Manufacturers remain cautious in how they describe their progress with artificial intelligence, yet the underlying data suggests a more deliberate and arguably more durable approach is taking shape. According to a new benchmark study from S&P Global Market Intelligence, conducted in collaboration with Vultr, the sector reports the lowest share of “transformational” AI adoption among surveyed industries, but also the lowest friction in getting there.
That combination is reshaping how manufacturers deploy AI. Rather than expanding the number of models in production, companies are consolidating portfolios, reworking platform architecture, and focusing on governance structures that support scale, reliability, and integration with core operations.
Lower Maturity Labels, Fewer Structural Barriers
The benchmark categorizes AI maturity into three stages: Operational, Accelerated, and Transformational. Only 31% of manufacturing respondents classify themselves as transformational, compared with higher shares in other sectors. About 25% remain in the early Operational stage, versus 19% across the full respondent base.
What stands out is not the maturity ranking, but the relative ease manufacturers report in overcoming the obstacles typically associated with AI adoption. Across measures including skills availability, data quality, security concerns, and cultural alignment, manufacturers consistently report lower barrier severity than peers in other industries.
Skills shortages, often cited as the primary bottleneck to AI progress, are reported by 46% of manufacturers, compared with 62% across all respondents. That gap suggests the sector’s long-standing investments in engineering talent, process discipline, and systems integration may be paying dividends as AI moves closer to production environments.
The data points to a sector that may be underplaying its progress while quietly laying the groundwork for broader deployment.
Rebuilding Platforms and Reducing Model Sprawl
One of the clearest signals of that groundwork is a shift in how manufacturers structure their AI platforms. Thirty-five percent have already built or are in the process of building internal platform-as-a-service (PaaS) infrastructure, a figure expected to rise to 45% within two years. Over the same period, reliance on hyperscaler-managed PaaS is projected to decline from 56% to 42%.
This is not a retreat from the public cloud. Manufacturers still run roughly 30% of training workloads and 28% of inference workloads on major hyperscalers, both well above industry averages. Instead, the move reflects a redesign of how AI is governed, orchestrated, and standardized across plants, business units, and geographies.
That architectural shift is accompanied by a notable contraction in model counts. The average manufacturer currently runs 242 AI models in production, a figure projected to fall to 189 next year. At a time when AI investment narratives often equate progress with proliferation, manufacturers are deliberately moving in the opposite direction.
Recent industry data shows that maintaining large numbers of bespoke models drives up validation costs, complicates cybersecurity oversight, and slows integration with operational systems such as MES, ERP, and quality platforms. Fewer models, if better governed and more widely deployed, can reduce technical debt while improving consistency across operations.
When Restraint Becomes an Operating Discipline
What deserves closer attention is how this consolidation reshapes decision-making speed and accountability. With fewer models in play, performance trade-offs become more visible, ownership becomes clearer, and AI outcomes are easier to tie to production, quality, or cost metrics that already govern plant operations. That discipline matters as regulatory scrutiny around industrial data use, cybersecurity, and model explainability increases. In that environment, the ability to defend why a specific model exists, and what operational risk it mitigates or return it delivers, may prove more valuable than the ability to deploy new ones quickly.
