Enterprise software & SaaS

Enterprise software & SaaS

Industry ID: 02 | Slug: enterprise-software | Researched: 2026-09-15 | Analyst: agent

Central question for this sector in 2026: is AI expanding or cannibalising software revenue? Short answer from the evidence on this date: it is expanding software spend and cannibalising services spend, while equity markets price the opposite for software. Gartner has 2026 software spend at $1,468B (+15.5%) and IT services at $1,570B (+5.3%); the MSCI World Software & Services Index fell more than 20% in the first two months of 2026. No disclosed customer metric at any major incumbent yet shows displacement. That gap is the sector's defining feature — and it is falsifiable.

1. Definition and boundaries

In scope. The application and tooling layer of the technology stack: horizontal SaaS (CRM, HCM, ERP, ITSM, collaboration), vertical SaaS, developer tools, data infrastructure and analytics, IT services and systems integrators, and the ERP/CRM/ITSM incumbents. Business models, pricing, distribution, labour and consolidation within that layer.

Explicitly out of scope, and who owns it. Frontier models, model training and AI compute belong to sector 01; this dossier treats model capability as an exogenous input and only records it where it changes application-layer economics (inference cost in gross margin, coding agents in developer tools). Pure cybersecurity — endpoint, network, SecOps, identity as a security product — belongs to sector 04; security-adjacent software appears here only where it is a governance or compliance feature of an enterprise application (e.g. data governance in a lakehouse, authorisation hardening in the MCP specification).

Boundary disputes, named.

1. Data platforms straddle 01 and 02. Snowflake and Databricks sell compute-metered infrastructure that increasingly serves model training and inference. They are kept here because their buyers, contracts and competitive set are enterprise data teams, not AI labs.
2. Hyperscalers are double-counted by construction. Microsoft appears here for Microsoft 365, Dynamics and GitHub, and in sector 01 for Azure AI capacity. Any revenue figure crossing that line must say which side it is on. Gartner's "software" line includes AI infrastructure software, which flatters application SaaS if quoted without that caveat — a trap this dossier flags repeatedly.
3. IT services sits uneasily here. Accenture, TCS, Infosys, HCLTech and Wipro are not software companies, but they are the delivery channel for enterprise software and the place where AI's revenue effect is currently most measurable. Excluding them would remove the best evidence in the sector.
4. Developer tools have left the sector's valuation gravity. After SpaceX paid $60B for Anysphere, dev tooling is being priced as strategic infrastructure by acquirers who see their own engineering cost base, not as SaaS by public-market comparables.
5. Identity and data governance are contested with sector 04. The rule applied here: if it is sold to a security team as a security control, it is 04; if it is sold to a data or platform team as a prerequisite for AI deployment, it is 02.

2. Subcategories

1. Horizontal SaaS — systems of record. CRM, ERP, HCM, ITSM. Multi-year contracts, high switching cost, seat-anchored pricing. Salesforce, SAP, Workday, ServiceNow, Oracle.
2. Horizontal SaaS — systems of engagement. Collaboration, marketing, support, design. Shorter cycles, lower switching cost, most exposed to AI-native substitution. HubSpot, Atlassian, Adobe, Intercom, Zendesk.
3. Vertical SaaS and vertical AI. Industry-specific workflow software — legal, clinical documentation, field service, restaurants, life sciences. Deep domain data, small individual TAMs. Veeva, Toast, Harvey, Abridge.
4. Developer tools. IDEs, coding agents, CI/CD, code hosting, package management. The fastest revenue-scaling subcategory in software history and the one with the most violent valuation re-rating. Anysphere/Cursor, GitHub, Atlassian, JetBrains.
5. Data infrastructure and analytics. Warehouses, lakehouses, catalogs, ETL, BI, observability. Consumption-priced by default, which is why it is structurally insulated from the seat problem. Snowflake, Databricks, Datadog, MongoDB.
6. Integration and middleware. iPaaS, API management, event streaming, and now agent gateways. The subcategory most directly threatened by MCP standardisation. MuleSoft, Boomi, Confluent.
7. IT services and systems integration. Consulting, application development and maintenance, managed services, BPO. Effort-priced, headcount-linked, and in active repricing. Accenture, TCS, Infosys, HCLTech, Wipro, Cognizant, Capgemini.
8. Agent platforms and AI application layer. Agentforce, Now Assist, Sierra, Fin. Defined by pricing unit rather than by function — this is where credits, per-resolution and outcome pricing are being tested.
9. Enterprise governance and compliance tooling. Data residency, model governance, audit trails — pulled forward by the EU AI Act and Data Act and by procurement's agent-risk questions.
10. Billing, metering and monetisation infrastructure. The unglamorous subcategory made suddenly strategic by the pricing shift: you cannot sell consumption you cannot meter.

3. Market structure

Concentration: winner-take-most within category, fragmented across categories. Each system-of-record category is a durable oligopoly — CRM, ERP, HCM and ITSM each have two to four credible enterprise vendors and essentially no new entrants in twenty years. Across categories the market is extremely fragmented: a large enterprise runs hundreds of applications. IT services is fragmented and price-competitive with no firm above roughly 5% share.

Market size — with modeller, date and caveat. Gartner (John-David Lovelock, published 2026-07-27) forecasts 2026 worldwide software spending of $1,468B, +15.5% (2025: $1,271B, +13.9%) and IT services of $1,570B, +5.3% (2025: $1,492B, +3.8%), within total IT spending of $6.37T, +14.2%. Two caveats that must travel with these figures: Gartner's software line includes AI infrastructure software, not only application SaaS; and Gartner revised its 2026 growth forecast four times in nine months — 9.8% (2025-10-22), 10.8% (2026-02-03), 13.5% (2026-04-22), 14.2% (2026-07-27). There is no independent audited counterpart series. Recorded as estimate/forecast, never as fact.

Where margin actually sits. Historically at the system-of-record application layer: 75-85% gross margin, 30%+ non-GAAP operating margin at scale (Salesforce 34.1%, Workday 31.1%, ServiceNow 29.5%, Datadog 23% in the most recent quarter). That position is being squeezed from two sides. Below, model and infrastructure providers capture inference economics that vendors do not separately disclose — so the gross-margin effect of AI features is currently invisible from outside. Above, buyers have a credible build threat at renewal for the first time since the move to cloud. Services margin is thinner (Accenture 17.0%) and compressing under AI deflation.

Barriers to entry. Still high for systems of record — data gravity, integration surface, compliance certification, reference customers, 7-15 year replacement cycles — and falling fast for systems of engagement, where coding agents have collapsed the cost of a credible v1 and MCP is collapsing the cost of integration. The moat is migrating from software to proprietary customer data plus distribution, which is precisely why Salesforce bought Informatica and SAP bought Reltio and Dremio.

Who has pricing power. Today: system-of-record incumbents at renewal (seat counts are contracted, switching is expensive), consumption platforms with high NRR (Snowflake 126%), and scarce developer-tool assets. Losing it: seat-priced systems of engagement, per-seat AI add-ons, and effort-priced services. Gaining it: buyers of large managed-services contracts, who per HFS are reopening deals within 24 months of signing rather than waiting for renewal.

4. Who matters

Leading companies. Microsoft (https://www.microsoft.com) · Salesforce (https://www.salesforce.com) · SAP SE (https://www.sap.com) · Oracle (https://www.oracle.com) · ServiceNow (https://www.servicenow.com) · Workday (https://www.workday.com) · Adobe (https://www.adobe.com) · Snowflake (https://www.snowflake.com) · Datadog (https://www.datadoghq.com) · Atlassian (https://www.atlassian.com) · HubSpot (https://www.hubspot.com) · Accenture (https://www.accenture.com) · TCS (https://www.tcs.com) · Infosys (https://www.infosys.com) · HCLTech (https://www.hcltech.com) · Wipro (https://www.wipro.com)

Notable startups and private companies. Databricks (https://www.databricks.com) · Anysphere/Cursor, now SpaceXAI (https://cursor.com) · Sierra (https://sierra.ai) · Intercom (https://www.intercom.com) · Harvey (https://www.harvey.ai) · Abridge (https://www.abridge.com) · Rippling (https://www.rippling.com) · Deel (https://www.deel.com) · Canva (https://www.canva.com)

Active investors. Andreessen Horowitz (https://a16z.com) · Thrive Capital (https://thrivecap.com) · Bessemer Venture Partners (https://www.bvp.com) · Thoma Bravo (https://www.thomabravo.com) · Vista Equity Partners (https://www.vistaequitypartners.com) · Nvidia as strategic investor (https://www.nvidia.com)

Platforms and standards bodies. Model Context Protocol (https://modelcontextprotocol.io) · Cloud Native Computing Foundation (https://www.cncf.io) · Apache Software Foundation, for Iceberg (https://iceberg.apache.org) · Linux Foundation (https://www.linuxfoundation.org) · OASIS (https://www.oasis-open.org)

Regulators. European Commission DG CONNECT (https://digital-strategy.ec.europa.eu) — Data Act and AI Act · US Securities and Exchange Commission (https://www.sec.gov) — disclosure regime · US Federal Trade Commission (https://www.ftc.gov) — software M&A review · UK Competition and Markets Authority (https://www.gov.uk/cma) · India's Ministry of Electronics and IT (https://www.meity.gov.in)

Research institutions and analysts. Gartner (https://www.gartner.com) · Everest Group (https://www.everestgrp.com) · HFS Research (https://www.hfsresearch.com) · Forrester (https://www.forrester.com) · IDC (https://www.idc.com) · DORA, Google Cloud (https://dora.dev) · US Bureau of Labor Statistics (https://www.bls.gov) · Indeed Hiring Lab (https://hiringlab.indeed.com)

Trade organisations. NASSCOM (https://nasscom.in) · BSA | The Software Alliance (https://www.bsa.org) · CISPE, for EU cloud interests (https://cispe.cloud) · TechUK (https://www.techuk.org)

Consumer and civil-society groups. Electronic Frontier Foundation (https://www.eff.org) — on automated decision-making · AlgorithmWatch (https://algorithmwatch.org) — EU AI accountability · European Digital Rights, EDRi (https://edri.org). Note: this is a predominantly B2B sector and civil-society engagement is thin except where employment-context AI is concerned.

5. Products, business models, technologies, customers

Major products. Systems of record (Salesforce Sales/Service Cloud, SAP S/4HANA, Workday HCM, ServiceNow Now Platform); productivity and collaboration (Microsoft 365, Slack, Atlassian); data platforms (Snowflake Data Cloud, Databricks Lakehouse, Microsoft Fabric); observability (Datadog); developer tools (Cursor, GitHub Copilot, Claude Code); and the new agent layer (Agentforce, Now Assist, Microsoft 365 Copilot, Intercom Fin, Sierra).

How money is made today. Multi-year subscription contracts priced per seat per month, with annual uplifts, tiered editions, and expansion sold through additional seats and additional modules. Revenue is recognised ratably; the forward-looking metrics that matter are cRPO and backlog.

How that is changing. Four distinct models now coexist, and the mix is the story:

• Per seat — still the contractual base everywhere. Microsoft 365 Copilot's 30M+ paid seats is the largest per-seat AI product in existence, so reports of the model's death are premature.
• Credits / metered consumption — Salesforce Flex Credits, HubSpot Credits, ServiceNow AI ACV uplift. The dominant incumbent response: metered, but not outcome-guaranteed.
• Pure consumption — Snowflake, Databricks, Datadog. Structurally immune to the seat problem; Snowflake's 126% NRR is the proof.
• Outcome / per-resolution — Intercom Fin at $0.99 per resolved conversation, Sierra. The only model that monetises the destruction of seats rather than being destroyed by it. Proven so far in exactly one domain (customer support) where the unit of work is already counted.

The measurement problem, stated plainly. No vendor discloses the split between seat revenue and metered revenue, and NRR has largely disappeared from seat-based vendors' disclosure. Among the major vendors examined here, only Snowflake published an NRR figure. The single most important question in this sector therefore cannot be answered from filings — only inferred from pricing pages, cRPO growth and backlog duration. That is a structural data gap, not a research shortfall.

Technologies that matter. Large language models as an input (sector 01); retrieval and semantic layers over enterprise data; agent orchestration and tool-calling; MCP as the integration substrate, whose 2026-07-28 revision made the protocol stateless with header-based routing and RFC 9207 authorisation hardening — the changes that make it deployable behind enterprise gateways; open table formats (Iceberg, Delta) commoditising the storage/compute boundary; and metering and billing infrastructure, newly strategic.

Customer segments and what they buy on. Large enterprise (>10,000 employees) buys on integration with the existing estate, compliance, vendor viability and reference customers — and is the segment where displacement is slowest and cRPO is holding. Mid-market buys on time-to-value and total cost, and is where AI-native substitution is most plausible. SMB buys on price and self-service, and is most exposed to being served by a general-purpose agent instead. Public sector buys on procurement frameworks and sovereignty. Developers — increasingly the actual buyer in dev tools and data infrastructure — buy bottom-up on individual productivity, which is why Cursor reached $2B ARR without a traditional enterprise sales motion.

6. Geography

Demand is broadly distributed but revenue recognition is US-centric: the US is the largest enterprise software market and the headquarters of most of the cohort. Europe is a large buyer and a comparatively small seller, with SAP the principal exception — and a useful one, because SAP's Q2 2026 (2026-07-23) showed current cloud backlog of EUR 22.9B, up 27% (+26% cc), cloud revenue +22% (+24% cc) and cloud ERP suite revenue +25%. European enterprise demand is therefore accelerating at least as fast as US demand, which argues against a demand-side reading of the software sell-off.

Production of labour and delivery concentrates in India, which is the sector's most important non-US market and the one where AI's revenue effect is already visible. This dossier's regional source is The Register's 2026-04-28 analysis (Simon Sharwood), which carries direct executive quotes: HCLTech CEO C. Vijayakumar on "AI deflation" cutting revenue 3-5% in the coming year, and TCS CEO K. Krithivasan describing the effect as "degrowth". Reported performance across the big four in that period: TCS -0.5%, Infosys +3.1%, Wipro +4% with lower margins on some deals, HCL +11.2% with headcount up nearly 2%. Critically, none implemented significant layoffs — which is why margin, not headcount, is absorbing the deflation. Infosys subsequently narrowed FY27 guidance to 1.5-3.0% constant currency.

Capital concentrates in the US: the Americas took 66% of Q2 2026 global VC funding, and the macro brief records extreme concentration (OpenAI and Anthropic alone at 43% of H1 2026 funding). Europe at $25.6B in Q2 remains the structurally weak leg, which is why European enterprise software produces buyers and one giant seller rather than a challenger cohort. Asia reached $50.8B, an 18-quarter high.

Regulation concentrates in the EU, decisively. The Data Act's switching obligations applied from 2025-09-12 across IaaS, PaaS and SaaS, and all switching charges are prohibited from 2027-01-12, with member-state penalties reaching 3-5% of global turnover in France and up to 4% or EUR 5M in Germany. The EU is, uniquely, legislating switching costs downward at exactly the moment challengers need lower exit friction. The counterweight: the same jurisdiction just deferred AI Act high-risk employment obligations from 2026-08-02 to 2027-12-02, so EU deadlines are not immovable.

Non-English and regional sources used. SAP SE investor disclosure (Walldorf, Germany); The Register (UK-based) on Indian IT; Indian issuer disclosure from TCS and Infosys; EUR-Lex is registered in the source registry as a multilingual primary feed. Honest gap: no Japanese, Korean, Chinese or Indian-language primary source was retrieved directly; NASSCOM's own 2026 publications were not accessed.

7. Historical trend patterns

This sector has a long record of predicted extinctions that did not happen, and of quiet structural shifts that did. Both matter for calibrating 2026.

What actually cycled, 2000-2026.

• 2000-02, the dot-com bust. Enterprise software revenue proved far more durable than internet advertising because it was contracted. The lesson repeated in every subsequent cycle: multi-year contracts create a 12-24 month lag between sentiment and revenue. That lag is exactly what makes 2026's evidence ambiguous today.
• 2005-2015, on-premise to SaaS. A genuine architectural displacement that took a decade. Salesforce was founded in 1999 and did not pass $1B revenue until 2009. Incumbents that were declared dead (SAP, Oracle, Microsoft) were not — they converted, slowly and expensively, and retained their customers. This is the single most relevant precedent for 2026: in the best-documented platform shift in this sector's history, incumbents lost multiple expansion but not customers.
• 2010-2015, "mobile will kill enterprise software." It did not. It became a feature.
• 2015-2020, the shift to consumption. AWS proved metered pricing at enterprise scale; Snowflake proved it for applications-adjacent infrastructure. Seat pricing survived alongside it. This is the precedent that argues 2026 produces coexistence rather than replacement.
• 2021-2022, the ZIRP bubble and its unwind. SaaS multiples reached 20-40x forward revenue and fell to single digits. Over-hiring in 2021 is a confounder that is still contaminating every "AI killed the jobs" analysis in 2026.
• 2023-2026, the AI wave. Currently unresolved.

Prior false positives in THIS sector — be specific.

1. "No-code/low-code will eliminate developers" (2018-2021). Gartner and a large vendor cohort forecast that citizen developers would displace professional development. Developer employment rose. The mechanism that failed: the hard part of enterprise software was never writing code, it was specification, integration, maintenance and accountability. This is the closest historical analogue to the 2026 coding-agent thesis, and it should temper it.
2. "RPA will automate the back office" (2017-2021). UiPath and Automation Anywhere reached enormous valuations on the claim; deployments proved brittle, expensive to maintain, and were later re-badged as "agentic". Gartner's 2026 finding that only ~130 of thousands of agentic vendors are genuinely agentic — the rest rebranding assistants, RPA and chatbots — is the same pattern recurring with new vocabulary.
3. "Blockchain for enterprise" (2016-2019). Near-total write-off. Instructive because the consortium press releases, pilot announcements and vendor claims looked identical to genuine adoption right up until the projects were cancelled. The tell that was available at the time, and is available now: pilots that never produce a disclosed production revenue line.
4. "Big data will replace the data warehouse" (2012-2016). Hadoop attracted enormous capital and ended as a feature of cloud data platforms. The incumbents absorbed the innovation. Relevant to anyone forecasting Snowflake or Databricks displacement today.
5. "Klarna replaced Salesforce and Workday with AI" (2024-2026). A live false positive still in circulation: CX Today established Klarna moved HR to Deel and CRM to a mix of other SaaS and in-house tools, retaining Slack. It is a vendor-swap story that has been cited for two years as the flagship evidence of AI-driven SaaS displacement.

What the pattern suggests. Every prior wave that was going to eliminate a layer instead became a feature of that layer, and the incumbents lost multiple rather than customers. The honest counter-argument for 2026: coding agents attack the cost of building the alternative, which none of the prior waves did. That is a genuinely different mechanism, and it is why this is scored as unresolved rather than dismissed.

8. What is changing now (as of 2026-09-15)

Six things are simultaneously true, and the tension between them is the sector's current state.

1. Spend is accelerating. Gartner has software at +15.5% to $1,468B in 2026, revised up four times in nine months. Reported results corroborate: ServiceNow +24.5% subscription, Snowflake +37% product revenue, Datadog +36%, SAP cloud +22%, Workday +13.9%, Salesforce +11%.
2. Valuations de-rated violently anyway. The MSCI World Software & Services Index fell more than 20% in the first two months of 2026; roughly $1T of software and services market value was removed in early February alone. The BVP Nasdaq Emerging Cloud Index sat at 1,876.37 on 2026-09-11, with an average revenue multiple of 7.5x on 20.6% average growth. Per F-Prime's Abdul Abdirahman, "this may be the first time in history that the terminal value of software is being fundamentally questioned."
3. AI revenue is real, disclosed, and small. Salesforce Agentforce ARR >$1.5B (+240%) and Agentforce + Data 360 ~$3.9B (+210%) against $46.1-46.4B FY27 guidance. ServiceNow AI ACV >$1B against ~$15.77B guided subscription revenue. Adobe AI-first ARR >$650M (+150%) against $27.5B total ARR. Microsoft 30M+ paid Copilot seats. Roughly 3-8% of the base, growing fast, not yet moving the aggregate.
4. Services is where cannibalisation is actually measurable. IT services spend growing 5.3% against software's 15.5%; HCL guiding to 3-5% AI-driven revenue decline; TCS -0.5%; Infosys FY27 guidance 1.5-3.0% cc; Accenture +3% local currency on ~799,000 people. HFS reports contracts being reopened within 24 months of signing.
5. The pricing unit is being renegotiated everywhere at once. Credits at Salesforce and HubSpot, ACV uplift at ServiceNow, pure consumption at Snowflake and Databricks, per-resolution at Intercom and Sierra, and per-seat still dominant at Microsoft. No convergence yet.
6. The macro is hostile to the growth-at-any-cost response. Per the macro brief: FOMC held at 3.50-3.75% on a 9-3 vote with three dissents in favour of a hike, core PCE ~3.3-3.4%, and the Fed explicitly flagged financial-stability risk from high AI-firm equity valuations and leveraged infrastructure financing. The 2026-09-16 FOMC decision falls one day after this research date and is unresolved. Any thesis in this sector that assumes cheap capital is untested.

The one-sentence synthesis. Enterprise software is being repriced on a terminal-value argument that current customer behaviour does not yet support, while the cannibalisation that everyone predicted for software is actually happening to IT services.

9. The five lists

(Overlap is explicit and noted. T-02-01, T-02-04 and T-02-06 each appear in three lists — that concentration is itself a finding: the pricing shift, services deflation and coding agents are the three forces doing most of the work in this sector.)

Five most important current trends

1. T-02-08 — Software budgets accelerating while services budgets stall (the aggregate answer to the central question)
2. T-02-04 — AI deflation in IT services: effort-based pricing repriced mid-contract
3. T-02-03 — Software equity de-rating ahead of any fundamental deterioration
4. T-02-01 — Per-seat licensing giving way to credit and consumption metering
5. T-02-06 — Near-universal AI coding tool adoption with falling developer trust

Five fastest-growing signals

1. T-02-11 — Strategic acquirers re-rating developer tools: the $60B Anysphere precedent
2. T-02-10 — Model Context Protocol as default integration substrate
3. T-02-02 — Disclosed AI revenue growing at triple-digit rates off small bases
4. T-02-09 — Outcome-priced AI agents reaching material revenue
5. T-02-14 — Vertical AI applications attacking horizontal SaaS budgets (fastest-growing by attention; weakest by evidence — see §10)

Five trends most likely to affect businesses

1. T-02-01 — Per-seat to consumption pricing (overlaps list 1: every renewal negotiation changes)
2. T-02-04 — Services repricing (overlaps list 1: immediate procurement leverage on large contracts)
3. T-02-16 — Seat-expansion-led NRR cooling (changes how software companies are underwritten)
4. T-02-13 — EU Data Act switching rules and the 2027-01-12 egress-fee ban
5. T-02-06 — Coding agents (overlaps list 1: changes engineering org design and build-vs-buy)

Five trends most likely to affect consumers (This is a B2B sector; consumer impact is mostly mediated through employment and service quality, and the honest scores in §5 of the trend records reflect that.)

1. T-02-07 — Software hiring contraction: postings a quarter below pre-pandemic levels (the largest direct consumer-facing consequence in this sector, via entry-level career paths)
2. T-02-15 — Software vendors cutting their own headcount with their own agents (support quality and job losses; overlaps T-02-07)
3. T-02-17 — Headcount-linked IT services decline (graduate hiring in India, a multi-million-person employment base)
4. T-02-06 — Coding agents (overlaps lists 1 and 3: reshapes a 1.9M-person US occupation)
5. T-02-09 — Outcome-priced AI agents (consumers increasingly interact with the agent, not a person, when they contact a company)

10. Overhyped / overlooked / cooling / reversing

Most overhyped — and the evidence that hype outruns substance.

1. "Enterprises are ripping out SaaS incumbents and replacing them with AI" (T-02-19). Roughly $1T of market value moved on this thesis. Against it: Salesforce cRPO +14% cc and described as accelerating, ServiceNow cRPO +21%, SAP current cloud backlog +27%, Workday 12-month backlog +14.2%. AllianceBernstein found "little sign of operational stress in customer behavior, retention metrics or reported financials". The flagship case study (Klarna) is a vendor swap. There is currently no disclosed customer metric at any major incumbent showing displacement.
2. Autonomous agentic AI in production (T-02-20). Gartner forecasts >40% of agentic AI projects cancelled by end-2027 on cost, unclear value and inadequate risk controls, and estimates only ~130 of thousands of self-described agentic vendors have genuine capability. Developer trust in AI output fell to 29%, down 11 points in a year, while usage rose to 84% — people are using it more and trusting it less, which is the opposite of an adoption curve approaching autonomy.
3. Honourable mention: vertical AI displacing horizontal SaaS (T-02-14). Not scored as overhyped because the direction is plausible, but every revenue figure in this cohort is private, self-reported and carried by aggregators and VC content. It is the place in this sector where a trend platform is most likely to launder a content-farm number into a decision.

Most overlooked — and why attention has missed it.

1. The disappearance of NRR from disclosure (T-02-16). Among the major vendors reviewed, only Snowflake — which is consumption-priced and therefore not seat-dependent — published a net revenue retention figure. A sector-defining metric quietly left the disclosure set and almost nobody commented, because the story everyone is chasing is AI, not accounting.
2. Workday's backlog duration. 12-month backlog +14.2% against total backlog +8.0%. Customers are committing for shorter periods. That is a cleaner early-warning indicator of confidence erosion than any AI metric, and it is sitting in plain sight in a press release.
3. Vendor pricing pages as a leading indicator (T-02-01). Packaging changes appear months before they show in revenue, are free to monitor, are dated, and are effectively unwatched systematically.
4. The EU Data Act's 2027-01-12 egress-fee ban (T-02-13). The only force actively legislating switching costs downward, landing exactly when challengers need it. Attention went to the AI Act.
5. Services deflation as a software opportunity. If services spend grows 5.3% while software grows 15.5%, budget is moving from people to product. The displacement story everyone is telling about software is actually happening to the consultancies — and software vendors are the beneficiaries.

Trends that appear to be cooling — with the indicator that turned.

1. Seat-expansion-led NRR (T-02-16). Indicator: Indeed software postings index at 76.12 against a 100 baseline, plus the disappearance of NRR from disclosure, plus Workday total backlog growth (8.0%) falling below 12-month backlog growth (14.2%).
2. The headcount-linked services pyramid (T-02-17). Indicator: executives now guide to it — HCL's "AI deflation" at 3-5%, TCS's "degrowth" — rather than denying it.
3. Standalone per-user AI copilot SKUs (T-02-18). Indicator: Adobe net new ARR down 36-37% YoY with management explicitly declining pricing actions in favour of freemium reach, plus the migration of AI onto credit pools at Salesforce and HubSpot. Counter-indicator that keeps this at "cooling" not "dead": Microsoft's 30M+ paid Copilot seats.

Trends that may reverse — and the mechanism.

1. The software de-rating (T-02-03) could reverse fast. Mechanism: two or three consecutive quarters of incumbent cRPO reacceleration alongside visible AI ARR would remove the terminal-value discount as quickly as it was applied. Salesforce's Q2 FY27 cRPO already accelerated. This is a sentiment trade on an unfalsifiable-in-the-short-run claim, and such trades unwind violently.
2. Consumption pricing could reverse toward committed contracts (T-02-01). Mechanism: CFOs hate unbudgetable variable cost. The 2015-2020 cloud precedent is instructive — consumption pricing ended up wrapped in multi-year committed spend agreements that behave like subscriptions.
3. The dev-tool valuation premium (T-02-11) could reverse. Mechanism: the $60B Anysphere price was paid in the stock of a newly-listed acquirer. If SPCX derates, the reference price retroactively becomes an artefact of paper rather than a market clearing level.
4. The EU Data Act egress deadline could slip (T-02-13). Mechanism: precisely what the EU just did to the AI Act's high-risk employment obligations, deferring them ~16 months from 2026-08-02 to 2027-12-02.

11. Risks and major uncertainties

Sector-specific risks.

• Gross-margin compression that is currently invisible. Inference cost sits inside cost of revenue and is separately disclosed by nobody. A vendor reporting 80% blended gross margin may be selling AI features at far lower incremental margin. This is the most important undisclosed number in the sector.
• Consumption and outcome pricing transfer forecast risk to the vendor. Revenue becomes less predictable at the exact moment the market is punishing unpredictability.
• Capital cost. Sticky 3-4% inflation and a hawkish-leaning Fed (three dissents for a hike in July 2026) mean growth-stage software cannot assume refinancing. The Fed has itself flagged AI-firm valuations as a financial-stability concern.
• Concentration risk in the evidence base. Gartner is effectively the sole modeller of IT spend; Crunchbase and KPMG disagree by ~10% on VC totals; private ARR figures are unaudited throughout.
• A visible agent failure in a regulated process would set the entire agentic category back years, independent of any individual vendor's quality.
• Concentrated standard stewardship. MCP has lead maintainers but no identified formal foundation governance, while carrying near-half-a-billion monthly SDK downloads.

Genuine unknowns — and the distinction that matters.

Things we don't know but could find out: the seat-versus-consumption revenue split (vendors have it; they don't disclose it); AI feature gross margin (same); actual enterprise switching rates under the EU Data Act (measurable from 2027); real production agent deployment depth as opposed to pilots (Gartner's cancellation forecast will be testable by end-2027); whether Adobe's net new ARR decline is strategy or erosion (two more quarters will tell).

Things nobody can know yet: whether the terminal value of application software is genuinely impaired — this depends on model capability trajectories that do not exist yet, and no amount of research on current filings resolves it. Whether coding agents lower the cost of building enterprise software enough to change build-vs-buy at scale, or whether specification, integration, maintenance and accountability remain the binding constraints as they did through the no-code wave. Whether the 2026 equity de-rating is a correctly-early signal or a narrative overshoot; the multi-year contract structure of this sector means the answer is structurally unavailable for 12-24 months. Anyone claiming certainty on these three in September 2026 is selling something.

12. Scenarios to 2030

Base — "coexistence and multiple compression" (most likely). AI is additive to software revenue and subtractive to services revenue, as the 2026 data already shows. Seats survive as the contractual base with credits layered on top; outcome pricing stays confined to domains with a countable unit of work. Incumbents retain customers but not multiples, repeating the 2005-2015 SaaS transition in which SAP and Oracle converted rather than died. Software spend compounds high single to low double digits; services grows low single digits and consolidates. Falsifiable early indicator: incumbent cRPO growth stabilises in the 10-20% band for four consecutive quarters while AI ARR passes ~15% of total revenue at two or more incumbents.

Upside — "AI expands the software wallet." Agents make software useful to populations that never had seats — frontline workers, customers, partners, machines — so the addressable unit count rises even as employee seats stall. Consumption pricing captures it. Gartner's four upward revisions in 2026 prove to be the start of a trend rather than a catch-up, and software spend grows faster than the 2026 forecast through the decade. Falsifiable early indicator: two or more incumbents disclose metered/consumption revenue growing faster than total revenue and total customer count rising faster than seat count — which requires them to start disclosing the split, itself the signal.

Downside — "terminal value impairment is correct, with a lag." The 2026 de-rating turns out to have been early rather than wrong. Renewal-cycle pressure shows up in FY28-29 as cRPO decelerates, discounting deepens, and seat counts fall with customer headcount. Vendors respond by harvesting margin, which accelerates the de-rating. Falsifiable early indicator: cRPO or current-backlog growth at two or more of Salesforce, Workday, ServiceNow or SAP falling below 8% year over year, combined with contract-duration shortening of the kind already visible in Workday's 8.0% total backlog growth.

Disruption — "build beats buy for the long tail." Coding agents cut the cost of an internal alternative far enough that enterprises stop buying the middle and long tail of their application estate, keeping only systems of record. MCP makes the remaining pieces interchangeable. Horizontal systems of engagement are hit hardest; vertical AI takes the high-value workflows. Falsifiable early indicator: a documented enterprise case — unlike Klarna — of a Fortune 500 company retiring a material commercial application in favour of an internally-built AI system, with disclosed cost and maintenance figures. None exists as of 2026-09-15.

Regulatory — "interoperability mandates reset switching." The EU Data Act's 2027-01-12 egress-fee prohibition binds on schedule and is enforced with real penalties; MCP or a successor becomes a de facto interoperability requirement; switching costs fall structurally across the EU and the norms export. Incumbent pricing power erodes for regulatory rather than technological reasons — the underrated path. Falsifiable early indicator: a first material enforcement action under the Data Act's switching provisions in 2027, or EU enterprises publicly reporting completed platform migrations citing the Act.

Failure — "the agentic cancellation wave." Gartner's >40% cancellation forecast is realised; agent-washing is exposed; a high-profile agent failure in a regulated process triggers procurement freezes. AI ARR growth rates collapse from triple digits to flat as pilots fail to convert, and the sector discovers it has been booking pilot revenue as ARR. Services firms partially recover as remediation demand returns. Falsifiable early indicator: sequential decline in disclosed AI ARR at any major incumbent, or a vendor quietly ceasing to disclose an AI metric it previously headlined — the same disclosure-disappearance pattern already observed with NRR.

13. Data gaps and limitations

Could not verify.

1. The seat-versus-consumption revenue split at any vendor. This is the central question of the sector and it is not a disclosed metric anywhere. All conclusions about the pricing shift rest on pricing pages, packaging changes and executive language, not on reported revenue.
2. Net revenue retention across the seat-based cohort. Only Snowflake (126%) published an NRR figure among the major vendors reviewed. No credible independent NRR benchmark series exists — the entire public space for SaaS multiples and NRR benchmarks is dominated by SEO content farms with undisclosed samples, which this dossier refused to cite.
3. AI feature gross margin / inference cost. Disclosed by nobody.
4. Oracle Q1 FY2027 (reported 2026-09-10) specifics. Fetches were blocked (403); RPO and cloud figures circulating are Tier C. Oracle is therefore materially under-covered here despite being a top-five vendor.
5. Adobe Q3 FY2026 figures reached this research only via an aggregator (BigGo, Tier C) after Adobe IR and major outlets returned 403. Recorded as single_source with the tier flagged.
6. Private ARR and valuation figures throughout — Cursor, Sierra, Intercom Fin, Harvey, Abridge, Databricks, Rippling, Canva. All are self-reported or press-reported, none audited. Cursor's ARR is actively contradictory (see below).
7. Exact press-release dates for ServiceNow Q2 2026 and Snowflake Q2 FY2027 were not captured from the SEC-hosted exhibits; dates recorded (2026-07-22 and 2026-08-26) are inferred from the reporting period and should be confirmed in Phase 2.
8. Infosys FY27 guidance (1.5-3.0% cc) rests on a single Tier C outlet; the Tier A/B versions were robots-blocked. Flagged single_source.
9. AI coding tool market share. No Tier A or B source was obtainable; the entire space is content farms. Deliberately omitted rather than laundered.
10. No primary non-English-language source was retrieved. SAP's disclosure is German-origin but English-language; The Register is UK-based reporting on India. NASSCOM's own 2026 publications were not accessed. This is a real gap for a sector whose delivery base is India.

Paywalled or licence-restricted. Gartner's underlying forecast documents (only press summaries are free); Everest Group and HFS Research full research; Crunchbase's API; sell-side research. Phase 2 needs a licensing budget for Gartner and Crunchbase specifically — they are unavoidable here.

Where the numbers conflict. Four material contradictions are recorded in the trend records rather than silently resolved:

• Software developer employment direction. Indeed/FRED postings index 76.12 (~24% below the Feb 2020 baseline) versus BLS OOH 1,905,400 jobs and +10% projected to 2035. Flow versus stock, and a decade horizon versus a current reading. Quoting either alone gives a false picture.
• Expansion versus cannibalisation. Gartner software spend +15.5% versus MSCI World Software & Services -20% in two months. Realised spend versus priced terminal value.
• Cursor's revenue base. $2B ARR (Feb 2026) versus Morgan Stanley's projection of up to $13B of SpaceX revenue by 2027. The $60B price cannot be reconciled to disclosed ARR without accepting the forward figure on trust.
• Databricks run-rate. $6.9B (CNBC) versus $7B (Everest Group) for the same month — both trace to the company's own unaudited disclosure.
• Plus the programme-level Crunchbase ($510B) versus KPMG ($560.4B) H1 2026 VC divergence from the macro brief, which is a definitional difference in deal inclusion.

Methodological caution specific to this sector. Search results for SaaS pricing, SaaS multiples, NRR benchmarks, AI coding market share, MCP adoption statistics and "vertical AI eating horizontal SaaS" are almost entirely Tier C content farms publishing unsourced CAGR and benchmark figures. These were excluded, which is why this dossier leans so heavily on SEC filings, company IR, FRED and BLS. The upside: this is the best-instrumented sector in the programme for primary data.

14. Ranking scorecard

15. Sources

1. ServiceNow Reports Second Quarter 2026 Financial Results — ServiceNow / SEC EDGAR — https://www.sec.gov/Archives/edgar/data/1373715/000137371526000072/erq2fy26.htm — 2026-07-22 — A
2. Salesforce Delivers Record Second Quarter Fiscal 2027 Results — Salesforce Investor Relations — https://investor.salesforce.com/news/news-details/2026/Salesforce-Delivers-Record-Second-Quarter-Fiscal-2027-Results/default.aspx — 2026-08-26 — A
3. Snowflake Reports Financial Results for the Second Quarter of Fiscal 2027 — Snowflake / SEC EDGAR — https://www.sec.gov/Archives/edgar/data/1640147/000164014726000033/fy2027q2earnings.htm — 2026-08-26 — A
4. Microsoft FY26 Q4 Earnings Press Release — Microsoft Investor Relations — https://www.microsoft.com/en-us/investor/earnings/fy-2026-q4/press-release-webcast — 2026-07-29 — A
5. Workday Announces Fiscal 2027 Second Quarter Financial Results — Workday Investor Relations — https://investor.workday.com/news-and-events/press-releases/news-details/2026/Workday-Announces-Fiscal-2027-Second-Quarter-Financial-Results/default.aspx — 2026-08-27 — A
6. Datadog Announces Second Quarter 2026 Financial Results — Datadog Investor Relations — https://investors.datadoghq.com/news-releases/news-release-details/datadog-announces-second-quarter-2026-financial-results — 2026-08-06 — A
7. SAP Announces Q2 and Half-Year 2026 Results — SAP SE News Center (Walldorf, Germany) — https://news.sap.com/2026/07/sap-announces-q2-and-half-year-2026-results/ — 2026-07-23 — A (regional/non-US source)
8. Accenture Reports Third-Quarter Fiscal 2026 Results — Accenture plc via Business Wire — https://www.businesswire.com/news/home/20260618029271/en/Accenture-Reports-Third-Quarter-Fiscal-2026-Results — 2026-06-18 — A
9. Software Developers, QA Analysts, and Testers — Occupational Outlook Handbook — US Bureau of Labor Statistics — https://www.bls.gov/ooh/computer-and-information-technology/software-developers.htm — 2026-08-27 — A
10. Software Development Job Postings on Indeed in the United States (IHLIDXUSTPSOFTDEVE) — Indeed Hiring Lab via FRED, Federal Reserve Bank of St. Louis — https://fred.stlouisfed.org/series/IHLIDXUSTPSOFTDEVE — 2026-09-04 — A
11. Bessemer Venture Partners Nasdaq Emerging Cloud Index (NASDAQEMCLOUD) — BVP/Nasdaq via FRED — https://fred.stlouisfed.org/series/NASDAQEMCLOUD — 2026-09-11 — A
12. Gartner Forecasts Worldwide IT Spending to Grow 14.2% in 2026, Totaling $6.37 Trillion — Gartner, Inc. — https://www.gartner.com/en/newsroom/press-releases/2026-07-27-gartner-forecasts-worldwide-it-spending-to-grow-14-point-2-percent-in-2026-totaling-6-point-37-trillion — 2026-07-27 — B
13. Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027 — Gartner, Inc. — https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027 — 2025-06-25 — B
14. The 2026-07-28 Specification — Model Context Protocol project blog — https://blog.modelcontextprotocol.io/posts/2026-07-28/ — 2026-07-28 — A (standards body)
15. Mind the gap: Closing the AI trust gap for developers — Stack Overflow — https://stackoverflow.blog/2026/02/18/closing-the-developer-ai-trust-gap/ — 2026-02-18 — A
16. New DORA Report Claims Strong Engineering Foundations Drive AI Return on Investment (DORA 2026.01) — InfoQ, reporting Google Cloud DORA research — https://www.infoq.com/news/2026/05/dora-roi-ai-assisted-dev-report/ — 2026-05-11 — B
17. EU Data Act: Significant New Switching Requirements Due to Take Effect for Data Processing Services — Latham & Watkins LLP — https://www.lw.com/en/insights/eu-data-act-significant-new-switching-requirements-due-to-take-effect-for-data-processing-services — 2025-08-19 — B
18. SaaS in, SaaS out: Here's what's driving the SaaSpocalypse — TechCrunch (Dominic-Madori Davis) — https://techcrunch.com/2026/03/01/saas-in-saas-out-heres-whats-driving-the-saaspocalypse/ — 2026-03-01 — B
19. 'AI deflation' comes to India's tech services giants — The Register (Simon Sharwood) — https://www.theregister.com/software/2026/04/28/ai-deflation-comes-to-indias-tech-services-giants/5225686 — 2026-04-28 — B (regional source: India)
20. Software's Big Sell-Off: Structural Risk or Narrative Noise? — AllianceBernstein (Hargis, Winkelmann, Fogarty, Qiu) — https://www.alliancebernstein.com/us/en-us/investments/insights/investment-insights/softwares-big-sell-off-structural-risk-or-narrative-noise.html — 2026-03-05 — B
21. Cursor in talks to raise $2B at $50B valuation after hitting $2B ARR in three years — The Next Web (Ana Maria Constantin) — https://thenextweb.com/news/cursor-anysphere-2-billion-funding-50-billion-valuation-ai-coding — 2026-04-18 — B
22. SpaceX completes record $60 billion acquisition of AI coding platform Cursor — Yahoo Finance — https://finance.yahoo.com/technology/ai/articles/spacex-completes-record-60-billion-131311785.html — 2026-08-14 — B
23. SpaceX to acquire the AI coding startup Cursor for $60 billion — CNBC — https://www.cnbc.com/2026/06/16/spacex-spcx-cursor-acquisition-ipo.html — 2026-06-16 — B
24. AI drives IT contract renegotiations within 24 months as pricing models shift — HFS Research — https://www.hfsresearch.com/news/ai-drives-it-contract-renegotiations-within-24-months-as-pricing-models-shift/ — 2026-06-30 — B
25. Picking an enterprise data platform in a market with no clear winner — Everest Group (Agarwal, Modi, Menon) — https://www.everestgrp.com/blogs/picking-an-enterprise-data-platform-in-a-market-with-no-clear-winner — 2026-09-08 — B
26. Klarna Didn't Replace Salesforce & Workday with AI; It Replaced Them with Alternative SaaS Apps — CX Today (Charlie Mitchell) — https://www.cxtoday.com/crm/klarna-didnt-replace-salesforce-it-replaced-them-with-alternative-saas-apps/ — 2024-12-11 — B
27. Databricks revenue growth tops 80% to $6.9 billion annualized — CNBC — https://www.cnbc.com/2026/06/16/databricks-revenue-growth-tops-80percent-to-6point9-billion-annualized.html — 2026-06-16 — B
28. Salesforce CEO confirms 4,000 layoffs 'because I need less heads' with AI — CNBC — https://www.cnbc.com/2025/09/02/salesforce-ceo-confirms-4000-layoffs-because-i-need-less-heads-with-ai.html — 2025-09-02 — B
29. Adobe Q3 2026 Earnings Call: $6.76B revenue, AI-first ARR up 150% — BigGo Finance (aggregating Adobe earnings call) — https://finance.biggo.com/news/US_ADBE_2026-09-10 — 2026-09-10 — C (aggregator; Adobe IR returned 403 — flagged single_source)
30. The BVP Nasdaq Emerging Cloud Index — Bessemer Venture Partners — https://cloudindex.bvp.com/ — 2026-09-15 — B
31. Intercom's Fin AI agent nearing $100M ARR on outcome-based pricing ($0.99/resolution) — Enterprise DNA AI Pulse — https://enterprisedna.co/resources/ai-pulse/ai-pulse-2026-08-02-intercom-s-fin-ai-agent-is-nearing-100m-arr-roughly-half-of/ — 2026-08-02 — C (aggregator; unaudited, flagged single_source)
32. Infosys Q1 results: FY27 revenue growth guidance narrowed to 1.5%-3.0% in constant currency — Upstox — https://upstox.com/news/market-news/earnings/infosys-q1-results-fy-27-revenue-growth-guidance-narrowed-to-1-5-3-0-in-constant-currency/article-197461/ — 2026-07-23 — C (Tier A/B versions robots-blocked; flagged single_source)
33. Global startup exits, IPO, M&A soar with AI — Q2/H1 2026 — Crunchbase News — https://news.crunchbase.com/venture/global-startup-exits-ipo-ma-soar-ai-q2-h1-2026/ — 2026-07-03 — B

Tier distribution: A = 12, B = 17, C = 4. No claim in this dossier rests on a Tier C source alone without being explicitly flagged single_source with the tier named in the trend record.

More research in this sector · Explore trend records