The enterprise AI market crossed €31.4 billion in 2025. Most of that money isn't buying software — it's buying workers.
- AI is commoditizing work itself, not just automating tasks — a new category called Work-as-a-Service is emerging where enterprises subscribe to outcomes rather than license tools.
- Spending is accelerating dramatically: global enterprise AI spend hit €31.4B in 2025 (3.2× year-over-year), yet 95% of organizations report no measurable P&L impact.
- The gap between investment and results won't close through better models. It will close through better orchestration, data pipelines, and the interoperability standards being built right now.
The Numbers That Don't Add Up — Yet
Digital labor is no longer theoretical. In 2025, enterprises spent €31.4B on generative AI, up 3.2× year-over-year. The results are arriving in forms that feel less like software upgrades and more like new hires.
Salesforce says Agentforce ARR reached €677M in Q4 FY2026 (+169% year-over-year), with 29,000 Agentforce deals closed, and has branded Agentforce "the first digital labor platform for enterprises". The headcount implications are already visible: Salesforce disclosed it had reduced its own customer support staff from 9,000 to 5,000 through agentic AI. Microsoft 365 Copilot reached 15 million paid seats, with seat adds up over 160% year-over-year. Cognition reported Devin revenue scaling from €0.85M ARR (September 2024) to €61.9M ARR (June 2025) — a trajectory no software product reaches in nine months.
And yet the ground truth demands honesty. An MIT Project NANDA survey found 95% of organizations reported no measurable P&L impact from their generative AI initiatives. McKinsey's global survey found only 23% of organizations are scaling AI agents, while 39% are still experimenting.
This is the central paradox of AI in 2025: massive, accelerating investment; minimal, stubbornly hard to measure returns. Understanding why requires understanding what, exactly, is being bought.
From Software to Labor: The Naming Shift That Matters
The terminology around AI is converging on something specific. Thoughtworks frames it as Service-as-Software: paying for autonomous outcomes rather than software seats. Others use Agentic AI as a Service (AaaS) or Agent-as-a-Service to describe the delivery of goal-driven agents via platforms and APIs.
A clearer framing for the category is Work-as-a-Service (WaaS). The unit you buy is work itself — tickets resolved, briefs drafted, invoices reconciled, campaigns shipped — delivered by AI agents as an operational service rather than a software license.
Pricing is the clearest confession of intent. Salesforce lists Agentforce at €1.70 per conversation. Companies like 11x sell named "digital workers" such as Alice and Julian, explicitly targeting hiring budgets rather than software procurement. OpenAI has reportedly explored pricing specialized agents up to €17,000 per month. These are not software pricing models. They are headcount pricing models.
McKinsey estimates that currently demonstrated technologies could automate activities equivalent to ~57% of US work hours. The World Economic Forum projects 170 million jobs created and 92 million displaced by 2030, a net gain of 78 million with substantial skills disruption in between.
This matters beyond terminology. When the unit of value is work completed rather than software licensed, the economics of hiring, procurement, and organizational design change fundamentally.
The Infrastructure Behind the Shift
Venture capital does not invest in vibes. It invests in trajectories. Crunchbase estimates AI companies raised €179B in 2025, roughly half of all global venture funding, up ~85% year-over-year. Gartner projects worldwide AI spending to reach $2.5 trillion in 2026.
The standards layer is solidifying rapidly. Anthropic's Model Context Protocol (MCP) had 10,000+ active servers by late 2025. The Agent2Agent (A2A) protocol standardizes how agents communicate with each other. The Agentic AI Foundation (AAIF), co-founded by Anthropic, OpenAI, and Block under the Linux Foundation, has grown to 146 members. The AI agents market is estimated at €6.47B in 2025, projected to reach €42.7B by 2030.
Why the Gap Will Close — and Where
The 95% figure deserves more than shock. It deserves interrogation. If the technology is working at the product level (Agentforce is closing 29,000 deals; Devin went from €0.85M to €61.9M ARR in nine months), why isn't it registering at the P&L level?
The answer is not the models. It is the layers around the models: the data pipelines that feed them, the orchestration infrastructure that coordinates them, and the identity and governance systems that let enterprises trust them enough to give them consequential tasks. Gartner warns that 63% of organizations either do not have, or are unsure they have, the right data management practices for AI, and expects 60% of AI projects to be abandoned without AI-ready data.
The P&L impact of Work-as-a-Service will become measurable when the connective tissue catches up — when orchestration protocols are standardized, when enterprise data is AI-ready, and when governance frameworks give organizations the confidence to deploy agents on consequential tasks rather than sandboxed demos. That process is underway, and the speed of standardization in 2025 (MCP, A2A, AAIF in a single year) suggests it will move faster than most enterprise technology transitions.
The question then is not whether WaaS arrives. It is who controls the infrastructure on which it runs — and whether the current US-China duopoly in AI is the only architecture on offer.
This article is part of a series drawn from the Mycel AI white paper "Anatomy of a Third Force: The AI Stack, Digital Labor, and the India-Europe Corridor." The companion articles will cover the willseven-layer architecture of the full AI stack, and why the India-Europe corridor may be the most underexploited opportunity in global AI.