Why VCs Are Pulling Back from SaaS and Doubling Down on AI Infra

Pitchbook released its Q1 2026 venture data in late March and the shift is no longer subtle. SaaS funding as a category declined 34% year-over-year, while AI infrastructure funding - compute, training pipelines, model serving, enterprise AI platforms - increased 89%. The allocation gap between those two numbers is the story of where venture capital thinks the next decade of software value is going to be created.

Having spent the quarter talking to partners at a dozen funds, from micro-VCs to multi-billion-dollar growth funds, I can tell you the reasoning is more nuanced than "AI is hot." There are structural arguments for why the venture math on traditional SaaS has changed.

The Multiple Compression Problem

Traditional SaaS companies built their valuations on a relatively stable set of multiples: 8-15x ARR for growth-stage companies, 20-40x for the fastest growers with clear category leadership. These multiples were supported by the durability of subscription revenue, the predictability of net revenue retention metrics, and the structural moat created by switching costs.

Those multiples have compressed significantly. The median growth-stage SaaS multiple in Q1 2026 sits at approximately 6x ARR, down from 12x in 2021 and still declining in several subsegments. There are two drivers:

First, AI is genuinely threatening the defensibility of certain SaaS categories. If a well-prompted LLM can do 80% of what a $200/month vertical SaaS tool does, the switching cost of the SaaS tool is reduced. Investors are repricing this risk.

Second, the public market comparables are compressed. SaaS public multiples set the ceiling for private market multiples, and public SaaS has been under sustained multiple pressure since 2022. The free money environment that supported 30x ARR multiples does not exist anymore.

The implication: for VCs building a fund-returning investment thesis, SaaS requires better companies to generate the same returns it did in 2020. The denominator problem is real.

Why AI Infrastructure Is Different

The AI infrastructure thesis rests on a different set of structural arguments.

Capital intensity creates barriers. Training a frontier model costs hundreds of millions of dollars. Deploying it at scale costs hundreds of millions more. These capital requirements create moats that most potential entrants cannot cross. Investors in AI infra are betting on a relatively small number of winners in each layer of the stack, each with pricing power that SaaS companies rarely achieve.

The enterprise sales motion is just beginning. Companies like Salesforce and ServiceNow are spending billions to embed AI into their existing platforms. Every enterprise software category is being rebuilt with AI capability. The companies that own the underlying model infrastructure, training tooling, and enterprise deployment platforms are upstream of all of this activity. Upstream ownership in technology transitions tends to be more valuable than application-layer ownership.

The switching cost is higher, not lower. Enterprises that train proprietary models on their data, build fine-tuning pipelines, and integrate inference infrastructure into their production systems have created meaningful switching costs. The OpenAI API is not the whole picture - the switching cost is the months of engineering work to rebuild integrations.

The SaaS Categories Still Attracting Capital

Not all SaaS is being abandoned. Investors are being more selective, and the categories still attracting significant interest share common characteristics: clear ROI proof, strong enterprise motion, and limited AI displacement risk.

Compliance and regulatory SaaS: GDPR, SOC 2, and emerging AI regulation create demand for specialized compliance software that cannot easily be replaced by a general AI. Human accountability requirements mean humans need tools, not just AI.

Developer infrastructure: CI/CD, monitoring, security tooling. The market for Cursor (AI-native developer tools) represents the intersection of SaaS and AI that is still attracting significant capital, because it is not competing with AI - it is distributing AI capability to developers.

Vertical SaaS with deep workflow integration: Industry-specific software where the switching cost is embedded in workflows, not just data. Healthcare systems, construction management, legal practice management. These are harder for AI to displace because the compliance and workflow integration is the product.

What This Means for Founders

The funding environment has changed in ways that require different strategies for SaaS founders.

The "grow fast and optimize later" model that worked in 2020-2021 is less available. Investors are looking for capital efficiency - ARR per dollar raised, path to profitability on current capital, evidence of durable retention. The Series A bar has moved.

Differentiation on AI is table stakes, not a moat. Every SaaS pitch that doesn't include meaningful AI capability is immediately questioned. But "we use AI" has become so common it generates skepticism rather than excitement. The differentiation has to be specific: which AI capabilities, how they are defensible, what makes your implementation better than a competitor adding the same LLM API calls.

For founders considering a raise in Q2 2026: the HubSpot playbook of building a category-defining product with strong NRR and clear enterprise motion is still fundable. The "disrupting a legacy category with a better UX" story is much harder. Investors have seen too many companies disrupted in turn by the AI they embedded.

The honest advice from most partners I spoke with: if you are at $1M ARR or below and building horizontal SaaS without a clear AI-native moat, the funding environment will be challenging in 2026. If you are at $3M+ ARR with strong NRR and a specific vertical or workflow integration, capital is still available - just at lower multiples than 2021.

The AI infrastructure wave is real and the capital allocation reflects it. But the best SaaS companies will continue to get funded, because revenue predictability never stops being attractive to investors - even in a paradigm shift.