Enterprise AI is splitting into two economies: leaders redesigning operations and spenders chasing ROI that never arrives

Most enterprise AI programs are spending more and delivering less. The organizations actually capturing value are not running more pilots or licensing more models. They are redesigning how their businesses operate. That distinction, between tool-level deployment and organizational redesign, is now the clearest dividing line in enterprise technology, according to Forbes contributor Brian Solis writing on July 27.

The ROI gap is an architecture gap

Solis argues that AI leaders are treating the technology as a reason to rethink operating models, not just automate existing tasks. The result is compounding advantage: better data, faster decisions, and workflows that improve as the AI learns. Spenders, by contrast, layer tools onto unchanged processes and measure ROI in productivity percentages that never quite materialize at the organizational level.

Forbes contributor Vivian Toh, writing earlier this month, framed a related dynamic around what she calls the "enterprise AI reckoning." Enterprises that relied heavily on third-party frontier models are now reassessing. Data ownership concerns and platform dependency risk are pushing a cohort of larger organizations toward what Toh describes as "sovereign AI": owning the infrastructure, fine-tuned models, and proprietary knowledge bases that constitute their intelligence stack. The driver is not ideology but operational control.

Forbes contributor Sarah Elk has described the destination these leaders are building toward: a "learning system" in which agentic AI compounds its own advantage over time. Rather than deploying individual agents for discrete tasks, forward-looking IT and operations leaders are architecting systems that self-improve, turning each automated decision into training signal for the next one.

Agent gateways: the control layer that security and IT teams now need

As agentic AI moves from pilots into production, a new governance problem has surfaced: who controls what agents can access, call, and act on? Forbes contributor Janakiram MSV identified agent gateways as an emerging product category built to answer that question. Writing in early July, MSV described the agent gateway as a control plane that sits between AI agents and the models, APIs, and enterprise tools they interact with, providing centralized auditing, access management, and security policy enforcement.

For CIOs and IT operations leaders, this category matters now. Enterprises that have deployed more than a handful of agents are finding that without a centralized control plane, auditability collapses and security posture degrades. The agent gateway is not a nice-to-have for mature deployments. It is a prerequisite for scaling agentic AI responsibly.

The governance imperative runs parallel to the sovereignty trend. If an enterprise is moving to own its AI stack, it also needs to govern every interaction that stack has with the outside world. Agent gateways are the mechanism for doing that at scale.

ERP as an execution engine, not just a ledger

The role of ERP in this shift is harder to see but operationally significant. Forbes contributor Robert Kramer, also writing on July 27, makes the case that modern ERP has crossed a threshold: it is no longer just a system of record that captures what happened, but an execution layer that initiates what happens next. AI embedded in ERP connects transaction data, operational workflows, and external signals in ways that allow the system to take autonomous action, routing a purchase order, triggering a replenishment, or escalating a compliance flag, without waiting for a human prompt.

For procurement directors and supply chain leaders, this reframes how they should evaluate ERP vendors in 2026. The question is no longer whether a platform supports AI add-ons. It is whether the platform's data model and workflow engine are designed from the ground up to support autonomous decision loops. That capability gap between legacy ERP and modern AI-native platforms will widen over the next 18 to 24 months as agentic capabilities mature.

Nvidia's role in the infrastructure that makes all of this possible

None of the above works without infrastructure, and on July 27 CNBC reported on Nvidia's deepening position as what correspondent Kristina Partsinevelos described as the financial backstop of the AI buildout. A separate CNBC report the same day noted that Nvidia is set to back OpenAI's data center expansion. The practical implication for enterprise operators is straightforward: Nvidia's hardware and financing structures are now embedded in the supply chain for AI capacity in ways that affect procurement timelines and vendor leverage across the industry.

Enterprises building sovereign AI stacks, deploying agent gateways, or modernizing ERP execution layers all depend on GPU availability and data center capacity that flows, in large part, through Nvidia. Procurement and IT teams evaluating multi-year AI infrastructure commitments should factor in that dependency when assessing vendor risk and contract terms.

The operational picture heading into the second half of 2026 is one of structural divergence. The technology itself is not the bottleneck. The bottleneck is whether an enterprise's data architecture, governance model, and operating structure are built to let AI act, not just advise. The organizations that solve that problem first will be the ones setting benchmarks that everyone else measures against.