Buyer caution and AI ops limits

Big companies are getting choosier about tech spending just as AI projects are getting harder to run. Atos reported an 11% organic revenue drop in the first quarter, and Datadog said operational complexity is now the main obstacle to scaling AI reliably. (atosgroup.com) (datadoghq.com) Atos said first-quarter revenue was €1.739 billion, or €1.640 billion on its go-forward perimeter, with about 3 percentage points of the decline tied to low-profitability contracts it is exiting. Chief executive Philippe Salle said a “volatile macroeconomic environment” is slowing commercial decision-making, especially in North America, where some clients have taken a “wait-and-see approach.” (atosgroup.com) The French services group also said its book-to-bill ratio reached 87% in Q1, up about 4 points from a strong Q1 2025, and its qualified pipeline grew by about €900 million in the quarter. Atos kept its 2026 operating-margin target at about 7% and narrowed its full-year organic revenue outlook to a decline of 1% to 5%. (atosgroup.com) AI systems are now often built from several models, tools, and software agents working together, more like a supply chain than a single chatbot. Datadog’s State of AI Engineering 2026 report said that setup is creating reliability problems in production even when the underlying models are improving. (datadoghq.com) (markets.businessinsider.com) Datadog said 69% of companies now use three or more AI models, about 5% of AI model requests fail in production, and nearly 60% of those failures come from capacity limits. The company also said OpenAI remained the most widely used provider at 63% share, while adoption of Google Gemini and Anthropic Claude rose by 20 and 23 percentage points. (markets.businessinsider.com) The report said agent framework adoption doubled year over year, while the amount of data sent to models per request more than doubled for median-use teams and quadrupled for heavy users. Datadog chief product officer Yanbing Li said AI is starting to resemble the early cloud era, when programmability expanded faster than companies’ ability to monitor and control complex systems. (markets.businessinsider.com) (itwire.com) Those two updates land a day apart, on April 21, 2026, from vendors selling different parts of the enterprise technology stack. One is seeing customers trim contracts and delay commitments; the other is telling customers that AI budgets now need monitoring, capacity planning, and failure management, not just better models. (atosgroup.com) (datadoghq.com) Atos is still pitching growth areas including agentic AI, cybersecurity, and sovereign services, and it said liquidity stood at €1.736 billion at the end of March after the Bull transaction closed on March 31. Datadog, for its part, is using the report to argue that observability software is becoming part of the basic plumbing for AI in production. (atosgroup.com) (datadoghq.com) The near-term test is whether enterprise buyers loosen spending and whether AI teams can cut failure rates as usage rises. For now, the signals from both companies point to the same constraint: fewer impulse purchases, and more scrutiny of what can run cleanly at scale. (atosgroup.com) (markets.businessinsider.com)