The agentic AI gap: Vendors sprint, enterprises crawl

The geopolitical dislocations ripping through the stock market are filtering right down to information technology budgets in the shape of increased uncertainty.

Evidently every quarter of budget optimism is followed with some external event that causes organizations to tighten their belts. Specifically, we’ve seen the increased momentum from January’s chief information officer sentiment survey on spending, pull back from 4.6% growth to three.6%. War, oil prices, the specter of inflation and even the prospect of Fed tightening now loom larger.

Although big tech continues massive capital expenditures – and the real enthusiasm from this month’s Nvidia GTC and RSAC events continues to be being felt – mainstream enterprises are once more expressing caution of their spending intentions. Along with economic and world affairs, artificial intelligence success still eludes most mainstream organizations.

Our remark is that the tech industry is within the third inning of the AI wave, which began in earnest mid last decade with DeepMind and other significant research milestones that led to the ChatGPT and subsequent moments corresponding to Claude Code and OpenClaw. Meanwhile, organizations are still in the primary inning and rightly cautious about deploying AI at scale.

The info suggests that though virtually all firms are leaning into AI, those realizing return on investment at large scale remain the mid- to low teens. Despite leading thinkers corresponding to Nvidia Corp. Chief Executive Jensen Huang advising to not give attention to ROI, and let innovation flourish no matter hard dollar returns, the truth is that within the land of enterprise customers, tangible returns and risk management remain key governors of spending.

On this Breaking Evaluation, we share recent survey data from Enterprise Technology Research that quantifies macro spending and AI adoption within the enterprise. And we put forth a thesis as to why the gap exists between AI enthusiasm and enterprise adoption — and the way the software stack will evolve to make adopting and securing AI simpler. Finally, we draw on the insights from GTC 2026 and what Jensen called “crucial slide” of his keynote. It puts forth a brand new revenue model that potentially unlocks a brand new wave of enterprise value.

Access the complete slide deck

Macro IT spend and IT budget sentiment

The primary chart is the one we keep coming back to since it shows a time series on macro spending sentiment. It’s from ETR’s quarterly drill-down survey of expected IT spending changes, going back to the COVID era, with a big sample size (N = 1,543). The story is considered one of a whipsaw between optimism and caution – and the way quickly sentiment moves when world events get in the way in which.

Coming out of COVID, the info above shows a giant uptick in budget flexibility. IT spending expectations surged into the 7.3% to 7.5% range. Then the air got here out as rates rose and uncertainty took hold. By 2022, expectations compressed and ultimately bottomed at 2.9%, inversely proportional to rates of interest. This was a reminder that when the macro tightens, IT budgets tighten with it.

From there, the chart becomes a map of confidence shocks. Because the Fed began to lower rates, spending expectations improved, however the recovery wasn’t smooth. We saw periodic pops – 4.3%, then as much as 5.3% – followed by pullbacks as recent uncertainties corresponding to Ukraine and tariffs surfaced. Essentially the most recent example is seen exiting December at 4.0%, rising to 4.6% in January, then falling back to three.6% now with the war within the Middle East. We feel that’s a meaningful swing in a brief time period, especially given AI’s overall momentum.

We imagine the fitting technique to interpret the info is IT spending is sensitive to the business climate, and the business climate is being shaped by rates, geopolitics, policy noise and headlines. It’s not all the time possible to prove causation with a single chart, but over many cycles the market data appears consistent – when uncertainty rises, budget confidence softens, and IT spending follows that trend.

AI adoption: Productivity is the goal, decision support is rising

The slide below is a check on how organizations say they’re using AI in the present climate (N = 1,573), and it has been asked consistently since July 2025. The highest answer is what you’d expect – enhancing workforce productivity through automation or task augmentation. That answer has stayed consistently within the low 70% range range and has been durable across multiple quarters. The query we see organizations asking is: How can we make recent breakthroughs beyond early use cases? In other words, firms are seeing early wins but they’re desperate to see them compound.

Essentially the most impressive movement is the regular rise in supporting worker decision-making with AI-driven analytics and insights. That’s a logical next step after productivity since it builds on the work organizations have already done modernizing analytics. When data is organized and accessible, AI can amplify it quickly.

That is where the fashionable data stack players have see real tailwinds – vendors corresponding to Snowflake Inc. and Databricks Inc. are the poster children for consolidating analytics into usable platforms, with Oracle Corp. and others corresponding to IBM Corp. also relevant within the broader market. The info suggests more organizations are actually pushing AI into the “insights” layer, not only the “automation” layer.

Two other points stand out:

• The share saying they usually are not currently leveraging AI in any of those areas has fallen from 10% last July to six% now. That doesn’t mean those firms aren’t using AI in any respect, but it surely does reinforce the larger point that adoption is becoming near-universal, even though it’s uneven and sometimes implicit.
• On labor, the info is consistent with what we’ve been seeing elsewhere – that AI is limiting future headcount growth greater than it’s driving immediate headcount reduction. Firms will often message layoffs as “AI-driven,” but the info suggests the truth is generally headcount avoidance and slower hiring fairly than dramatic cuts directly attributable to AI.

The underside line in the info is AI is within the constructing. Productivity stays the first use case, decision support is gaining momentum, and the job impact is showing up first in hiring plans, not sudden mass reductions. Our take is these are predictable and comparatively straightforward early wins, but they’re not game-changing. Later on this post we posit an emerging recent architecture that may support more dramatic organizational change as AI becomes simpler and safer to adopt.

ROI reality check: Pilots in all places, scale still rare

The chart below gets to the center of the agentic gap – what type of ROI organizations are literally reporting from AI initiatives up to now (N = 1,573). ETR splits the info into two approaches: constructing in-house solutions on the left and buying external vendor solutions on the fitting. In each cases, “no adoption or traction” is declining, but it surely stays meaningful – especially on the in-house side. Embedded AI and vendor-delivered capabilities appear to have an path to adoption, which shows up within the lower “no traction” bars on the right-hand side.

The more telling story is what happens after initial adoption:

On the in-house side, “adoption, ROI not yet realized” sits across the 30% range and has been stuck. That means a meaningful slice of organizations are constructing, experimenting, and learning – but not getting payback yet.

You then hit the dominant category in each side of the chart: ROI in pilots or limited use cases, but not yet at scale. It’s roughly 33% for in-house and 39% for vendor solutions. That’s the clearest indicator that AI is working in pockets, but most enterprises are still struggling to industrialize it.

Finally, the metric everyone cares about – sustained ROI at scale – sits within the low to mid-teens in each cases, about 13%. That’s the headline. Whether organizations construct or buy, only a small minority say they’ve durable ROI at scale.

This supports the broader point we’ve been making in that vendors are moving fast – from RAG-based chatbots to reasoning to agentic workflows – and enterprises are absorbing that shift more slowly. The constraint isn’t enthusiasm for AI or lack of vision. Reasonably, it’s operational readiness – AI governance, safety, security, integrating data, hardening processes and constructing repeatable deployment muscle memory so pilots can convert into production outcomes at scale.

The brand new AI software stack: Closing the ‘agentic gap’

We’ve argued in previous Breaking Evaluation segments that as we moved from on-premises perpetual software models to software as a service, it modified firms’ technical, operational and business models. We further argue that a more profound change is coming with AI that may affect not only IT departments but entire organizations. We’ve written extensively in regards to the infrastructure shift from general-purpose computing to accelerated architectures.

On this section we go further up the worth chain and drill into the emerging AI software stack. Here we specifically project the technical layers we see emerging that may support more rapid AI adoption.

The slide below ties the ROI data to a deeper architectural shift – the enterprise is moving from an app-centric world to an intelligence-centric one. The diagram lays out a four-layer topology and, in our view, it explains what’s missing in today’s software stack and what has to emerge to assist simplify adoption, support recent business models and help organizations which might be stuck in pilots. Organizations are smitten by AI and so they’re funding it. That’s not the constraint. The issue is that the majority enterprises try to bolt agentic workflows onto yesterday’s software stack, while the stack itself is being rearranged.

At the highest sits the Frontier Model – the scarce, capital-intensive layer that produces tokens. It runs on essentially the most advanced hardware, improves rapidly and is increasingly concentrated in a small variety of providers (OpenAI Group PBC, Anthropic PBC, Google LLC and xAI Holdings Corp.). For many enterprises, constructing this layer isn’t a viable objective. The economic reality is that frontier models are enabled by AI factories – and most firms will devour them, not replicate them.

The more underappreciated layer is the Cognitive Surface. We’ve often referred to this layer because the System of Intelligence or SoI. That is where intent gets shaped, context gets assembled, constraints get enforced and outputs get become actions. Additionally it is where the enterprise requirements live – security, policy, compliance, auditability, latency control and integration to existing systems.

That is the layer that turns “a sensible model” into something operable inside a regulated enterprise. Additionally it is the layer that determines switching costs, because policy, semantics and power integration get hardened here. As such, switching vendors will turn out to be much harder, in our view.

We expect this layer to be distributed – but controlled. Large enterprises will want instances closer to their data for latency, sovereignty and regulatory reasons. But they won’t own the evolution of the cognitive surface. They’ll configure it, operate it and integrate it – inside guardrails defined by the frontier model provider. That preserves enterprise control over data and policy while stopping semantic drift.

Below that sits the Transactional Substrate – the systems of record. This layer is important since it stores truth and executes transactions. The change we project is that intelligence migrates upward. The apps and databases don’t disappear, but their role focuses around state, service level agreement guarantees and execution.

Finally, the Edge eventually becomes necessary because sensing and physical execution occur there. Capability at the sting will lag initially, but it surely becomes strategic as agents and automation demand local motion and native autonomy when disconnected. That is where smaller language models will thrive, in our view.

The opposite key point is that the dearth of a mature cognitive surface contributes to the agentic AI gap. We see this within the ROI data. Enterprises are being asked to maneuver to a world where intelligence is produced in AI factories as tokens, accessed through application programming interfaces and governed through a cognitive surface. Until organizations (and SaaS players) construct the control, governance and integration muscle memory in that middle layer, they’ll keep shipping pilots – and struggling to show them into repeatable ROI at scale.

We see this model evolving and the 4 frontier labs shall be fundamental in supporting this recent software stack. We imagine OpenAI, Anthropic and Google will aggressively compete for enterprise traction, while xAI is best positioned for edge workloads, in our view — leveraging Elon Musk’s flywheel of Tesla, Optimus and SpaceX.

The token business model: Throughput, interactivity and ‘where you sit on the curve’

We imagine the deeper shift underway isn’t just architectural – it’s economic. AI is catalyzing a model where intelligence is manufactured in AI factories as tokens, accessed through APIs, and paid for as a first-class line item. On the macro, today firms spend roughly 4% of their revenue on technology. We imagine this figure will rise to 10% or more inside the following decade. Spend will move away from general-purpose computing toward accelerated computing – supported by extreme co-design across central processing units, graphics processing units and networks – with power because the governing constraint.

Because of this Jensen said the slide below from GTC was crucial. The vertical axis is throughput normalized by energy (tokens per second per megawatt). The horizontal axis is interactivity – responsiveness that’s broader than easy latency but latency is the motive force of user experience. In a power-constrained world, moving up the curve on the vertical axis means dollars to operators. This what we wrote about as “Jensen’s Recent Law” in a previous Breaking Evaluation.

Access to the most recent and best systems from Nvidia may be the difference between a stalled AI program and one which scales. Hyperscalers, AI clouds and frontier labs have known this for years. Getting on the Nvidia technology curve is critical for leadership. The annual cadence from Hopper to Blackwell to Rubin is the important thing – massive step-function improvements on a 12-month cycle, not an 18- to 24-month Moore’s Law clock. The 35X improvement called out on the slide below is the type of delta that changes unit economics overnight if you will have a set power budget. The large capex spenders know this and the dynamic will migrate to enterprises as described by Jensen and shown on the horizontal axis.

That’s the opposite a part of the chart where the business model starts evolve. Training monetizes totally on the vertical axis – maximize throughput per megawatt and customers “buy more, save more” or “buy more, make more” depending on whether or not they’re constructing models or selling capability to model builders.

Interactivity is a second monetization opportunity on the X-axis. It creates tiers – free, medium, high, premium, ultra – where users pay more for higher responsiveness, and where essentially the most demanding workloads drive the best willingness to pay. Low-latency inference becomes a priced product and a service delivered through software.

That’s where the Groq integration and Nvidia’s $20 billion Groq investment starts to is smart. Rubin + LPX extends the curve further to the fitting – preserving throughput while improving interactivity. The implication is that the platform that may move the curve right without collapsing the curve down gets to charge to be used cases which might be sensitive to responsiveness, especially in agentic workflows and edge inference. The spectrum goes from freemium (free ChatGPT) to paid ($20/month) to higher-paid tier ($200/month) to coding assistance to super-low-latency agentic– ultra-expensive but value it since it drives revenue.

The takeaway for enterprises is that this isn’t something they will absorb overnight. They’ve to choose the use cases that matter, construct the systems, validate safety and controls, operationalize them, prove ROI, then scale. At the identical time, the price model changes. Token spend becomes a part of cost of products sold – the way in which cloud costs became a part of SaaS COGS – and organizations start managing token budgets as a core operating discipline.

Because of this cautious IT budget sentiment can coexist with AI enthusiasm. Customers don’t wish to over-invest in legacy spend, and so they don’t wish to over-rotate into the brand new spend until they understand where they sit on the curve – and how you can translate throughput and interactivity into unit economics, outcomes and predictable revenue returns.

Going forward this may create recent revenue models and start to interrupt down organizational silos that exist today due to technology constraints. Many departments construct their very own custom tech stack to support their specific mission. Processes are developed and arranged around this tech stack. Data lives of their siloed department and humans then integrate the info via extract/transform/load processes, data pipelines, data science workflows and the like.

Increasingly, we imagine these silos will dissolve to an excellent extent as organizations gain access to intelligence in the shape of tokens to power their agentic enterprises. They are going to construct digital representations of their organizations and the operational model will evolve to support this recent reality.

Jensen said something profound at GTC. Every CEO must understand where they sit on this Pareto curve. Are you monetizing on the vertical axis, the horizontal axis or each? Today a brand new worker gets a laptop and access to systems. In the long run they are going to get a token budget to direct their revenue-producing agents. A software engineer paid $300,000 to $500,000 who spends only $5,000 annually on tokens could be like a chip designer eschewing modern design tools and using graph paper as a substitute. They’d be fired.

That sounds absurd, however the analogy holds for the long run of business. Profit-and-loss managers, sales pros, operational staff, logistics planners, finance pros and others will all be managing armies of agents and burning tokens. Closing the agentic gap requires recent technology, business and operational models that may be executed securely and safely.

That day is coming. Where do you sit on the pareto and how briskly are you able to get there?

Image: theCUBE Research/Gemini

Disclaimer: All statements made regarding firms or securities are strictly beliefs, points of view and opinions held by SiliconANGLE Media, Enterprise Technology Research, other guests on theCUBE and guest writers. Such statements usually are not recommendations by these individuals to purchase, sell or hold any security. The content presented doesn’t constitute investment advice and shouldn’t be used as the premise for any investment decision. You and only you might be chargeable for your investment decisions.

Disclosure: Lots of the firms cited in Breaking Evaluation are sponsors of theCUBE and/or clients of theCUBE Research. None of those firms or other firms have any editorial control over or advanced viewing of what’s published in Breaking Evaluation.