A recent SUSE survey presented at the Bitport CIO conference finds that enterprises are deploying AI rapidly, but investments in the supporting infrastructure lag behind. Among respondents, 34.1% reported running some form of AI as a pilot, and the same share said they already use AI in production.
By contrast, only 20.5% identified AI/ML infrastructure preparation as a modernization priority, suggesting that many organizations are launching pilots and specific use cases before establishing a long-term infrastructure strategy.
Why robust infrastructure matters
SUSE emphasizes that broad, enterprise-level AI deployment requires a stable foundation that supports secure operation, flexible scaling and cost transparency. Naveen Chhabra, a lead analyst at Forrester, noted that infrastructure leaders now face more complex questions: can their infrastructure reliably move AI from pilot to production without runaway costs, operational burdens or risks?
International data echo this gap: the Stanford HAI 2026 AI Index Report shows that in 2025, 88% of organizations used AI in at least one business area, but only 3–10% reported full enterprise-scale AI adoption depending on organization size.
Cloud or on-premises GPUs?
Running AI requires substantial compute, typically GPUs. In the SUSE survey, 29.5% of decision-makers said they plan to rely on on-premises GPU capacity, the same proportion (29.5%) selected public cloud GPUs, and 27.3% had not yet decided which model to choose.
Among organizations already operating AI in production, the split is clearer: 46.7% use public cloud GPUs and 33% run on-premises solutions.
The larger cloud share for early production projects likely reflects the ease and elasticity of cloud compute for initial deployments. At the same time, the sizable on-premises presence indicates that direct control over data, local data handling and predictable long-term operations remain priorities for many companies.
Running AI on-premises: control and cost visibility
On-premises deployment can be especially important where organizations handle sensitive business information, customer data or significant internal knowledge assets. To address these needs, SUSE offers the SUSE AI Factory, an open infrastructure platform for private, hybrid and air-gapped enterprise AI deployments.
SUSE AI Factory enables organizations to run applications and models on their own infrastructure, providing tighter control over data, the technology stack and available resources. The platform includes monitoring for application token usage and associated costs, as well as GPU performance and utilization metrics, which can improve visibility into actual capacity needs and help control spending.
An open approach also gives companies flexibility in choosing models and technology providers—a key advantage in a rapidly evolving AI market.
Conclusion
The SUSE survey highlights a growing discrepancy between the pace of AI adoption and the preparedness of enterprise infrastructure to support it at scale. Decisions about cloud versus on-premises GPUs and transparent resource management remain central to moving AI reliably from pilots into production and sustaining it over time.
