble charges for workload units on its own servers. FlutterFlow charges a flat subscription. Neither platform's pricing is tied to the token economics of frontier AI models. If AI costs go up, your Bubble app costs the same.

3. The Maintenance Reality

As one developer in the Bubble community put it: "Everyone's impressed by how fast AI can build version 1. I'm more interested in who still understands version 127."

AI-generated code is fast to produce and cheap to create when model costs are low. But maintaining, debugging, and extending AI-generated code over months and years requires either ongoing AI interaction (at whatever the prevailing token costs are) or human engineering expertise. If AI costs rise during a correction, the ongoing maintenance cost of AI-built software rises with them.

Visual no-code platforms produce applications that are maintained in the same visual environment where they were built. The maintenance cost is your time, not token fees. The person who built the app at version 1 can still understand and modify it at version 127 because every workflow, every data relationship, and every piece of logic is visible and inspectable.

The Counterintuitive Thesis: A Burst Strengthens No-Code

Here is the part that most AI bubble analysis misses: if the AI investment bubble deflates, it does not mean the demand for building software without engineers goes away. That demand is structural. There are millions of founders, operators, and businesses that need custom software and cannot afford or access engineering teams. That need existed before ChatGPT and will exist after any correction.

What changes is which tools serve that demand. During the current AI boom, AI app builders have captured attention because they are fast, novel, and funded by the same investment surge driving the broader bubble. If that funding contracts, if AI model costs stop falling or start rising, if some AI tool providers do not survive the correction, the builders who depended on those tools will need alternatives.

The alternatives that are already built, already profitable, already independent of AI model economics — platforms like Bubble.io, FlutterFlow, and Webflow — become more valuable in that scenario, not less. They are the "boring" tools that just work, regardless of what happens in the AI investment cycle. Their value proposition does not depend on trillion-dollar valuations or continued exponential growth in model capabilities. It depends on letting people build software visually, which works today and will work tomorrow whether or not the AI bubble pops.

This does not mean AI tools are bad bets. Lovable, Replit, and Cursor are genuinely useful products that may have perfectly sustainable economics. It means that when choosing how to build your product, the AI bubble question is a real factor in your risk assessment. Building on a platform whose existence depends on continued AI investment momentum is a different risk profile than building on a platform whose existence depends on people wanting to build apps visually.

How to Build With the Bubble Question in Mind

Diversify your dependency. If you are building with an AI tool, make sure you can export your code and migrate if needed. Lovable, Replit, and Cursor all produce exportable code. If your AI platform disappears or becomes unaffordable, you have something to take with you.

Evaluate platform economics, not just features. Does the platform make money? Can it survive without continued VC funding? Bubble.io has been operating since 2012 and charges customers directly for a profitable service. A platform that launched last year on VC money and charges below cost to acquire users is a different bet.

Consider maintenance costs over the full product lifecycle. The cheapest tool to build with is not necessarily the cheapest tool to maintain for three years. Factor in what it costs to keep your product running and evolving, not just what it costs to launch.

Use AI for prototyping, use established tools for production. A common pattern emerging among experienced builders: use Lovable or Replit to validate an idea quickly, then rebuild on Bubble or FlutterFlow for production. You get the speed of AI for exploration and the stability of established platforms for operation.