Key Takeaways
- The Intelligent Economy Is Being Enclosed. Over the past two quarters, ServiceNow, Workday, SAP, and their peers have moved, one by one, to meter or gate the access of outside AI agents. In early May the pattern became unmistakable, and the trade press converged on a single word: tollgate. In our analysis, enclosure adds a second sorting variable to the bifurcation of the AI Premium, the widening divide between structurally advanced companies and the rest. Structural readiness determines who can generate the premium; terrain ownership increasingly determines who retains it. Durable value is migrating to those who control the terrain on which AI agents must operate; everyone else pays the toll.
- The Bifurcation of the AI Premium Holds, and It Is Hardening. Nearly half of technology deals carry an AI rationale, and 70% of private equity buyers in one survey backed out of at least one active deal over AI exposure concerns. The question a quarter ago was largely who deploys AI well. The sharper question now is who controls the data, the workflow, and the access chokepoints that agents must cross.
- The Judgment Premium Is Rising. As AI distributes analytical capability broadly, the competitive premium on those tools compresses. What does not compress is the premium on judgment. AI may provide better inputs. Judgment can provide better decisions.
Introduction
The global economy has entered a structural era shaped by the convergence of artificial intelligence, data infrastructure, and automation. At Viorum, we call this the Intelligent Economy. It is reshaping how value is created, defended, and ultimately transacted, and our focus is on where the shift is most acute: technology, software, and technology-enabled services. Artificial intelligence has moved from a question of adoption to a question of architecture, and architecture is now becoming a question of access. A front that had been building since late 2025 hardened this spring: metered access and permissioning layers that external AI agents must pass through.
The data points in a consistent direction. More than half of companies report zero payoff from AI investment, while a narrow cohort classified as structurally advanced has historically reported 1.7x higher revenue growth and 3.6x stronger shareholder returns relative to peers.1, 2 Leading institutions now describe the next phase as agentic AI, where autonomous systems are restructuring workflows and, increasingly, decision-making itself.3, 4
This edition examines the enclosure of the Intelligent Economy: the emergence of access, permissioning, and pricing controls around the platforms AI agents must traverse. The bifurcation of the AI Premium, a thesis we introduced in our Spring 2026 edition, remains the foundation; enclosure adds the axis that determines who retains that premium and who pays a portion of it away in tolls.
In our advisory work, we observe three enterprise archetypes in the Intelligent Economy. The AI-Architected Enterprise has deliberately restructured around AI, accepting integration and execution risk in exchange for structural foundations beyond surface-level adoption. The AI-Native Innovator was designed from inception with AI as its central operating thesis, unburdened by legacy technical debt but exposed to market timing risk, capital intensity, and unproven unit economics. The AI-Constrained Enterprise is defined by siloed data, compounding technical debt, and surface-level deployment, and faces a widening valuation gap with a narrowing window to reposition. Pursuing AI-Architected status carries its own material risks, and not every company’s context calls for the same pace or depth of integration.
In our advisory experience, AI-Architected and AI-Native firms tend to align with the premium side of valuations while AI-Constrained firms face headwinds, and the factors that determine placement are increasingly structural rather than discretionary. One posture sits outside these three archetypes: the established platform incumbent that is neither AI-Constrained nor AI-Native but monetizes enclosure, resetting the access economics within which every other archetype operates.
The data cited throughout this Perspective is drawn from third-party research sources identified in the Endnotes. These figures reflect the methodologies and conditions of their respective studies and should not be treated as projections applicable to any individual company.
The Valuation Shift: The Bifurcation of the AI Premium
If the factors that separate premium from compressed are structural, the evidence should be visible in performance data, capital flows, and transaction behavior. The market data suggests that it is. Organizational AI adoption reached 88% in 2025, up from 78% the prior year, and private AI investment in the U.S. surged to $285.9 billion, part of $581.7 billion in total global corporate AI investment.5 But adoption data can mislead. A narrow cohort, roughly 5% classified as structurally advanced, has reported 1.7x higher revenue growth and 3.6x stronger shareholder returns relative to peers,2 while 60% of companies remain stagnant in AI maturity, more than half report zero payoff, and only 12% of CEOs say AI has delivered both cost and revenue benefits.1, 2 CEO confidence in revenue growth has fallen from 56% in 2022 to 30% in 2026, a five-year low.1Deployment does not equal transformation, and this distinction is where the bifurcation begins.
Where Returns Are Concentrating
The implication is not that AI is overhyped. Returns are concentrating. AI is becoming a foundational input, much as cloud computing once did, where access is increasingly universal. Advantage accrues to those who embed it at the core of the operating model, a Harvard Business Review identified as early as 2020.6 Structurally advanced firms have deployed their AI initiatives at several times the rate of those further back in the maturity curve, and commitment is not retreating: every industry BCG tracks plans to increase AI spending this year, and only 6% of executives would pull back if 2026 initiatives do not pay off.2, 7
Concentration also explains the enclosure dynamic at the center of this edition: when returns pool around deep integration, the platforms beneath that integration acquire every incentive to meter access to it. Generating the return and retaining it are becoming separate questions.
The Agentic Acceleration
Agentic AI is compressing the timeline on which structural advantages compound. The length of tasks AI systems can reliably complete has doubled approximately every seven months since 2019 and every four months since 2024,8 and nearly three-quarters of companies now plan agentic deployment within two years.9 The open-weight advance has accelerated in parallel over the past quarter. DeepSeek released a preview of its V4 family in late April: a 1.6 trillion parameter flagship and a 284 billion parameter variant, both MIT-licensed and with serving support on Chinese silicon.10 Per MIT Technology Review, Chinese open-weight models accounted for 17.1% of global AI model downloads over the year ending August 2025, narrowly surpassing the US share for the first time,11 and Alibaba reports that its Qwen family surpassed one billion cumulative downloads on Hugging Face as of January 2026.12 The direction is what matters: open-weight models are collapsing the cost of frontier capability, and in our analysis defensibility is shifting to proprietary data, workflow position, and access.
Infrastructure Economics as a New Axis
A parallel axis of bifurcation is emerging around infrastructure economics. The IEA reports that data center electricity demand grew 17% in 2025, with demand from AI-focused facilities surging 50%, and that the capital expenditure of the largest technology companies exceeded $400 billion in 2025 and is expected to rise a further 75% in 2026.13 AI data center power capacity has already reached 29.6 GW, roughly what it takes to power the state of New York at peak demand.5 Companies dependent on significant cloud compute face rising and increasingly unpredictable costs; those that have shifted inference to the edge, a market projected to grow at a compound annual rate above 21% through 2033, may be building a structural cost advantage.14
How the Bifurcation Is Reaching Transaction Markets
The bifurcation is registering directly in how companies are valued and transacted. Product cycles are compressing,15 and organizations treating data as a formal strategic asset achieve two to three times the return on investment of peers that do not.16 Almost half of all technology deals now carry an AI rationale, and one in five strategic acquirers have walked away from a deal because of AI’s anticipated impact on the target’s business.17 In Lazard’s survey of private equity participants, 70% backed out of at least one active deal based on AI exposure concerns in the past year, and performance-linked structures are increasingly used to bridge AI valuation gaps.18 The window during which AI capabilities and data infrastructure are treated as incremental upside rather than baseline expectations is narrowing.
The first half of 2026 confirmed the direction, unevenly. Bain’s midyear data show global deal value up 41% year over year to $2.4 trillion through May; deal count rose only 2%, median valuations held flat at 11.6x EV/EBITDA, and Bain frames the year’s defining tension as a winner’s paradox: acquirers must fund an AI transformation and a complex integration at the same time.19 The market is K-shaped, and the lower middle market sits on the flatter limb.20 The counterpoint is volume: Software Equity Group (SEG) recorded a record 2,698 SaaS transactions in 2025, up 28%, with 659 more announced in the first quarter of 2026.21, 22 SEG’s strongest public categories ended 2025 near 6.7x to 6.9x EV/Revenue against a broader index at 4.8x,21 and private equity consolidators remain active buyers of AI-enabled roll-ups, with AI-driven margin expansion cited as the primary value creation lever in one survey of roll-up investors.23 How buyers apply these lenses, and how ownership of access chokepoints flows into multiples, is the subject of our forthcoming analysis of Valuation Architecture in the Intelligent Economy.
The Enclosure of the Intelligent Economy
If the bifurcation describes who generates the AI Premium, enclosure describes who retains it. The two do not sort identically: a structurally advanced company that does not own the terrain its agents traverse pays the toll, and the toll compresses its premium. The moves accumulated over two quarters and converged this Spring. Workday began charging for agent work through Flex Credits in late 2025 and governs it through an Agent System of Record.24 Snowflake has metered AI usage through dedicated AI Credits,25 and Datadog capped agent traffic on its Model Context Protocol (MCP) server.26 SAP published a revised API policy in late April that prohibits third-party autonomous agents from planning or executing sequences of API calls.27 Then, in early May, the pattern became unmistakable: at its Knowledge 2026 event, ServiceNow unveiled Action Fabric, a governed layer through which any external AI agent can act inside its platform, with every action metered in the same consumption currency that prices its own AI products.28 JPMorgan's software analyst characterized the charge as “a tax on customers”; the trade press converged on the word tollgate.29
The structural logic is straightforward. As open-weight models collapse the cost of raw capability, the scarce assets are the ones agents cannot route around: proprietary data, the workflow systems of record where enterprise processes actually live, and now the metered access layers built on top of both. Lazard’s deployment-era framework points in the same direction, separating systems of record, which benefit from data gravity and compliance moats, from workflow wrappers facing commodity risk.18 The technical substrate is consolidating as well: the MCP, the open standard through which agents request access to enterprise tools and data, surpassed 97 million monthly SDK downloads before Anthropic donated it to the Linux Foundation’s Agentic AI Foundation in December 2025.30 The tollgates are being built at exactly this layer.
In our view, durable value is migrating to those who own the chokepoints agents must cross: the data, the workflow, and now the toll.
The enclosure model is contested. SAP’s German-speaking user group has called for clarification of the API policy, Forrester has urged CIOs to resist platform gatekeeping,31 and several vendors are running countervailing open strategies. What is not in question is that access economics now exist where none did a year ago. For lower middle market companies, the dynamic cuts both ways: they are payers of new tolls in every multi-vendor AI deployment, and in vertical niches they are potential owners of chokepoints of their own. A company whose proprietary data and workflow position are genuinely embedded in its clients’ operations holds, in miniature, the same asset the incumbents are now monetizing at scale. Which side of the tollgate a business sits on is becoming a first-order strategic question.
The New Competitive Moat: Compounding Architecture
What separates companies on the premium side of the bifurcation from those on the compressed side? In our analysis, durable advantage rests on three compounding factors: what we call the Compounding Architecture framework, a proprietary analytical lens rather than a predictive model. The first is proprietary data: data uniquely generated, curated, or enriched through a company’s operations, whose value lies in exclusivity and context rather than volume. The enclosure dynamic is itself market evidence: incumbents are erecting tolls precisely because the data and workflow behind the gate are the assets that compound. The second is continuous learning loops: the capacity of systems and teams to improve through iterative feedback.32 Among high-performing firms, 55% fundamentally reworked processes when deploying AI, nearly three times the rate of other organizations.3 The third is deployment speed: the velocity with which a company translates insight into operational change; in compressed markets, the interval between recognizing a shift and acting on it has material consequences.8, 33 Building all three requires time, capital, and organizational commitment, and execution risk is material.
Unlike traditional moats, proprietary data, continuous learning loops, and deployment speed can compound with use rather than erode with competition.
Recent research identifies further reinforcing dimensions, including organizational plasticity and workflow ownership, where companies embed AI into the systems their clients depend on, building switching costs that compound over time.34 A dimension we call Perimeter Intelligent Architecture (PIA) cuts across all three factors: a bounded architecture in which AI processing remains confined within secure, localized boundaries, enabled by confidential computing, edge computing, and on-device AI. The unifying logic is simple: bring the intelligence to the data, and never let sensitive information cross the perimeter.
Bring the intelligence to the data. Never let sensitive information cross the perimeter.
The product architecture shift is measurable. Gartner projects that by 2027, organizations will use small, task-specific AI models three times more frequently than general-purpose LLMs,35 and the IEA finds the energy required per AI task is falling by at least an order of magnitude annually, which favors architectures designed to run small models close to the data.13 On-device deployment can deepen proprietary data advantages without regulatory exposure, and local inference tightens learning loops by eliminating cloud latency, though on-device deployment constrains model complexity and not every workload suits distributed processing. Where PIA does apply, regulatory readiness remains a deployment speed advantage under the EU AI Act, though the compliance clock has recently moved.36
The Diagnostic: Identifying Your Strategic Inflection Point
An inflection point is not a crisis. It is a signal that the current playbook is losing fit with the market. In our advisory experience, most firms encounter one or more recurring signals, what we refer to as Strategic Inflection Signals. These are frames for structuring strategic attention, not diagnoses or recommendations.
Viorum identifies ten recurring signals in the Intelligent Economy: the Capability Gap (client demand outpacing AI or data capabilities), the Growth Plateau (expansion flattening as markets consolidate around platforms), the Generational Competitor (an AI-Native resetting unit economics), the Data-Rich Insight-Poor problem (valuable data locked in legacy systems), the Unsolicited Offer (a buyer approaching with terms that discount strategic option value), the Leadership Bandwidth Limit (growth trajectory outpacing leadership capacity), the Regulatory Inflection (compliance reshaping product architecture), the Agentic Transition (autonomous AI systems restructuring workflows and economics), the Operating Model Obsolescence Point (internal cost structure and staffing model falling out of step with AI-enabled peers as benchmarks reset across the sector), and Platform Enclosure (metered agent access reshaping the economics of every multi-vendor AI deployment).
The first six signals are structural and recur across market cycles. Among them, the Generational Competitor, where an AI-Native Innovator resets unit economics on an architecture incumbents cannot replicate incrementally,10 the Data-Rich, Insight-Poor problem of valuable proprietary data locked in legacy systems,16 and the Leadership Bandwidth Limit, where growth or the technical demands of AI integration outpace leadership’s capacity, are the most frequently active in our current advisory conversations.
The Regulatory Inflection has reversed direction. As recently as this spring, the EU AI Act's high-risk obligations were set to take effect in August 2026. Under the Digital Omnibus, agreed provisionally in May and formally adopted at the end of June, those obligations were deferred: standalone high-risk systems to December 2, 2027, and high-risk systems embedded in regulated products to August 2, 2028, with the amendments entering into force upon publication in the Official Journal.37 The deferral does not empty the calendar. Transparency obligations still take effect on August 2, 2026, watermarking requirements follow in December 2026, and penalties still reach €35 million or 7% of global turnover.36, 37 The framing that matters is deferral plus fragmentation: the high-risk deadline moved out by roughly sixteen months while US state-level legislation continues to splinter, and the deferral may reward companies that keep building compliance capacity now.
The Agentic Transition captures the shift from tool-based AI to autonomous systems that restructure workflows and service economics. The acceleration is measurable: on the SWE-bench Verified benchmark, agent performance rose from roughly 60% to near 100% of the human baseline in a single year.5 The signal has two faces. Adopting agents inside the business is an execution question. Being disintermediated by agents operating on behalf of clients or platforms is a business model question, and it is the harder of the two.
The Operating Model Obsolescence Point emerges when a company’s internal cost structure and staffing model fall out of step with what AI-enabled peers are demonstrating. Major technology incumbents have initiated workforce reductions tied explicitly to AI-driven productivity gains rather than cyclical demand weakness, resetting benchmark expectations for headcount ratios, gross margins, and delivery economics across the sector. For lower middle market companies, the reference model for what a comparable business should look like has moved, and the diligence implications follow.
Platform Enclosure is the tenth signal, new to this edition. It emerges when the platforms a company depends on, or the platforms its agents must traverse, begin metering access: per action, per conversation, per resolution. Gartner projects that at least 40% of enterprise SaaS spend will shift to usage-, agent-, or outcome-based models by 2030, a third-party projection cited via Deloitte.38 Intercom/Fin (now being acquired by Salesforce) prices its support agent per resolution; Salesforce launched Agentforce at a per-conversation rate before broadening to a multi-model structure.39, 40 The signal carries a dual reading: rising access costs on one side and, in vertical niches, the possibility of owning a chokepoint of one’s own.
These signals are not mutually exclusive: a capability gap may compound into a growth plateau, the agentic transition intensifies nearly all of them, and Platform Enclosure adds an access-cost line to every agentic deployment while redrawing the boundary between platform and feature. The diagnostic value lies in recognizing which signals are active and how they interact within your specific context. A detailed framework for assessing them, including the Viorum Strategic Options Quadrant, follows in our companion Strategic Transitions Perspective.
The Role of Judgment in the Intelligent Economy
When every participant has access to sophisticated models and automated analysis, the competitive premium on those tools compresses. What does not compress is the premium on judgment: the capacity to weigh strategic trade-offs involving incomplete information and the contextual understanding that comes from years of operating in a specific sector. Yet the nature of decision-making itself is being restructured. By 2027, AI agents are projected to augment or automate half of all business decisions.41
The evidence is instructive. Executives using generative AI produced worse forecasts than peers who deliberated collaboratively, because the AI’s authoritative tone reduced the perceived need for dissent.42 Eighty percent of organizations have encountered risky behavior from AI agents, and most agentic systems reviewed in one comprehensive index disclosed no results from internal safety investigations.43, 44 In one 2026 industry survey, 88% of organizations deploying agents reported confirmed or suspected security incidents within the past year.45 For a founder weighing a transaction, the number that matters most in our view is that last one: agent governance that cannot be evidenced reads, in diligence, much the way an unaudited financial statement does. The transparency problem is worsening: the Foundation Model Transparency Index dropped from 58 to 40 points year-over-year, and over 90% of notable AI models are now created by private companies.5 The second International AI Safety Report reaches a compatible conclusion: capability is advancing faster than the ability to implement effective safeguards.46
The corrective is deliberate: leaders who maintain decisional authority, treat AI outputs as analytical inputs rather than conclusions, and preserve the deliberative processes that complex strategic decisions require. For founders and operators navigating inflection points, senior, experienced advisory is not diminished by AI. In our view, it is amplified. AI may provide better inputs. Judgment can provide better decisions. The outcomes that matter most tend to emerge from the disciplined combination of both.
The Strategic Imperative for Leaders
Better inputs and better judgment still require action. The bifurcation and the enclosure dynamic outlined in this Perspective are structural, not cyclical; the compounding dynamics that define the premium side accelerate as they mature. For AI-Constrained enterprises, deferred action may carry increasing opportunity cost.8 Four strategic pathways, capability acquisition through M&A, architectural transformation to rearchitect the operating model, a capital raise to accelerate investment, or organic development to deepen differentiation, each carry material trade-offs. Our companion Strategic Transitions Perspective examines each in detail.
Clarity and sequencing provide stability in uncertain markets. The aim is not to predict the future, but to position the organization to adapt as conditions evolve, from a foundation of strategic strength rather than reactive necessity. Strategic inflection signals mark moments of choice. In our experience, the companies that navigate them with care share one characteristic: they treated the inflection point not as an interruption, but as the catalyst for their next phase of value creation.
