Reading NVIDIA’s 2026 Retail AI Survey: The Numbers That Matter and the Ones That Need Context

Seventy-nine percent of retail and CPG respondents to NVIDIA’s third annual State of AI survey said open-source models and software are moderately to extremely important to their AI strategy. That figure, published in January alongside headline adoption numbers that have drawn far more coverage, is the most consequential finding in the report for operators and brand teams trying to understand where AI deployment in this industry is actually headed.

The survey gathered hundreds of responses from retail and CPG professionals between August and December 2025, covering AI adoption rates, budget intentions, business impact, and emerging application areas including agentic and physical AI. NVIDIA is a semiconductor and platform company with direct commercial interest in accelerated AI adoption across enterprise verticals. That does not make the survey’s findings false. It means the numbers deserve the same reading any rigorous practitioner would apply to research underwritten by a company whose customers are the survey’s subjects. The survey is useful. It is not neutral, and its headline figures are not straightforwardly comparable to independent market research. Genuine signal exists in the data for retail and CPG leaders making AI investment decisions right now, but locating it requires reading past the top-line percentages.

The Open-Source Finding

The context for the 79% figure comes directly from the survey report, where Jason Goldberg, chief commerce strategy officer of Publicis Groupe, described the pattern the industry went through to arrive at this position. Retailers, Goldberg said, started with proprietary AI vendors and found themselves without control over their own data or the ability to adapt models to their specific needs. His framing of what open source changes — the ability to apply proprietary data, avoid vendor dependency, and draw on community-driven model improvements — explains why 79% is not a cost story but an architecture and data ownership story.

That distinction matters for how organizations read their own position. A company that built its AI stack on proprietary platforms two or three years ago and has not revisited that decision is likely operating under the constraints Goldberg described: models it cannot fine-tune on its own data, deployment dependent on a vendor’s terms, and integration paths that may not align with where its retail partners are now building. A company that made the move to open infrastructure, or is making it now, has more flexibility on each of those dimensions. The 79% figure represents how far the industry’s center of gravity has shifted on this question.

For CPG brands working within retailer AI ecosystems, the open-source shift has a secondary effect worth tracking. When a retail partner moves its demand forecasting, assortment optimization, or personalization infrastructure to open and interoperable frameworks, the integration requirements for supplier data can change more frequently than they would under a locked commercial system. Maintaining clean, structured, and well-attributed product and sales data is not a new requirement, but the bar for how consistently it needs to be met gets higher as retailer AI systems become more capable and more actively applied.

Who Responded, and Why That Shapes Every Number

NVIDIA disclosed its survey methodology in the companion cross-industry report published the same week. Respondents were sourced from NVIDIA’s own distribution lists and, for China and Japan, through a third-party agency. The survey reached 3,200 respondents across five industries; the retail and CPG segment represented a subset of that total, with NVIDIA describing the response count as “hundreds.”

Respondents spanned three seniority bands: C-suite and VPs at 27%, directors and managers at 33%, and AI practitioners at 40%. Company size ranged from large enterprises with more than 1,000 employees at 39%, mid-sized organizations at 27%, and smaller companies with fewer than 100 employees at 34%. Geographic distribution ran 32% APAC, 26% North America, 21% EMEA, and 20% elsewhere.

Two features of that sample warrant attention before reading the results. First, sourcing respondents from NVIDIA’s own distribution lists means the sample skews toward organizations already engaged with NVIDIA’s ecosystem, companies that are by definition further along in AI evaluation than the broader retail and CPG market. Second, AI practitioners made up the largest single respondent group. People whose professional identity is tied to AI adoption are not a representative cross-section of retail and CPG decision-makers. Both factors push the adoption and sentiment figures higher than a randomly sampled industry population would likely produce.

None of this makes the survey unreliable. It makes it a measure of AI adoption among companies already in the conversation, not a census of the industry at large. For practitioners, that framing is actually more useful: the survey describes what the more engaged, AI-active segment of the market is doing, which is a reasonable proxy for where broader adoption is heading.

The Revenue and Cost Numbers

The 89% revenue increase and 95% cost reduction figures are the ones getting the most coverage, and they are softer than they appear. Both are self-reported assessments of directional impact, not measured outcomes against a control group. The survey asked respondents whether AI has helped increase revenue or decrease costs, a yes/no sentiment question, not a request for documented financial results.

That matters for how the figures should be used. A respondent who deployed an AI-assisted email campaign and saw engagement improve is technically reporting that AI increased revenue. So is a respondent who implemented demand forecasting and reduced overstock by a measurable percentage. The survey treats both the same. What the 89% figure actually captures is that the vast majority of organizations in this sample believe AI is contributing positively to their revenue performance, which is meaningful as a sentiment indicator but not a reported financial result.

The more granular data is where the survey produces genuine insight. Among the 89% who reported revenue increases, 30% said the increase exceeded 10%. Among the 95% reporting cost decreases, 37% said costs had fallen by more than 10%. Those figures suggest a meaningful cohort of respondents has moved past marginal improvement into material financial impact, even if the precise measurement basis is unknown. The 54% who cited improved employee productivity and 52% who cited operational efficiencies as AI’s primary business contributions provide better texture: these are concrete operational outcomes that practitioners can evaluate against their own programs.

Agentic AI: What the 47% Figure Actually Describes

Nearly half of survey respondents, 47%, said their organizations are either using or evaluating agentic AI. The breakdown matters: 20% said agents are already active, 21% said deployment is coming within the year, and the balance are still in assessment. That distribution means the “using or evaluating” figure combines active production deployments with pre-deployment interest, which covers a wide range of organizational readiness.

The survey’s stated goals for agentic AI are more instructive than the adoption rate for understanding where actual deployment is happening. Process speed and efficiency led at 57% of respondents, followed by enhanced customer personalization and improved real-time decision-making, each at 40%. The ordering is worth noting. The primary value being sought from agents in retail and CPG is operational throughput, not a better customer-facing interface. Inventory rebalancing, dynamic pricing, demand signal processing, vendor coordination: these are the applications where measurable ROI justifies the deployment complexity, and they are also the ones that run in the background, invisible to the customer but consequential for the P&L.

Chris Walton, co-CEO of Omni Talk, is quoted in the NVIDIA report advising executives to prioritize use cases that solve specific P&L problems and prove value before scaling. The survey data supports the logic behind that sequencing: the applications producing documented results in this sample are operational, not experiential, and the 57% who named process speed and efficiency as their primary agentic goal are further along in deployment than the 40% focused on customer personalization.

One concrete operational implication the survey raises but does not fully develop concerns product data quality. Agentic systems operate on product catalogs and inventory feeds in real time, and an AI agent handling dynamic pricing or automated reordering is only as accurate as the data it reads. At NRF in January, NVIDIA announced an open-source Retail Catalog Enrichment blueprint, describing the problem it addresses as “sparse data,” meaning basic product images and limited attribute information that produce unreliable outputs when AI systems act on them. The announcement is an implicit acknowledgment that catalog data quality is a widespread operational gap, and for CPG brands it represents a near-term operational exposure as retailer agent deployments expand.

The Supply Chain Findings

Sixty-four percent of survey respondents said supply chain challenges increased year over year. The pressures they named, geopolitical instability, labor constraints, demand volatility, and regulatory complexity, are persistent conditions, not a 2025-specific development. What the survey adds is the industry’s stated response: 51% named supply chain operational efficiency and throughput as their primary AI use case, making it the leading application area ahead of customer experience and traceability.

For CPG companies operating within retail supply chains, the relevant question that figure raises is how AI-driven operational changes at the retailer or major CPG level are affecting the planning assumptions and lead times that govern supplier relationships. The survey documents the direction of those changes. The pace at which they affect day-to-day supplier coordination is what practitioners need to be tracking.

What the Budget Intentions Reflect

Nine in ten survey respondents said they plan to increase AI budgets in 2026, with half reporting planned increases of 10% or more. North American respondents were more aggressive still, with 48% planning increases of 10% or more, according to the cross-industry data.

Budget intention figures in vendor surveys are among the least reliable metrics for gauging actual spending. Respondents reporting optimistic intentions to a technology company’s research team is not the same as approved capital allocation. What the figures do indicate, taken alongside the adoption data, is that organizations already in the survey sample are not pulling back. Active users are scaling up, and the share still in assessment mode is declining.

The more useful signal from the survey is the distribution of outcomes among those already deployed. Thirty percent of the revenue-impact group reported gains exceeding 10%. Thirty-seven percent of the cost-impact group reported reductions exceeding 10%. For executives using the survey to calibrate their own investment timing, those figures are more instructive than the budget intention data: they describe what a substantial portion of this sample has already produced, in conditions recent enough to be relevant to decisions being made now.

Conversations On Retail

Conversations On Retail is a gathering place and resource center for retail and CPG executives, built to make it easier to stay current, discover the technologies and solutions shaping the industry, and connect with the people driving it forward.

We publish news, views, and reviews from staff editors, contributing experts, and trusted partners. Some articles are developed internally, while others are submitted by industry contributors or adapted from interviews and recorded conversations with industry leaders.