McKinsey says enterprise AI is finally 'on the road to ROI'

McKinsey’s State of AI in 2026 survey of 1,719 leaders finds AI ROI remains elusive: 37% attribute at least some EBIT impact to AI (flat vs. 2025), while only 6% qualify as “high performers” (>=5% of EBIT and “significant” impact), also unchanged. Despite convictions and increased agentic AI scaling (40% of >$1B revenue firms vs. 27% last year), 20% say high AI operating costs constrain usage, and expectations for AI-driven workforce cuts rise (39% vs. 32% in 2025). Overall message: productivity gains for individuals are not yet translating into broad, enterprise-level financial returns, suggesting a cautious outlook for near-term profit impact.

Analysis

The market implication is less about an AI demand shock than a widening separation between capex and monetization. Boards will keep funding AI to avoid strategic underinvestment, but delayed EBIT contribution means CFOs will pressure vendors for measurable payback, which is negative for generic workflow SaaS and consulting-heavy implementation models. The second-order effect is in-house substitution: coding agents can erode low-differentiation software purchases before they reduce labor, so ARR gets cannibalized while the customer still carries most of the personnel cost.

Over the next 1-3 months, the key catalyst is earnings guidance, not survey sentiment. If enterprise buyers keep saying AI is strategic but not yet accretive, multiple compression should hit application software harder than infrastructure, because the compute stack can still sell picks-and-shovels even when ROI is delayed. Over 6-18 months, winners are vendors tied to hard-dollar savings—security, code review, cloud optimization, contact-center automation—while the broad "AI-enabled" software basket risks a slow fade as procurement becomes more disciplined.

The contrarian miss is that slow ROI is not necessarily bearish for the whole stack; it can actually extend the spending cycle for semis and hyperscaler cloud while making downstream software buyers more selective. The thesis breaks if large enterprises start reporting real margin lift or headcount reduction from AI in the next two quarters, or if capex guides roll over materially. In that case, the market will stop treating AI as a future promise and start pricing it as a current earnings contributor.

Key Decisions for Investors

• Enter a long SMH / short IGV pair over the next 2 earnings seasons. Rationale: AI spend remains sticky even if monetization lags, but application/software valuation is more exposed to ROI disappointment and in-house feature substitution.
• Reduce exposure to high-multiple SaaS names that sell broad productivity promises rather than measurable savings; use CRM, WDAY, and ADBE as watchlist names into upcoming guidance. Re-enter only if management quantifies AI-driven expansion in net retention or operating margin.
• Buy NVDA or AVGO on post-earnings weakness only if hyperscaler capex commentary remains firm. Risk control: exit if two consecutive large cloud customers guide capex growth lower or if AI-server order comments soften.
• Add to cybersecurity/AI-control beneficiaries only on pullbacks, not strength: PANW and CRWD are better positioned than generic SaaS if enterprises internalize coding but need more code scanning, governance, and breach protection.
• Set an alert for any enterprise management team claiming 5%+ EBIT contribution from AI. If that shows up in a meaningful sample, cover software shorts quickly; that would be the first evidence the ROI curve is inflecting earlier than expected.