软件增速放缓有多少是预算流向Anthropic和OpenAI？

Look at that chart. Almost every major public software company has printed lower growth rates every single year since the 2021 peak. Not every other year. Not in patches. Every year, in sequence, without exception. Salesforce went from 25% to 8%. Snowflake from 106% to 24%. HubSpot from 47% to 17%.

The conventional explanation is that vibe coding is eating software. That’s nonsense — for the moment.  It just can’t show up in the dollars yet.

Developers are building their own tools in Replit and Claude, but the idea that the enterprise software stack is already getting hollowed out is just way, way too early to show up in revenue. It’s a clean narrative. It also doesn’t match the timing, the data, or the mechanics.

The vibe coding thesis is wrong. Not as a long-term risk — as a near-term cause.

The growth deceleration you see in that chart started in 2022. Claude Code launched publicly in May 2025. Cursor didn’t hit mainstream enterprise adoption until 2025. The SaaS slowdown predates vibe coding by three years. You can’t explain a 2022 problem with a 2025 tool.

Beyond the timing, there’s the practical reality. Nobody is vibe coding their way out of Salesforce. We’ve built 10+ production apps with AI-assisted development in the last few months — things that would have taken a team weeks, we did in days. But none of them replace an ERP, a CRM, or a compliance system. Shipping a v1 is maybe 2% of the work. The other 98% is data integrity, permissions, audit trails, integrations, change management, and the grind of keeping something running in production at scale. Enterprise software exists because that 98% is brutally hard. Vibe coding doesn’t change that.

What’s actually happening is simpler and more damaging: CIOs have finite budgets, and a new category just ate a massive chunk of them.

A lot of it. More than most people are admitting. And the numbers just got dramatically bigger.

The narrative around the software stock collapse of early 2026 has focused heavily on existential fear — AI agents replacing SaaS, Claude Cowork threatening to gut the enterprise software stack, the end of the seat-based model. All of that is real and worth taking seriously. But there’s a more prosaic, near-term force doing serious damage right now: CIOs have finite budgets, and they’re pouring an unprecedented share of those budgets into AI infrastructure and foundation model providers. That money has to come from somewhere.

The Numbers Are Unprecedented

Start with what happened to software stocks. The SaaS index fell 6.5% in 2025, while the S&P 500 rose 17.6%. That’s not a slight underperformance — that’s a 24-point gap while everyone around you was printing money. Then in early 2026, it accelerated. By early February, the IGV ETF was down nearly 20% year to date and close to 30% off its September peak. The sector’s forward P/E collapsed from roughly 35x at the end of 2025 to around 20x.

Now look at where the money went instead.

As of early March 2026, Anthropic crossed $19 billion in annualized revenue run rate — up from $9 billion at the end of 2025, and up from $14 billion just a few weeks before that. CEO Dario Amodei confirmed at the Morgan Stanley TMT conference that the company added $6 billion in run rate in February alone. Sacra’s data shows the customer base reflects this: 300,000+ business accounts, over 500 customers spending more than $1 million annually (up from a dozen two years ago), and 8 of the Fortune 10 now using Claude. One in 5 businesses on Ramp now pay for Anthropic, up from 1 in 25 a year ago.

Claude Code alone is now at $2.5 billion in annualized revenue — a product that didn’t exist publicly 10 months ago, growing at a pace that has more than doubled since the start of 2026.

OpenAI’s trajectory is parallel. Their CFO confirmed $20 billion ARR for 2025, and their March 2026 run rate is estimated at $25 billion. Epoch AI projects that at current growth trajectories, Anthropic could actually surpass OpenAI in annualized revenue by mid-2026 — a crossover point that would have seemed absurd 18 months ago.

Add it up: Anthropic and OpenAI together are now collecting somewhere between $40-50 billion in annualized enterprise spend — spend that barely existed two years ago. With roughly 80% of Anthropic’s revenue coming from enterprise and API customers, you’re looking at $35+ billion hitting corporate IT budgets directly and recurringly. That money existed before. It was flowing to Salesforce, ServiceNow, HubSpot, Datadog, and a thousand other vendors.

Where CIOs Are Actually Cutting

The aggregate data confirms this. Morgan Stanley’s Q4 2025 CIO survey found IT budget growth moderating from 3.5% in 2025 to 3.4% in 2026. That headline number looks almost flat, which is why it’s misleading. The composition of that flat budget has changed dramatically.

Gartner projects AI spending will rise 80.8% in 2026, while communications services spending rises just 4.7% and device spending just 6.1%. You cannot have one category explode by 80% while overall budgets grow by 3.4% without something else shrinking in real terms.

Now layer in what’s actually happening at Anthropic. Their run rate jumped from $9B to $19B in roughly one quarter — a $10 billion annualized step-up in enterprise spend that landed faster than any single product category has ever grown. Dario Amodei confirmed $6 billion of that was added in February alone. No other software company in history has added that much incremental run rate that fast, at scale. And that’s before you add OpenAI’s parallel trajectory.

Deloitte’s 2025 tech value research found that more than half of enterprise respondents now allocate between 21% and 50% of their digital initiative budgets to AI — averaging 36%. For a company with $13 billion in revenue, that works out to roughly $700 million going to AI. That budget existed last year too. It just wasn’t going to AI.

Gartner’s John-David Lovelock framed it bluntly: runaway AI spending has put pressure on CIOs to ratchet down spending elsewhere, and many have turned to their services budgets first. The expectation now is that service firms are using AI to cut their own costs, and enterprises are demanding those savings be passed back. Software vendors are facing the same pressure from both directions.

The Structural Shift Under the Fear

There are two separate forces hitting software stocks simultaneously, and conflating them leads to bad analysis.

The first is the budget displacement effect: AI is eating into IT spend that used to flow to application software. This is happening right now, is measurable, and is directly tied to the growth of Anthropic, OpenAI, and the hyperscaler AI buildout. Approximately 75% of new hyperscaler infrastructure spending in 2026 — over $450 billion — is targeting AI infrastructure, redirected in significant part from enterprise software budgets. CIOs only have so much money. When Anthropic’s enterprise deals and OpenAI API costs start hitting six and seven figures per year, something else gets cut or delayed.

The second is the substitution fear: that AI agents will make traditional software irrelevant over time. This is legitimate but much harder to time. Macquarie analyst Steve Koenig described it bluntly: “The SaaS companies are wholeheartedly embracing agentic AI and putting a lot of investment dollars into this, but adoption is going really slowly. You’ve got this disconnect between what the enterprise software companies are saying and the reality of agentic AI on the ground.”

The market is pricing both of these simultaneously and not distinguishing between them. Budget displacement is a near-term, real earnings headwind. Substitution is a multi-year thesis that may or may not materialize at the scale the market fears. Wall Street is treating them as the same event.

What the Actual Earnings Are Showing

The fingerprints of budget pressure — not product displacement — are visible in the earnings data. Retention rates at enterprise software firms remain relatively high. Customers haven’t left en masse. They’ve slowed their buying.

Median SaaS revenue growth fell to 12.2% by Q4 2025, down from the 21% annual growth the sector was posting in 2023, with forecasts pointing to a further slowdown through at least Q2 2026. Median NRR across public SaaS has compressed to 101% — still positive, but barely, and down meaningfully from the 110%+ era of 2021-2022.

The pattern is consistent: deals are getting delayed, procurement is getting pickier, and “let’s add another module” conversations have been replaced by “what can we consolidate.” That’s not what churn looks like. That’s what budget exhaustion looks like.

Strip out price increases from recent SaaS earnings and the organic volume growth picture looks significantly worse than the headline numbers. The “growth” you see in many reports is pricing-driven, not seat-driven. That gets a different multiple, and the market is figuring that out.

How Much Is the Budget Shift vs. the Disruption Fear?

With Anthropic at ~$20B ARR and OpenAI at ~$25B ARR, and both still growing at rates that would have been unimaginable at this scale 12 months ago, the budget displacement math is now unavoidable.

Total global enterprise software spend is roughly $800-900 billion. Anthropic and OpenAI together are now pulling $40-50 billion annually from that pool — nearly all of it net new spend that didn’t exist two years ago. That’s 5-6% of the entire enterprise software market redirected to two companies in roughly two years. In a world where software budgets are growing at 3.4% total, that math is brutal. Some of that AI spend is additive. A meaningful chunk is substitutive.

The near-term budget displacement effect is probably responsible for 60-70% of the actual growth deceleration you’re seeing in software company earnings.

The other 30-40% is a mix of macro caution, deal elongation, tech stack consolidation (which was already happening before AI), and legitimate competitive pressure from AI-native alternatives at the lower end.

The stock price collapse, however, is still probably 80% fear and 20% fundamentals. Bank of America’s Vivek Arya called it an “indiscriminate selloff” driven by beliefs the firm considers “internally inconsistent” — that AI capex will simultaneously produce weak ROI and be so powerful it makes all existing software obsolete. Both outcomes cannot occur at once.

That point is worth holding onto. If AI is powerful enough to make ServiceNow irrelevant, then the $660 billion in AI infrastructure spend is justified and the hyperscalers win. If AI ROI disappoints and the bubble deflates, SaaS budgets come back. The market is priced as if both happen simultaneously to different companies. That’s not coherent.

Traditional Software Deals Are Also Often Now Just … Smaller

If you’re running a B2B software company right now, the honest read is this: your customers are not leaving you for AI. They are spending a meaningful chunk of their budget on AI that used to fund the expansion of your product. The deals aren’t disappearing — they’re getting smaller, slower, or deferred.

That is a very different problem than existential disruption, and it calls for a very different response.

Specifically:

• Expect slower NRR from expansion. The incremental seats and modules that used to sell easily are competing directly with AI budget. With median NRR already compressed to 101%, there is almost no cushion left.
• Double down on demonstrating ROI. CIOs are forcing every line item to justify itself now. Software that can’t show concrete, measurable business value is the first thing cut when Anthropic’s invoice shows up.
• Watch your new logo motion closely. New business is getting harder because AI experimentation is consuming the discretionary budget that used to fund software evaluations.
• Usage-based and outcome-based pricing models are structurally advantaged. Per-seat pricing is under pressure because AI is reducing headcount in some functions. Tie your pricing to value delivered, not humans using the product.

The companies getting hit hardest are those with weak ROI stories, high seat counts in functions AI is beginning to automate, and no clear path to becoming infrastructure rather than application layer. That’s not all of software. But it’s a large portion of the $400B+ mid-market SaaS economy.

The underlying thesis of software — that enterprises need reliable, integrated, mission-critical systems — hasn’t broken. What broke is the assumption that the software budget pool would keep growing 15-20% a year forever. That pool is now shared with a new category that went from zero to $40+ billion in two years, and is not slowing down.

That’s the real story. Not the apocalypse. Not a simple buying opportunity. A structural rebalancing that’s going to take several years to sort itself out — and is going to hurt a lot of software companies before it does.

Data sourced from: Bloomberg (March 3, 2026), Sacra, Epoch AI, Morgan Stanley Q4 2025 CIO Survey, Gartner 2026 IT Spending Forecast, Deloitte 2025 Tech Value Research, Aventis Advisors SaaS Valuation Report, Bain & Company, CIO.com, SaaStr