Software stocks are now trading below the S&P 500 for the first time in history.

The forward price-to-earnings multiple for application software peaked at 84x in mid-2020. By March 2026 it hit 22.7x — below the market average. The iShares Software ETF is down 30% from its September peak. Over a trillion dollars in market capitalization gone.

This is not a correction. Corrections are cyclical. This is structural. The market is not saying "software is temporarily overvalued." It's saying "we're not sure the business model still works."

The SaaS model was built on one assumption: every person who does work needs a software seat. Hire more reps, buy more licenses. Revenue grew with headcount. The entire model compounded predictably for two decades.

Automation Anywhere published data in April from over 70 enterprise deployments: AI agents resolve more than 80% of employee service requests autonomously, cutting IT service management licensing costs by up to 50%. Publicis Sapient is reducing SaaS licenses by approximately 50%, replacing major platforms including Adobe with generative AI tools. Workday announced 8.5% layoffs not because AI failed but because it succeeded — customers need fewer seats when fewer people do more work.

The mechanism is simple. When one employee with AI agents does the work of five, your productivity gain is the vendor's revenue leak. The more AI works, the fewer seats you need. A Databricks survey found multi-agent system usage spiked 327% in early 2026. This isn't a trend. It's a phase transition.

The vendors know. Some are shifting to usage-based pricing. Others to outcome-based. Adobe moved to "Generative Credits." Salesforce is trying to become the orchestrator of AI agent workforces. None of these transitions are smooth — sales compensation, board presentations, investor models, everything was built around predictable per-seat growth.

The vendors caught in the worst position are the ones whose core value was executing a task AI can now do directly: basic data analysis, template reporting, commodity workflow automation. If your product's value was "a human uses our interface to do X," and now an agent does X without the interface, you don't have a pricing problem.

You have an existential problem.

Here is where it stops being a Wall Street story and starts being a European competitiveness story.

The CEPR published new data in April quantifying the AI adoption gap with unusual precision. 43% of US workers have adopted AI tools. Italy: 26%. Germany: 31%. France: 28%. UK: 36%. But the intensity gap is starker — 5.2% of US work hours are spent using AI. In Germany, France, and Italy, less than a third of that.

Eurostat: 19.95% of EU enterprises used AI in 2025. OECD: large firms at 40%, SMEs at 11.9%. US productivity has grown nearly 90% since 1995. Euro area: 30%. The Draghi Report identified AI as the window to close this gap. The data says the window is being watched, not used.

Now map this onto the SaaS repricing.

US enterprises at 88% AI adoption are the ones actively compressing their SaaS spending — replacing seats with agents, building internal tools, renegotiating contracts. They're driving the repricing.

European SMEs at 13% adoption are still buying SaaS at pre-disruption prices. Full subscription costs for tools designed around the assumption that humans do the work — while American competitors operate with leaner stacks and lower per-unit costs. European SMEs aren't driving the repricing. They're going to absorb it.

Asia adds another dimension. China and India lead enterprise deployment at 58% and 57% respectively. South Korea made the single biggest jump of any country in H2 2025. The Asia-Pacific AI market is expected to grow at the highest CAGR through 2034. Asian competitors are not waiting for European regulatory frameworks to settle.

Everything above would be serious enough if AI infrastructure were priced at cost. It's not.

OpenAI generated $3.7 billion in revenue in 2025 and lost $5 billion. Hyperscalers committed over $600 billion in AI infrastructure capex for 2026. Amazon: $200 billion. Google: $175-185 billion. Microsoft: $145 billion. Meta: $115-135 billion. Average enterprise LLM spend hit $7 million per company, nearly triple the year before.

This is the classic platform subsidy cycle. The playbook is the same every time:

Bill Gurley calls it the "negative gross margin" problem. The current generation of AI tools is priced to maximize adoption, not profitability. The providers are racing for market share on the bet that once your processes, your data pipelines, your agent architectures are built around their APIs, you won't leave when prices rise.

They're probably right. Switching AI providers is not like switching SaaS vendors. Your prompts, your fine-tuning, your integration logic — all model-specific. The lock-in is subtler than a SaaS contract but deeper.

The question isn't "how cheap is the AI?", it's "what will this cost in three years when the provider needs to justify a $183 billion valuation?"

European SMEs are walking into a trap with two jaws.

First: paying for SaaS that's being structurally devalued by AI adoption they're not participating in. Every month of delayed adoption is a month of paying pre-disruption prices for tools whose competitive value is declining. The gap compounds.

Second: when they do adopt AI — and they will, because competitive pressure will force it — they'll adopt into an infrastructure controlled by a handful of American companies running a subsidy playbook. The ECB's latest survey shows euro-area SMEs plan to allocate 9% of total investment to AI in 2026. That investment will flow overwhelmingly to US-based providers. The money flows out. The dependency flows in.

Adopt now, not later. The European instinct to wait for regulatory clarity before deploying is understandable and expensive. The AI Act, the Product Liability Directive, the DSA — none of them prohibit adoption. They regulate it. There's a difference between guardrails and walls. European SMEs are building walls while their competitors build roads.

Diversify your AI infrastructure. Open-weight models — Llama, Mistral, DeepSeek, Qwen — are narrowing the gap with proprietary systems. Self-hosting inference is increasingly viable for predictable workloads. The same logic that makes single-vendor SaaS dependency risky makes single-vendor AI dependency riskier.

Audit your SaaS stack before it's audited for you. The repricing in public markets is a leading indicator. The operational repricing — where the tools you're paying for deliver less competitive advantage — follows with a lag. The companies that run this audit in 2026 will have choices. The companies that run it in 2028 will have constraints.

Price the subsidy into your planning. If your unit economics work at 3x current API costs, you're building on solid ground. If they only work at today's prices, you're building on a promotion.

The SaaS model isn't dying everywhere. Durable platforms with deep integration, proprietary data, and genuine workflow lock-in will survive — they'll become the infrastructure AI agents operate through, not the interfaces humans click on.

But "SaaS will survive" is cold comfort if you're the one still paying per-seat prices for a tool your competitor replaced with three API calls and a script.

For European SMEs, the stakes are specific.

AI is the most powerful productivity lever available to close the gap with the US. But adopting it through American infrastructure, at subsidized prices, without diversification or self-hosting capability — that's not closing the gap. That's financing someone else's growth while hoping they don't raise prices.

The SaaS era taught one lesson that transfers directly: whoever controls the platform captures the margin. In the AI era, the platform is the model.

Fabio Lauria

CEO & Founder, ELECTE

Every week we explore AI without the hype — with data, analysis and an independent perspective.

SaaS market repricing SaaStr: The SaaS Rout of 2026 Is Even Worse Than You Think | SaaStr: The 2026 SaaS Crash — It's Not What You Think | Intellectia: Will AI Disrupt the SaaS Business Model?

Seat compression and enterprise data Automation Anywhere: AI Agents Force Rethink of SaaS Pricing (April 2026) | Built In: How AI Agents Are Disrupting SaaS | Orbilontech: AI Agents Replacing SaaS Tools 2026

The AI adoption gap CEPR/VoxEU: Differences in AI Adoption in Europe and the US (April 2026) | CEPR/VoxEU: How AI Is Affecting Productivity and Jobs in Europe | OECD: AI Adoption by Small and Medium-Sized Enterprises | ECB: AI and the Euro Area Economy | World Economic Forum: Europe Is Lagging in AI Adoption

AI infrastructure economics SoftwareSeni: The AI Inference Market in 2025 | Oplexa: AI Inference Cost Crisis 2026 |

Reference Medium/Chernychko: Is AI Killing SaaS? Why Most SaaS Products Are Built to Last