The AI bubble narrative has been loud, confident, and everywhere since 2023. Dot-com comparisons. Valuations untethered from reality. CapEx bonfires with no return in sight. The skeptics had a coherent story — and for a while, the data didn’t clearly contradict them.
That window is closing fast.AI bubble revenue growth in 2026 is no longer a projection or a promise. It’s arriving in earnings calls, usage metrics, and enterprise contracts. The structural revenue that the bears said would never materialize is materializing. That doesn’t mean every AI valuation is justified. It means the wholesale “it’s all a bubble” take has become intellectually untenable — and the evidence is specific enough that we can now name exactly where the skeptics were wrong, and where they still have a point. This is that reassessment.
The Bubble Thesis, Summarized Fairly — Before We Dismantle It
Credit where it’s due: the bearish case was never stupid.The core argument ran like this — AI market valuation had decoupled from fundamentals. Companies were burning through capital at rates that assumed a total transformation of the global economy, on a timeline that no technology in history had ever achieved. The hype cycle looked uncomfortably familiar to anyone who lived through 1999. And the revenue, while growing, wasn’t growing fast enough to justify the multiples being assigned to it.
In 2023 and into 2024, those concerns had real merit. Enterprise procurement cycles were slow. Many companies were experimenting with AI rather than deploying it at scale. The killer use case outside of ChatGPT’s consumer novelty wasn’t obvious. And the infrastructure spend from Microsoft, Google, and Meta was reaching numbers that required extraordinary outcomes to pencil out.
The skeptics were stress-testing a thesis that deserved to be stress-tested. That’s not the problem. The problem is that some of them stopped updating.
The Anthropic Revenue Story Changes the Calculus
Anthropic is the clearest single data point that forces a reassessment.
The company’s revenue trajectory over the past 18 months has been steep by any standard. Anthropic’s annualized revenue run rate has grown from roughly $100 million in early 2023 to reportedly over $2 billion by late 2025, with acceleration continuing into 2026. The velocity matters as much as the number — this isn’t slow, linear growth. It’s compounding.
InsiderXP Fact: Anthropic’s annualized revenue run rate grew from approximately $100 million in early 2023 to over $2 billion by late 2025 — roughly a 20x increase in under three years, with growth continuing to accelerate into 2026.
1. What’s more telling is the revenue mix. Anthropic isn’t propped up by one enterprise mega-deal or a single consumer product. API revenue from developers, enterprise licensing through Amazon Bedrock and Google Cloud, direct enterprise contracts, and consumer Claude subscriptions are all contributing. That breadth signals genuine demand, not a concentrated bet.
2. Anthropic revenue at this scale and with this growth rate would have been dismissed as fantasy by most bubble skeptics three years ago. It’s not fantasy. It’s a quarterly earnings conversation.
3. Crucially, Anthropic isn’t an outlier. OpenAI crossed $3.4 billion in annualized revenue in 2024, with the company citing accelerating enterprise adoption as the primary driver. (Source: Reuters) The application layer is generating real money. The “where’s the revenue?” question has an answer now.
Claude Code Is the Product That Breaks the “No Real Use Case” Argument
If Anthropic’s revenue numbers update the macro picture, Claude Code growth updates the thesis at the product level.
- Claude Code — Anthropic’s agentic coding tool — has seen developer adoption accelerate sharply since its launch. (Source: Anthropic) Usage volumes are climbing.
- Enterprise engineering teams are embedding it into workflows, not just experimenting with it in sandboxes. The comparison to early GitHub Copilot adoption is apt, but Claude Code is hitting similar milestones faster and with broader task capability.
- Why does this matter for the bubble debate? Because agentic coding tools represent a qualitatively different revenue signal than a chatbot subscription.
When an engineering team builds Claude Code into their development pipeline, it becomes a workflow dependency. Churn risk drops. Usage scales with the team’s output. The revenue is recurring, embedded, and tied to something teams actively don’t want to lose. That’s not the usage pattern of a novelty product — that’s the usage pattern of developer infrastructure.
InsiderXP Fact: Agentic coding tools like Claude Code embed directly into engineering workflows, creating switching costs and recurring revenue tied to team output — a fundamentally different retention profile than consumer AI subscriptions.
Claude Code growth isn’t a marketing story. It’s a retention and expansion story. And those are the unit economics that justify long-term valuation.
The “no real use case” argument was always weakest in software development. It’s now essentially dead there.
AI Investment Returns Are Starting to Show Up in the Numbers
The CapEx critique has always been the most legitimate skeptical argument. Hundreds of billions flowing into GPU clusters and data centers, with returns that were largely theoretical & that’s changing.
1. Enterprise surveys from 2025 and early 2026 are documenting measurable productivity gains in software engineering, legal document review, customer support automation, and finance operations. These aren’t projected ROI figures from vendor case studies. They’re showing up in earnings call language from companies that have actually deployed at scale — reduced headcount growth against increased output, faster software release cycles, lower cost-per-resolution in support functions.
2. Goldman Sachs, McKinsey, and several independent enterprise research firms have all published data in the past 12 months showing that companies with mature AI deployments are seeing productivity improvements that are starting to justify the spend. Not universally, not evenly — but measurably.
3. The infrastructure investment returns timeline was always going to lag the CapEx. That’s true of every platform transition. The question was whether the returns would arrive before the capital markets lost patience. In the key application areas, they’re arriving.
The revenue curve is bending toward the CapEx curve. Not perfectly, not everywhere — but the direction of travel is no longer in serious doubt.
AI Market Valuation vs. Reality — Is the Market Still Mispriced?
Here’s where intellectual honesty requires some nuance.
Yes, parts of the AI market are still richly valued against current fundamentals. Infrastructure plays — companies whose entire thesis rests on continued exponential AI spend — carry significant speculative risk. If CapEx moderates, or if model efficiency improvements reduce the hardware requirements faster than expected, those AI market valuations compress hard. This dynamic was made visible in early 2025 when the emergence of DeepSeek’s low-cost model architecture triggered a sharp sell-off in AI infrastructure stocks. (Source: Financial Times)
But the application layer looks different now. Companies with documented revenue growth, expanding enterprise contracts, and high-retention product lines are trading on multiples that, while not cheap, are increasingly defensible against updated revenue projections. Applying a 2023-era “this is all speculative” lens to 2026 application-layer AI companies is a category error.
The honest AI market valuation assessment in 2026 is this: it’s not one market. Infrastructure is still a faith-based investment to a meaningful degree. Applications, increasingly, are not. Treating them as the same thing — as much bubble coverage still does — is lazy analysis.
What the Skeptics Got Wrong & What They’re Still Right About
The specific bubble predictions that haven’t materialized deserve naming.
The enterprise spending freeze didn’t happen. The ChatGPT plateau that was predicted in 2023 didn’t hold — AI assistant usage has continued to grow and diversify. The “no sticky use cases” argument has been directly falsified by developer tools, enterprise copilots, and vertical AI applications with documented retention metrics.
But the skeptics aren’t entirely wrong, and pretending otherwise would be its own form of hype.
Commoditization risk is real. As foundation model costs continue to fall and more players enter the market, margin compression is a genuine threat for companies without strong distribution moats. The concentration of value in a small number of players — Anthropic, OpenAI, Google DeepMind, Microsoft — raises questions about the rest of the ecosystem. And the assumption that current growth rates are permanent rather than front-loaded adoption curves is worth interrogating.
The bubble debate has evolved. The question in 2026 isn’t “will AI revenue arrive?” It has arrived. The question is who captures it at durable margins, which companies have built real moats versus riding a wave, and whether the infrastructure spend ultimately proves proportionate to the market it enables.
That’s a harder, more interesting question than “is this all a bubble?” — and it’s the question the serious analysts are now asking.
Frequently Asked Questions
1. Is there still an AI bubble in 2026?
The blanket “AI bubble” thesis has become increasingly difficult to defend in 2026. While parts of the market — particularly AI infrastructure and hardware plays — remain richly valued relative to current fundamentals, the application layer is now generating documented, recurring revenue at scale. Anthropic surpassed a $2 billion annualized run rate, OpenAI crossed $3.4 billion, and enterprise deployments are producing measurable productivity gains. The more accurate framing is that AI is not one market: some segments remain speculative, others are no longer speculative at all.
2. What is Anthropic’s current annual revenue run rate?
As of late 2025 and into 2026, Anthropic’s annualized revenue run rate has been reported at over $2 billion, up from approximately $100 million in early 2023. That represents roughly 20x growth in under three years. Revenue comes from multiple streams — API access for developers, enterprise licensing via Amazon Bedrock and Google Cloud, direct enterprise contracts, and consumer Claude subscriptions — making it structurally diversified rather than dependent on a single deal or product.
3. How fast is Claude Code growing compared to other AI coding tools?
Claude Code has seen accelerating developer adoption since launch, with enterprise engineering teams embedding it into core development workflows rather than treating it as an experimental add-on. It is reaching adoption milestones comparable to GitHub Copilot’s early growth but on a faster timeline and with broader agentic task capability. The key growth signal isn’t just user numbers — it’s the transition from trial usage to workflow dependency, which drives low churn and usage that scales with team output.
4. Which AI market segments are most overvalued right now?
AI infrastructure — companies primarily exposed to GPU demand, data center build-out, and continued exponential model training spend — carries the most residual speculative risk. If model efficiency improvements (as demonstrated by architectures like DeepSeek’s) reduce hardware requirements faster than expected, or if enterprise CapEx moderates, valuations in this segment compress significantly. By contrast, application-layer companies with strong revenue growth, enterprise contracts, and high product retention are increasingly trading on multiples that can be defended against real fundamentals.
The data has moved. The narrative needs to move with it.
If you think the bull case is being overstated here — or if you’ve seen enterprise AI deployments that tell a different story — the comments are open. And if you want this kind of evidence-led AI analysis without the hype or the doom, Surf through InsiderXP.
By the InsiderXP Editorial Team | May 07, 2026
