The Seat Is Empty: Gartner's $234 Billion Forecast and the Quiet Collapse of Software Company Formation

$234 Billion Is the Number. The Formation Cliff Is the Signal.

Let us begin with the dollar figure that matters: $234 billion. That is Gartner's estimate, published last week, of enterprise application software spending now directly in the path of agentic AI disruption between now and 2030. The firm's managing VP, George Brocklehurst, was precise: "This breaks the link between user growth and revenue growth for many enterprise software vendors." That sentence deserves a slow read. It is not describing a pricing adjustment. It is describing the severance of the fundamental revenue equation on which two decades of SaaS valuation has rested.

Gartner is measuring the demand side of a structural break. The supply side tells a different story, and it is quieter, and more alarming.

AIBD analysis of Companies House SIC 62.02 filings, the classification covering computer programming and bespoke software activities, the seedbed from which enterprise SaaS startups emerge, recorded just 429 new company registrations in Q3 2026. That represents a 94.5% collapse versus the prior comparable period. Nearly the entire cohort of would-be challengers has stopped forming. The pipeline that historically replenished the enterprise software market with insurgent competition has, by this measure, effectively ceased.

Christensen's framework is useful here, if applied precisely. The Innovator's Dilemma predicts that incumbents are disrupted from below, by entrants offering simpler, cheaper products to overlooked customers. What the SIC 62.02 data suggests is something categorically different: new entrants are not forming traditional SaaS companies because the organisational unit of SaaS, a team that builds, ships, and licences a discrete software product, may itself be the thing being disrupted. The challenger is not a startup. The challenger is an agent.

The Pricing Architecture Is Already Breaking

The Gartner forecast is reinforced, not contradicted, by the pricing decisions already visible in the market. In June, GitHub shifted from flat-rate premium request pricing to token-level metering for input, output, and cached tokens. Atlassian's standard plans now include 25 Rovo AI credits per user per month, with overage charges of $0.30 per conversation. Zendesk prices its AI resolution agent at $1.50 per resolved conversation. HubSpot charges $10 per 1,000 AI credits beyond tier allotments.

These are not experiments. They are the visible renegotiation of what the unit of value in software actually is. For thirty years, the answer was the seat. The seat implied a user; the user implied an interface; the interface justified the subscription. Gartner now projects that by 2030, at least 40% of enterprise SaaS spend will shift toward usage-, agent-, or outcome-based pricing models. Deloitte frames the end state with structural clarity: SaaS applications are evolving "towards a federation of real-time workflow services that can learn from their experiences."

Record labels in 2003 also believed their distribution model was defensible. The per-album transaction had, after all, worked for fifty years. What they failed to account for was that MP3s did not attack the music; they attacked the unit of commerce. AI agents are not attacking software functionality. They are attacking the seat.

The Margin Arithmetic Nobody Is Discussing

There is a second-order consequence here that receives insufficient attention: gross margin compression. Traditional SaaS achieved 75% to 85% gross margins because the marginal cost of serving an additional customer approached zero after development. That mathematics no longer applies when the product generates value through inference calls, model hosting, and orchestration layers, each of which scales with usage.

ICONIQ's 2026 State of AI survey estimates average gross margins for AI products at approximately 52%. That is a thirty-point compression relative to the SaaS norm. Vendors increasing prices to absorb AI infrastructure costs are, simultaneously, facing the structural devaluation of seat-based revenue as agents begin doing the work of multiple users from a single licence. Zylo's survey of IT leaders found that 61% of organisations cut projects or initiatives because of unplanned SaaS cost increases in the past twelve months. The customer is already in pain. So is the vendor. The margin is being consumed from both directions.

The Bifurcation That Valuation Markets Have Already Priced

The SaaS Capital Index fell from 7.0x at the start of 2025 to approximately 3.8x by March 2026, a 46% multiple compression the market has taken to calling, with characteristic drama, the SaaSpocalypse. The selloff erased roughly $1 trillion in aggregate market capitalisation. But the bifurcation within that rout is the more instructive data point: companies with net revenue retention above 120% and Rule of 40 scores exceeding 50 have maintained 7x to 9x multiples regardless of size. The market is not punishing software. It is punishing software whose value resides entirely in the interface layer.

This is precisely the distinction that TechRadar's July 9 analysis by Netcall's leadership articulates: enterprise platforms derive their durable value from managing structured data, enforcing permissions, executing workflows, and maintaining audit trails. Agentic AI does not remove that requirement. In many cases, it sharpens it. The system-of-record is structurally safer than the system-of-engagement. What AI changes is which categories are most exposed, not whether enterprise software survives at all.

Gartner reinforces this with a forecast that receives less coverage than the $234 billion headline: by 2030, 85% of enterprise agentic AI investments will be bundled into existing SaaS and cloud renewals rather than delivered through net-new contracts. The dominant incumbent is not necessarily dying. It is being transformed into infrastructure.

The Formation Silence Is the Leading Indicator

Return, then, to the 429 SIC 62.02 registrations. The historical function of new company formation in this classification was to generate the competitive pressure that forced incumbents to innovate, to reprice, and to serve previously ignored segments. Christensen called this the mechanism of disruption; Geoffrey Moore called the resulting pressure the chasm's other side. The 94.5% collapse in that formation rate does not signal a healthy consolidation. It signals that the traditional mode of organising software innovation, a founding team, a product, a SaaS subscription, has lost its economic rationale for a significant portion of would-be entrants.

Some of that talent is building AI agents instead of software companies. Some of it is building on top of foundation models rather than distributing its own stack. And some of it has simply concluded that the unit economics of a new SaaS entrant, competing against incumbents who are bundling AI at scale, no longer pencil out. Ben Thompson's aggregation theory implies that when distribution consolidates, the cost of reaching an audience without an existing platform becomes prohibitive. The platform, in 2026, is increasingly the model provider.

The structural prediction, then: within 36 months, the enterprise software market will complete its bifurcation into two defensible categories. Systems of record with deep data moats and compliance requirements, commanding premium multiples. And AI-native vertical agents, priced on outcomes, scaling faster than any SaaS cohort in history. Everything between those poles, the dashboard, the workflow tool, the horizontal point solution, is already repricing toward zero or toward acquisition. The seat-based model is not dead. It is just no longer the architecture of new value creation. The Gartner number is $234 billion. The Companies House number is 429. Read them together.