What 148 Ai4 Attendees Told Us About AI Priorities, Compute and Infrastructure
At Ai4 2026, we surveyed 148 attendees from 131 companies to get a clearer picture of where organizations stand with AI today.
Rather than focusing on AI adoption in the abstract, we wanted to understand what teams are actually working toward, what is slowing their next project down, and how infrastructure fits into those decisions.
The responses do not represent the entire market, nor do we think 148 responses from a single conference should be treated that way. But they do provide a useful snapshot of what practitioners working across AI strategy, development, infrastructure, procurement and day-to-day use are experiencing right now.
And one theme stood out: organizations are increasingly focused on putting AI into production, but the path to getting there is constrained by budget, security, access to compute and the infrastructure decisions underneath it.
AI teams are focused on moving beyond experimentation
The most commonly selected AI priority among respondents was moving AI projects into production, selected by 56%.
What is your organization’s biggest AI priority right now?
| Priority | Responses | Percent |
|---|---|---|
| Moving AI projects into production | 83 | 56% |
| Building internal AI skills | 66 | 45% |
| Exploring potential use cases | 57 | 39% |
| Managing infrastructure costs | 39 | 26% |
| Expanding compute capacity | 31 | 21% |
| Not sure | 14 | 9% |
Respondents could select more than one answer, so percentages may total more than 100%.
The results suggest that experimentation is still very much part of the AI landscape, but a significant share of these organizations are now thinking about what it takes to operationalize those projects.
That shift matters because moving an AI workload from an experiment into production introduces a different set of requirements. Compute capacity, infrastructure cost, security, internal approvals and access to the right hardware can all become more consequential as workloads scale.
Budget leads the list of AI project constraints
When respondents were asked what was most likely to limit their next AI project, budget was the most common response at 40%.
But it was far from the only constraint respondents identified.
What is most likely to limit your next AI project?
| Constraint | Responses | Percent |
|---|---|---|
| Budget | 59 | 40% |
| Privacy/Security Concerns | 46 | 31% |
| Access to compute | 39 | 26% |
| Internal Approval | 36 | 24% |
| Time | 29 | 20% |
| Talent | 19 | 13% |
| No Limitations | 18 | 12% |
Respondents could select more than one answer, so percentages may total more than 100%.
Budget clearly led the results, followed by privacy and security concerns at 31% and access to compute at 26%.
The ability to select multiple constraints also reflects how AI projects tend to operate in practice. An organization may be working through budget approval while simultaneously evaluating security requirements, infrastructure availability and internal resources.
Looking at the overall results, budget comes first.
But looking more closely at the people closest to infrastructure tells a slightly different story.
Compute becomes a bigger issue closer to the infrastructure
When we segmented the responses based on each respondent’s role in AI initiatives, one group stood out. Among respondents managing infrastructure and operations, access to compute rose to 41% and became the leading project constraint, ahead of budget.
That group was also particularly focused on getting AI into production, with 69% identifying production as a priority.
26% → 41%
Access to compute was selected by 26% of respondents overall, but rose to 41% among those managing AI infrastructure and operations.
That difference is worth paying attention to.
It does not mean budget stops mattering. Instead, it suggests that the challenges surrounding AI infrastructure can look different depending on how closely someone is involved with sourcing, managing and deploying the underlying compute.
Across the broader survey, access to compute was selected by 26% of respondents. Among those working directly with infrastructure, however, it moved to the top.
That does not prove that compute access is a larger industry-wide constraint than budget. The sample is too limited to make that conclusion.
But it does raise an important question for organizations moving AI projects closer to production: once the budget is approved, is the required compute actually available?
Organizations may need the right hardware in a specific configuration and quantity, within a timeframe that aligns with the project. Funding the infrastructure and obtaining the infrastructure are related challenges, but they are not necessarily the same challenge.
AI infrastructure is not moving in one direction
Among that same segmented group involved with infrastructure and procurement, the survey also showed that there is no single infrastructure model emerging as the default.
What best describes your organization’s current compute environment?
| Infrastructure model | Responses | Percent |
|---|---|---|
| Cloud | 16 | 42% |
| Hybrid | 11 | 29% |
| On-Prem | 9 | 24% |
| Still evaluating | 2 | 5% |
Those results reinforce a reality that often gets lost in conversations about cloud versus on-premises infrastructure. For many organizations, the answer is likely to depend on the workload and future initiatives.
Cloud infrastructure can provide flexibility and speed without requiring a large hardware investment upfront. Owned infrastructure can make sense for steady workloads, greater control or situations where ongoing cloud costs become difficult to justify.
Hybrid environments give organizations another option, allowing them to place workloads where they make the most operational and financial sense.
Rather than showing a wholesale movement toward one model, the responses point toward organizations weighing several approaches as their AI requirements become clearer.
AI use cases also vary considerably
Internal productivity and automation was the most commonly reported AI use case, selected by 61%.
How is your organization currently using AI?
| AI use case | Responses | Percent |
|---|---|---|
| Internal productivity and automation | 67 | 61% |
| Research and development | 44 | 40% |
| Data analysis and decision-making | 42 | 38% |
| Customer-facing products or services | 36 | 33% |
| We are still exploring use cases | 36 | 33% |
Respondents could select more than one answer, so percentages may total more than 100%.
The responses to this question came from a segmented group of 110 participants involved in building AI solutions, leading AI strategy, or using AI tools in their day-to-day work.
The differences underneath those numbers are particularly interesting.
People building AI solutions were more likely to identify research and development as their leading use case, while those focused on AI strategy or using AI tools in their day-to-day work were more concentrated around internal productivity.
That helps illustrate how differently AI can look across the same organization.
For one team, AI may mean developing and training models. For another, it may mean improving employee productivity. For the infrastructure team, the immediate concern may be whether the organization has enough compute to support any of it.
Eventually, those priorities intersect.
The infrastructure lifecycle deserves more attention
Another area we explored was what happens to AI infrastructure when organizations are finished using it.
Responses showed a wide range of approaches to managing retired hardware. Some organizations rely on internal teams, while others work with IT Asset Disposition partners or recyclers. Others are still evaluating their process or may not have full visibility into what happens to equipment after it is retired.
The survey does not tell us how effectively any of those approaches are recovering value.
What the responses do highlight is that the end of the hardware lifecycle deserves a place in the broader AI infrastructure conversation.
GPUs, servers, networking equipment and other enterprise technology can retain residual value after an organization is finished using them. A defined process for evaluating, securely decommissioning and remarketing retired equipment can potentially return capital that can be applied toward future infrastructure investments.
As organizations invest more heavily in AI infrastructure, planning for what happens to that equipment can become part of the investment strategy from the beginning.
That creates a connection between acquiring the compute an organization needs today and making the most of that investment when those assets are eventually retired.
What these responses tell us
This survey was not designed to forecast the AI market.
The survey reflects responses from 148 Ai4 attendees who participated as part of a giveaway. The results offer a useful snapshot of those respondents.
Even with that limitation, several patterns are worth watching.
Organizations are increasingly focused on moving AI into production. Budget remains a major constraint. Security, internal approval and time continue to create friction.
And among respondents closest to the infrastructure supporting these initiatives, access to compute became even more prominent.
As AI projects move beyond experimentation, infrastructure decisions are likely to become harder to separate from the broader AI strategy.
Organizations will have to determine where workloads should run, how much capacity they need, how quickly they can obtain it, and how to make the most of the infrastructure investments they already have.
Securing budget may be one part of that equation.
Making sure the compute is there when the organization is ready to use it is another.
