The State of Observability, 2026

300 IT leaders weigh in on spending, AI adoption, and maturity.

Grafana Labs and ETR, a Futurum company, surveyed 300 IT and engineering decision-makers from North America, Japan, Latin America, and Western Europe to learn where their observability spend is going and how they’re tracking it, how mature their practices really are, and how seriously they're putting AI into production.

The big takeaway: budgets are climbing almost everywhere and FinOps needs to catch up, consolidation is now table stakes rather than a nice-to-have; and trust in AI isn't keeping pace with the rapid shift from experimentation to production.

78% of respondents say spending increased over the past year, and 73% expect it to increase again over the next 12 months. There are a few reasons for this: teams point to new products added to their observability stack (61%), expanding observability coverage as AI-accelerated development creates more to watch (60%), business growth (51%), and vendor price increases (49%).

Budgets keep growing, almost everywhere

Globally, 78% of respondents report increased observability spending in the past year, and 73% expect another increase in the next 12 months.

AI features are the top driver

AI is driving this spend increase from two directions at once: teams are adding new AI-related features to their stack, and separately, AI-accelerated development is expanding what needs to be observed in the first place. The two reasons are essentially tied for the top spot (61% and 60%, respectively) — a sign that AI's effect on budgets isn't a single line item so much as a broad pressure showing up everywhere at once.

Most leaders say the increases are paying off, but not all can prove it

Spending is up, but the payoff is there, with 97% seeing at least some value in their increased observability spend. Of those, 61% say the increase in value they're getting is proportional to the increase in spend, but the other 36% say they see value but can't yet quantify it cleanly.

53% of organizations already describe their observability approach as relying on a unified, organization-wide platform, and another 25% say they're in transition toward one. Only 20% remain siloed by team or function. Consolidation is overwhelmingly the right path, with 79% rating it as "very" or "extremely" important. And intent is translating into real results: 86% of organizations describe their consolidation efforts as "very" or "somewhat" successful, and that jumps to 96% among organizations that have completed the move to a unified, org-wide platform. The minority that are still siloed by team or function are the ones lagging, at just 66% reporting any success so far.

AI is genuinely part of the observability stack now. 68% of organizations have moved past investigation into pilots or production use, with 47% saying AI is already running in production somewhere in their environment or is core to their strategy. Trust doesn't scale at the same rate as adoption, though: comfort with AI acting autonomously falls as the stakes rise.

Read-only AI is broadly trusted, but trust drops sharply once AI can act autonomously

85% are comfortable with AI doing read-only investigation; 84% are comfortable with AI recommending a fix. Comfort levels fall to 56% for AI making code or config changes, 52% for full autonomous remediations, and 50% for deploying changes to production. Leaders see AI as a tool that can help, but there's clearly a divide between those who are fully prepared to trust the outputs and those who still want people making final decisions.

AI for on-call helps reduce fatigue

67% report that AI-powered tooling has improved the on-call experience for engineering teams, including 28% of respondents who indicate AI has “significantly reduced” on-call burden and alert fatigue.

Reliability, not features, is the trust unlock

The top barrier to adoption is that the technology isn't yet reliable enough (26%), followed by regulatory/compliance constraints (23%) and customer trust (21%). Explainability and a proven accuracy track record are what respondents say would move the needle most (45% each).

80% of organizations are now building or operating LLM-powered features or products, with 42% already in production and 38% in development or pilot. But how they're watching those systems is far behind how they watch everything else.

Most orgs are building or piloting LLM products, but what are they observing?

Among those building or piloting LLM-powered applications, 63% are monitoring model output quality or accuracy, 53% are tracing LLM calls for latency and token usage, 44% watch cost attribution, 42% monitor hallucination rates, 41% look at user feedback loops, and 39% track prompt versioning.

AI observability tooling is still improvised

45% have pieced together an AI observability solution from a variety of open-source projects; 30% built something completely in-house; just 25% run a purpose-built paid solution.

Cost attribution is the top pain point

Among those respondents who are building or piloting LLM-powered applications, the top challenge is unclear cost attribution across models and teams (25%), followed by non-deterministic output quality (23%) and tools that weren't built for this use case (19%).

Cost discipline looks like it's becoming more common, but it’s still split across several teams: 90% have some kind of approach to FinOps, whether that’s a dedicated FinOps tool (31%), a dedicated team with BI tooling (26%), spreadsheets (17%), or an engineering DIY toolkit (16%).

Assigning dollar-value to observability outcomes is not as easy as it sounds

27% of respondents who assign a specific dollar amount to observability outcomes say it’s an informal process and can be “inconsistent”.

When it comes to AI, predictability still beats outcome-based pricing

50% say a straightforward per-user-per-month model is easiest to budget for and defend internally—token-based (19%) and outcome-based pricing (11%) trail well behind.

AI is the throughline, not just a single trend

More than half of all respondents (53%) predict AI itself will be at the center of the most consequential trend in observability over the next two years — they're just split on which side of it matters more. 28% point to AI driving operations directly, or "systems that run themselves," while 25% point the opposite direction: the need to actually observe the AI agents and applications organizations are now shipping. Put together, that's not two competing predictions so much as one bigger one — AI is about to change both sides of the observability relationship at once.

Security convergence and cost intelligence aren't far behind

22% point to observability-security convergence and 17% to cost intelligence built directly into the observability layer — both real, adjacent priorities with their own constituency, not afterthoughts to the AI conversation.

The organizations pulling ahead are the ones closing the gap between what they say matters and what they've actually shipped: real consolidation instead of stated intent, a defensible dollar figure behind their observability spend, and a clear-eyed view of exactly how much autonomy they're ready to hand to AI, and where.

ETR, a Futurum company, surveyed 300 IT and engineering decision-makers — Director/VP level or above, each with budget or purchasing influence over observability tooling — across the United States (150), Japan (50), Latin America (50), and Western Europe (50) in June 2026. Figures are shown unweighted, and percentages may not sum to 100 due to rounding. Year-over-year comparisons draw on Grafana Labs' 2025 survey of 150 North American IT leaders, matched against the same tracked panel surveyed again in 2026.