AI is delivering 180 basis points of margin growth for S&P 500 companies that quantified it

The ROI debate is over for the companies that did the math. Twenty-five S&P 500 firms that explicitly quantified AI's contribution during Q2 2026 earnings reported an average margin improvement of 180 basis points, according to an analysis by 22V Research cited by Bloomberg. Strip out companies that bundled AI savings with broader productivity programs and the average still holds at 150 basis points. With roughly 90% of S&P 500 companies having reported for the quarter, that data point is the clearest enterprise-level proof yet that AI deployments are reaching the income statement.

From any industry, not just tech

Bloomberg's Geoffrey Morgan noted that the margin gains are appearing across sectors, not just in software or semiconductors. The detail in the headline, a waste-management company reporting a meaningful margin boost tied to AI-driven route optimization and operational efficiency, illustrates just how broadly the technology has penetrated enterprise workflows. For operations leaders outside the technology vertical, that cross-industry pattern matters: the question is no longer whether AI generates returns in their sector, but whether their current deployment is configured to capture them.

Fortune reported on August 8 that AI is changing the nature of work faster than official data can track, a dynamic that complicates internal benchmarking efforts for companies still relying on lagging productivity metrics. Operations teams that measure AI impact only through traditional KPIs risk undercounting gains that show up in throughput, error rates, or cycle time rather than headcount ratios.

The model-mixing shift rewriting procurement

Even as margin gains accumulate, a parallel cost correction is underway. The Wall Street Journal reported in late July that companies across industries have grown frustrated with the expense of running every workload through flagship models from OpenAI and Anthropic. The response is a deliberate multi-model strategy: routing simpler, high-volume tasks to cheaper alternatives, including open-source models and lower-priced offerings built in China, while reserving premium models for work that genuinely demands their capability.

Cursor, a software development platform whose product is compatible with multiple AI model providers, has become a visible beneficiary of this shift. A company representative told the Journal that the most powerful and expensive AI models are simply not necessary for relatively routine tasks. That framing is increasingly the operating logic at procurement and IT departments negotiating AI contracts: match model cost to task complexity rather than standardize on one vendor.

The Wall Street Journal described the dynamic as an "a la carte" approach to AI procurement. For CIOs and procurement directors managing AI spend, the practical implication is a vendor landscape that now rewards flexibility. Platforms and middleware layers that can route workloads across multiple model providers will command more evaluation time, and single-vendor commitments made in 2024 or 2025 are worth revisiting against current pricing.

Majority of businesses now in deployment, not evaluation

The backdrop to both trends is a market that has crossed a critical adoption threshold. A Tech.co report, cited in Bloomberg's coverage, found that 58% of business leaders have adopted AI tools to stay competitive. That figure suggests the majority of organizations are past the pilot stage and operating AI in production environments, which shifts the conversation from "should we invest" to "how do we optimize what we have deployed."

For enterprise operators, the combination of verified margin data and a maturing multi-model procurement market creates a specific set of decisions. Teams that deployed AI under the assumption of a single-provider architecture built on the highest-capability models available now have both the financial incentive and the vendor ecosystem to redesign that stack. The 180-basis-point average is a benchmark to measure against; the model-mixing trend is the method for getting there more efficiently.

What this means for your team

• Audit your current AI model contracts against actual task complexity: identify high-volume, lower-complexity workloads that could be routed to lower-cost models without degrading output quality.
• Use the 150-180 basis point margin benchmark from 22V Research as an internal target when building the business case for expanded AI deployment or renegotiating existing vendor agreements.
• Evaluate middleware and orchestration platforms that support multi-model routing, since the ability to shift workloads between providers is now a core procurement requirement rather than a nice-to-have.
• Revisit productivity measurement frameworks: if your team is relying solely on headcount or traditional output metrics, you may be missing margin gains already generated by AI deployments in throughput, error reduction, or cycle time.