10 min Reading Time

On 24 March 2026, the iShares Expanded Tech-Software ETF plunged 4.3% in a single trading day. The trigger? A report from The Information revealing that AWS is internally building AI agents for sales and business development. The news struck an industry already down $2 trillion in market capitalization since the start of the year. Simultaneously, Microsoft revealed that only 3.3% of its 450 million M365 users pay for the Copilot add-on. The era of per-seat pricing is ending – and DACH companies face a strategic inflection point with exactly four months to act.

TL;DR

• Massive stock decline: Software equities have lost roughly $2 trillion in market capitalization since late 2025. UiPath, HubSpot, and Atlassian each plunged 8-9% on 24 March (Bloomberg, March 2026).
• Copilot disillusionment: Only 15 million of Microsoft’s 450 million M365 users pay for Copilot. The share of paying subscribers fell from 18.8% in July 2025 to 11.5% in January 2026 (The Register, February 2026).
• AWS builds AI agents: Amazon is developing autonomous agents designed to replace tasks currently performed by thousands of technical specialists in sales and business development.
• Pricing upheaval: Intercom prices its AI agent “Fin” at $0.99 per resolved ticket. Salesforce responds with its “Agentic Enterprise License Agreement” – a flat-rate model.
• EU AI Act deadline: Compliance obligations for high-risk AI systems take effect on 2 August 2026. Companies deploying uncontrolled agents now will face serious regulatory issues in four months.
• Gartner forecast: By end-2026, 40% of enterprise applications will integrate task-specific AI agents – up from under 5% in 2025.

What’s behind the stock slump?

The news that AWS is building AI agents for internal sales processes wasn’t the software industry’s first shock. On 29 January 2026, SAP disappointed with its cloud revenue forecast. That same day, ServiceNow reported weak earnings. The European software giant and the US platform provider jointly triggered the sell-off that traders at Jefferies immediately dubbed the SaaSpocalypse.

On 3 February came the next blow: new AI automation tools wiped out $285 billion in market value in a single day. Since then, the downward spiral has accelerated. The IGV Index – the most important benchmark for publicly traded software firms – lost over 21% from the start of 2026 through 24 March 2026. Salesforce shed more than 30%, Workday 33%, and Atlassian 35%.

That SAP – a German company – stands at the epicenter makes the SaaSpocalypse deeply personal for DACH decision-makers. Anyone managing SAP in their portfolio, budgeting SAP licenses, or planning SAP projects feels this uncertainty directly.

Source: J.P. Morgan / Fortune, March 2026

The Copilot Warning: What Microsoft’s Numbers Reveal About the Seat Model

Before debating abstract AI agents, consider the industry’s largest AI experiment: Microsoft 365 Copilot. 450 million users. $30 per seat per month. The strongest distribution advantage in software history. And yet only 15 million users – just 3.3% – pay.

That figure alone would be alarming enough. But the trend worsens: the share of paying subscribers dropped from 18.8% in July 2025 to 11.5% in January 2026. Microsoft gains users in absolute terms – but conversion is falling. Activation stands at just 35.8%. Two out of three companies buying Copilot licenses don’t use them.

Source: The Register / Motley Fool, February 2026

What this means: Even Microsoft – with the industry’s strongest bundling advantage – cannot sustain per-seat pricing for AI. Companies buy Copilot licenses, realize ROI isn’t materializing, and let them lapse. This isn’t an adoption problem. It’s a pricing-model problem. $30 per seat per month for a tool that sometimes helps – and sometimes hallucinates – is hard for CFOs to justify.

Microsoft’s response? A new SMB SKU, “Copilot Business,” priced at just $21 per month. That’s not a discount – it’s an admission.

Per-Seat Is Dead: What Comes Next

The business model that powered the cloud industry for the past 15 years rests on a simple equation: more employees = more licenses = more revenue. AI agents break that equation. If one agent does the work of five human users, a company needs one seat – not five.

Atlassian reported its first-ever decline in enterprise seat counts in Q1 2026. This isn’t an outlier: Workday cut 8.5% of its workforce. Both serve precisely the workflows – task tracking, data entry, customer logging – that AI agents automate most efficiently.

Vendor responses reveal how seriously they’re taking the crisis. Salesforce has launched an “Agentic Enterprise License Agreement” (AELA): a flat-rate model moving away from seat counting. ServiceNow shifts to usage- and outcome-based pricing. Intercom prices its AI agent “Fin” at $0.99 per fully resolved ticket – not per seat, not per month, but per result.

“Wall Street bakes in a doomsday scenario that is likely exaggerated.”
Dan Ives, Managing Director, Wedbush Securities (March 2026)

Gartner forecasts that by 2030, roughly 35% of standalone SaaS solutions will be replaced – or absorbed into broader agent ecosystems – by AI agents. The agent-based AI market is expected to grow from $8.5 billion to $47 billion by 2030. Simultaneously, Gartner expects 40% of SaaS spending to shift to usage-, agent-, or outcome-based pricing models.

The Klarna Lesson: Why “AI-First” Failed

In February 2024, Klarna announced its AI assistant had handled 2.3 million customer service conversations in one month – across 23 markets and 35 languages – covering roughly 66% of all chats. Upfront cost: $2-3 million. Projected annual savings: $40 million. Seven hundred roles were eliminated. Tech media hailed it as proof AI agents could replace human labor.

One year later, CEO Sebastian Siemiatkowski publicly backtracked. Quality had suffered. Klarna rehired human staff – not to revert to the old model, but to adopt a human-hybrid approach: AI for standard cases, humans for complex issues.

The lesson for DACH CIOs is threefold. First: AI agents can automate routine customer service tasks. That’s proven. Second: Skipping quality control and scaling immediately exacts a price – in customer satisfaction. Third: Those $2-3 million implementation costs are only part of the bill. SearchUnify estimates integration and change management account for 35-45% of first-year TCO. Omitting those from planning leads to 40-60% underestimation of total cost.

The Klarna story isn’t an argument against AI agents. It’s an argument against shortcuts.

Which SaaS Categories Will Survive?

Not all software categories face equal risk. Vulnerability depends on how reliant a product is on human routine activity.

High Replacement Risk: Point solutions covering a single workflow – ticketing systems, basic CRM functions, expense management, meeting scheduling, data cleansing. Gartner estimates 35% of these standalone tools will be replaced by agents by 2030.

Medium Risk: Collaboration tools like Slack, Teams, or Asana. Here, the model shifts: instead of licensing ten human users, two people collaborate with three agents. Seat count falls – but the platform remains relevant as an orchestration layer.

Low Risk: Systems of Record like SAP, Oracle, or Workday HR. These store mission-critical enterprise data and comply with regulatory requirements. AI agents won’t replace them – they’ll use them as frontends. ERP vendors offering early agent APIs may even benefit.

Growth Potential: Infrastructure for agent deployment – monitoring, observability, identity management for non-human identities. This market has existed for only a few months – and is growing rapidly. In late March 2026, AWS launched Bedrock AgentCore: a stateful runtime with memory streaming for persistent agents. Google rebranded Agentspace as “Gemini Enterprise” and offers a no-code agent designer with over 1,000 prebuilt partner agents.

Bull Case vs. Bear Case

J.P. Morgan titled its analysis Software Collapse Broadens with Nowhere to Hide. Yet the bank also argued that full replacement of SaaS by AI agents is a narrative relevant only after 2028 – at best. Today, Copilot-style features dominate – not autonomous agents.

Bank of America identified a paradox: Investors punish hyperscalers because their AI investments may yield weak returns. Simultaneously, they destroy software valuations because AI adoption will be so pervasive that existing software becomes obsolete. Both can’t be true at once.

The bear case is zero-sum: AI agents compress existing SaaS revenue by eliminating seats. The bull case is positive-sum: surviving software firms unlock the $6-trillion knowledge-work market – previously inaccessible to software.

Truth likely lies between. And Microsoft’s Copilot numbers show precisely where: the market wants AI. It just doesn’t want to pay for it per seat.

The DACH Factor: Why Europe Is Affected Differently

The European SaaS market differs structurally from the US market. Bitkom estimates Germany’s enterprise software market at €35 billion in 2026. If Gartner is right – and 40% of SaaS spend shifts to new pricing models by 2030 – then €14 billion in Germany alone hangs in the balance. This isn’t an abstract US stock-market phenomenon. It affects every mid-market CIO’s budget planning.

Meanwhile, stronger regulation – NIS2, the EU AI Act, GDPR – acts as both brake and opportunity. European vendors embedding compliance-by-design into their agent platforms could gain a competitive edge over US providers. GDPR requirements complicate deploying cloud-based AI agents that process corporate data. On-premises and private-cloud agent solutions will play a larger role in DACH than in the US. Data sovereignty becomes a key differentiator.

SAP plays a dual role here. As the catalyst of the sell-off, the company symbolizes the per-seat model’s failure. Yet as a deeply embedded System of Record across DACH, SAP is also best positioned to ride out the agent wave. SAP systems store mission-critical data. AI agents won’t replace them – they’ll use them as frontends. Understanding the SAP ecosystem explains why the bear case for Systems of Record is overstated.

The Four-Month Deadline: EU AI Act and What It Means for Agent Deployments

On 2 August 2026, compliance obligations for Annex III high-risk AI systems under the EU AI Act take effect. This applies to AI agents making decisions in employment, credit, education, or critical infrastructure. Penalties: up to €35 million or 7% of global annual turnover – exceeding GDPR fines.

For CIOs evaluating AI agents today, a concrete problem emerges: high-risk AI systems processing personal data trigger both a Fundamental Rights Impact Assessment under Article 27 of the AI Act and a Data Protection Impact Assessment under Article 35 of the GDPR. Multi-step agents generate distributed logs – making traceability requirements significantly harder to meet. Deploying agents today without designing for compliance creates technical debt due in four months.

In June 2025, the BSI (Federal Office for Information Security) published a criteria catalog for using generative AI in federal administration. The catalog specifically addresses agent use cases: chatbots, text summarization, translation. Though not yet binding for the private sector, it defines the minimum standard regulators will reference.

The consequence: anyone planning agent deployments should treat the BSI paper as baseline – and embed Article 27 AI Act and Article 35 GDPR assessments into architecture design now. Not post-launch. Not in Phase 2. Now.

What DACH Companies Should Do Right Now

The SaaSpocalypse isn’t a US-only phenomenon. Every DACH company paying for SaaS licenses per seat must reassess its software strategy. Five concrete steps:

1. Audit your SaaS portfolio. Which tools are licensed per seat? Which address tasks an AI agent could perform? Rule of thumb: Any software whose core function is data entry, standard communication, or rule-based decision-making is on the chopping block.

2. Renegotiate pricing models. Signing three-year contracts with per-seat pricing today locks you into a model likely obsolete in 18 months. Usage- or outcome-based models offer greater flexibility. Reference: Intercom prices AI-resolved tickets at $0.99 each. Salesforce offers an agent flat-rate via its AELA. These are the benchmarks to bring into vendor negotiations.

3. Re-evaluate build-vs.-buy. Cost structures are shifting. An internal AI agent trained on proprietary data may cost less than five SaaS licenses. Prerequisites: clean data infrastructure and a platform engineering strategy, enabling agent deployments.

4. Build agent readiness. GPU capacity, API gateways with rate limiting, observability stacks for agent monitoring, and identity management for non-human identities. AWS Bedrock AgentCore, Azure AI Agent Service, and Google Gemini Enterprise each offer distinct frameworks. Ecosystem choice will be strategic over the next 12 months.

5. Bake compliance in from day one. Before any agent deployment: conduct Article 27 AI Act impact assessment and Article 35 GDPR DPIA. Use the BSI criteria catalog as baseline. Multi-step agents require traceability logs. This isn’t Phase 2. It’s Day 1.

Conclusion

The SaaSpocalypse is neither hysteria nor apocalypse. It marks the end of a pricing model that worked for two decades. Microsoft’s Copilot numbers deliver buyer-side proof: 3.3% uptake among 450 million potential users shows per-seat pricing fails for AI. Klarna’s reversal proves “AI-first” without quality control is an expensive detour. AWS, Google, and Salesforce signal the direction with their agent platforms.

The concrete next step: In your next budget cycle, allocate a line item for agent infrastructure – while simultaneously questioning every SaaS renewal beyond 12 months. Place the BSI paper and EU AI Act Article 27 on your desk. Companies actively shaping this transition will emerge with lower software costs and higher process automation. Those who wait will keep paying for seats nobody needs.

Frequently Asked Questions

What exactly is the SaaSpocalypse?

The SaaSpocalypse refers to the massive stock decline among software equities since early 2026. Triggers included disappointing quarterly results from SAP and ServiceNow on 29 January – and fears that AI agents undermine per-seat licensing. By mid-March 2026, the sector had lost roughly $2 trillion in market capitalization.

Why does Microsoft Copilot have only 3.3% uptake?

Despite the industry’s strongest bundling advantage (450 million M365 users), only 15 million pay for Copilot. Activation stands at 35.8%. The issue is less technology than pricing: $30 per seat per month for a tool with inconsistent quality is hard for CFOs to justify. Microsoft has already introduced a cheaper SMB SKU ($21).

What is AWS actually building?

In late March 2026, AWS unveiled “Partner Central Agents” built on Bedrock AgentCore: AI-powered co-selling with pipeline insights and automated CRM population. Concurrently, Bedrock AgentCore delivers a stateful runtime with memory streaming for persistent agents. An AWS spokesperson confirmed the agent is designed to aggregate deep expertise across all AWS domains.

What is Intercom Fin – and why does it matter?

Intercom prices its AI agent “Fin” at $0.99 per fully resolved ticket. This is outcome-based pricing: costs arise only when the problem is solved. Compared to per-seat licenses costing $50-$150 per month, it’s a paradigm shift. Salesforce responded with its “Agentic Enterprise License Agreement” – a flat-rate alternative.

Which software categories are most at risk?

Most vulnerable are point solutions: ticketing systems, basic CRM functions, expense management, meeting scheduling. Gartner estimates 35% of these standalone tools will be replaced by agents by 2030. Systems of Record (SAP, Oracle, Workday HR) face lower risk, as they store mission-critical data and comply with regulatory mandates.

What does the EU AI Act mean for agent deployments?

On 2 August 2026, compliance obligations for high-risk AI systems take effect. AI agents making decisions in employment, credit, or critical infrastructure require an impact assessment under Article 27 of the AI Act. Combined with the GDPR’s Article 35 DPIA, this creates dual compliance overhead. Penalties: up to €35 million or 7% of annual global turnover.

Should DACH companies cancel SaaS contracts now?

Not universally. First, audit your SaaS portfolio: which tools address tasks an AI agent could handle? Avoid new three-year per-seat contracts. Instead, negotiate usage- or outcome-based models – and include exit clauses. Klarna’s experience shows: plan for quality control and change management before scaling.

Further Reading

SaaS Crisis 2026: Why Salesforce Lost 26%

AI-Native Consulting: The Future of IT Advisory

Platform Engineering 2026: Internal Developer Platforms

Sovereignty Washing: Honestly Assessing Data Sovereignty

More from the MBF Media Network

Digital Chiefs: AI Liability at the Board Level

SecurityToday: Copilot as a Security Risk

MyBusinessFuture: Cyber Resilience Act

Header Image Source: Pexels / AlphaTradeZone (px:5784807)