Key Takeaways:
A new UBS report suggests the artificial intelligence gold rush may be slowing to a crawl inside the world’s largest companies, with the gap between ambition and reality widening significantly.
“The key finding is not just that progress is slow, but that firms are systematically over-optimistic about their own pace,” Arend Kapteyn, a UBS economist and the report’s author, said in the note. “This optimism gap is persistent and growing, which has direct implications for valuations in the AI sector.”
The semi-annual survey of 139 IT executives and data engineers found that as of March 2026, only 19% of enterprises had achieved a “scaled production deployment” of AI across multiple business units. This marks a linear, not exponential, increase from the 10% reported two years prior. The disconnect is most apparent when comparing expectations to results: a year ago, 84% of the same executives predicted their firms would reach scaled deployment within 12 months. The actual number that succeeded was just 5%.
For investors who have piled into AI-related software and services companies, the report serves as a crucial reality check. The findings suggest that the path from a working AI model to broad, revenue-generating commercial deployment is fraught with more friction than the market currently appreciates. This could challenge the high-multiple valuations of companies like Salesforce and ServiceNow, which are predicated on rapid enterprise AI integration.
The core of the UBS report is the stark contrast between corporate forecasts and on-the-ground execution. The survey data, collected from AI decision-makers across 26 industries, shows a consistent pattern of over-promising and under-delivering.
This isn't a one-time miscalculation. According to UBS, this "optimism bias" has appeared in every survey round, with the chasm between what executives expect and what their organizations achieve continuing to expand. While AI technology itself, measured by benchmarks from the Stanford AI Index, is advancing at a non-linear rate, enterprise adoption is following a far more modest, linear trajectory. The two-year, nine-percentage-point gain in scaled deployment averages out to just under 3% of firms graduating to the next stage every six months.
The survey identified six primary obstacles preventing companies from scaling their AI initiatives, with return on investment and complexity topping the list.
A majority of firms, 53%, cited an unclear return on investment (ROI) as a major barrier. This was followed by compliance and regulatory issues (48%) and a lack of qualified talent (42%). Notably, the challenge of “integration complexity” saw a significant jump, rising to 45% from a range of 37-38% in the two previous surveys. This indicates that as companies move from pilot projects to enterprise-wide rollouts, they are discovering that embedding AI into existing workflows and legacy systems is substantially more difficult than anticipated.
This software and integration bottleneck is creating opportunities for a new class of companies. Firms like SPARC AI (OTC: SPAIF) are developing software-only platforms designed to give existing hardware, such as drones, advanced capabilities like GPS-denied navigation without requiring costly new sensors. This approach directly targets the integration and cost issues highlighted by the UBS report.
For investors, the UBS data suggests a need to recalibrate growth forecasts for the AI software sector. The market has largely priced stocks for frictionless adoption, but the reality appears to be a multi-year slog defined by system integration challenges, talent shortages, and difficult ROI calculations. While the AI hardware buildout continues, with companies like Netweb Technologies (NSE: NETWEB) seeing a 460% surge in their AI infrastructure segment, the software and services layer that runs on top of that hardware faces a more uncertain timeline. Netweb currently trades at a demanding 122 times earnings, a valuation that reflects the market's confidence in the hardware layer. The UBS report questions whether that confidence has been misapplied to the software layer. The report implies that the real value may accrue not to the most advanced models, but to the companies that solve the mundane, complex, and expensive problems of making AI actually work at scale.
This article is for informational purposes only and does not constitute investment advice.
