2026-08-03T18:30:00.000Z
The window in which a European bank could treat AI governance as a documentation exercise closed on 2 August 2026. That was the date the European Commission's enforcement powers over general-purpose AI systems became active, and integration platforms sit directly in the exposure path because they are the layer where data crosses systems, jurisdictions, and models. The regulation names no integration vendor. It does not have to.
Kaiso Research's primary dataset values the AI integration platform market at USD 8.30 billion in 2025 and projects USD 159.61 billion by 2035 at a 34.40% CAGR, yet the constraint that decides who captures that growth is architectural rather than budgetary. The same dataset names legacy system compatibility and integration complexity as the restraints holding adoption back.
That tension is the story. Enterprises will buy these platforms at close to the forecast rate. Far fewer will realise the operational gain, and the dividing line runs through how much unretired integration debt sits underneath the purchase. Two organisations can sign the same contract in the same quarter and land in very different places by 2029.
Why the 34.40% CAGR Measures Demand and Not Delivered Outcomes
A 34.40% compound annual growth rate between 2026 and 2035 records what enterprises are committing to buy, not what they have running in production. The distinction matters because the primary data behind this coverage points to legacy system compatibility as an active restraint. Purchase volume and realised outcome are different curves. Most coverage of this market collapses them into one.
Software carries the larger share by component across the forecast period, with services trailing as an attachment rather than a driver. Within applications, Data Integration and Management leads, ahead of API and Workflow Automation, Customer Experience Management, and Business Process Optimization. The ranking is diagnostic. Enterprises are still solving the problem of moving data reliably before they solve the problem of acting on it automatically.
BFSI holds the leading end-use position, ahead of Healthcare, Retail and E-commerce, Manufacturing, and IT and Telecommunications. That result inverts the usual efficiency argument. Banking and insurance do not lead because they have the freest hand to experiment; they lead because reconciliation, reporting, and audit trails across fragmented core systems are mandated work with a deadline attached. Compliance pressure, not discretionary modernisation, funds the largest share of this category.
North America holds the largest regional share while Asia-Pacific grows fastest across the 2026 to 2035 window, and cloud deployment leads over on-premise throughout. Those two facts pull vendor roadmaps in opposite directions. A cloud-first product strategy suits Asia-Pacific greenfield buyers and collides with the data-residency posture of regulated accounts in North America and Europe.
Three Forces Pushing Integration Budgets and the One Pulling Them Back
Three demand-side forces explain the 34.40% growth rate recorded for 2026 to 2035, and one supply-side constraint explains why realised deployments will trail it. The forces are additive. The constraint is not.
The first force is application count. Enterprise software estates have grown by accumulation rather than design, and every new system of record adds a connection surface that someone has to maintain. A single insurer may run SAP SE for finance, Salesforce for distribution, and a mainframe policy system nobody has the budget to retire.
Data Integration and Management leads the application segment in the primary data for exactly this reason: the backlog is not glamorous work, but it's unavoidable work. Integration debt compounds quietly until a strategic initiative stalls on it.
The second force is the arrival of AI agents that need governed access to production systems. Gartner projects that 40% of enterprise applications will be integrated with task-specific AI agents by the end of 2026, up from under 5% in 2025. An agent that can read a CRM record but cannot write a validated order back to it is a demo.
Agentic AI is the demand driver behind that shift, and the integration layer is where it either works or stalls. Turning a demo into a production workflow requires precisely the connection, authentication, and error-handling layer this market sells.
The third force is real-time decisioning inside regulated processes. When a bank runs sanctions screening or a hospital network reconciles patient records across facilities, the batch reconciliation window becomes the compliance exposure. That pressure is why BFSI leads end-use adoption in the primary data rather than a sector operating under lighter oversight.
Against those three sits legacy system compatibility, which the dataset names as the leading restraint alongside integration complexity and data quality gaps. This is where the forecast and the field diverge. A 34.40% CAGR through 2035 is growing faster than most CIO organisations have staffed for in their three-year operating plans, and the gap gets filled with services spend, contractors, and slipped go-live dates.
The asymmetry is worth stating plainly. A platform goes live in weeks. The estate underneath it takes quarters, and inside a bank running a forty-year-old core system it takes years.
Microsoft, IBM, and MuleSoft Are Buying the Same Layer for Different Reasons
The vendor set in this market splits into three groups that compete for one budget line while selling to different buyers. Microsoft Corp., Google LLC, Oracle Corp., SAP SE, and IBM Corp. sell integration as an extension of a stack the customer already runs. MuleSoft LLC, SnapLogic Inc., Talend Inc., TIBCO Software Inc., Jitterbit Inc., Celigo Inc., WSO2 Inc., Workato Inc., and Tray.ai Inc. sell it as the neutral layer between stacks. Zapier Inc., Make Inc., n8n GmbH, Pipedream Inc., and Activepieces sell it as something an operations lead configures without filing a ticket.
The incumbents hold a structural advantage that has little to do with product quality. When SAP SE or Oracle Corp. already owns the system of record, the integration layer arrives inside an existing enterprise agreement, an existing security review, and an existing support relationship. Procurement friction is a moat. Specialists have to win on capability against a competitor whose paperwork is already signed.
IBM Corp. and Google LLC occupy a position the other incumbents do not. Neither owns the dominant system of record in most accounts, so both compete on the data tier rather than the application tier, which is consistent with IBM Corp. buying integration outright rather than building it. Their argument is that the connection layer should be independent of whoever won the ERP or CRM decision. That argument lands hardest in the accounts where those two decisions went to different vendors.
The specialists answer that with heterogeneity, and it is a real answer. MuleSoft LLC and SnapLogic Inc. earn their position in accounts running SAP SE, Microsoft Corp., and a mainframe simultaneously, which describes most large banks and most national healthcare networks. Talend Inc. and TIBCO Software Inc. hold ground where data quality and event streaming matter more than application connectivity. Neutrality is defensible only as long as the estate stays mixed.
The automation-first group is the segment most buyers underestimate. Zapier Inc., Make Inc., n8n GmbH, Pipedream Inc., and Activepieces entered through individual and departmental budgets rather than central IT, which means they arrive inside the enterprise before anyone evaluates them. That's a distribution advantage rather than a product one, and it is how shadow integration estates form. By the time governance catches up, those workflows are load-bearing.
Consolidation logic favours the incumbents, and pricing structure is the reason. An independent specialist has to justify a separate line item against a stack vendor that folds integration into an existing agreement and against an automation tool a department already expensed. That squeeze does not resolve through better features. It resolves through acquisition or through a data-engineering position the stack vendors have not built.
Five Product Releases Between July 2024 and April 2025 Reset the Feature Baseline
Between July 2024 and April 2025, five vendors shipped variations of the same capability, and it changed what a buyer can reasonably demand. MuleSoft LLC opened the sequence in July 2024 with generative AI that produces integration logic from a description, cutting development time by 60% on the workloads measured. Talend Inc. followed in September 2024 with AI-enabled data integration aimed at analytics pipelines.
Microsoft Corp. embedded Copilot into its enterprise integration platform services in November 2024, allowing workflows to be configured in natural language rather than through connector-by-connector setup. Google LLC brought AI-powered workflow automation to mid-market buyers in January 2025. Zapier Inc. closed the sequence in April 2025 with intelligent workflow suggestions that cut configuration complexity by 70%.
Read as a set, those five releases retired authoring effort as a competitive differentiator. Building a connector by hand was the thing that made integration projects expensive, and generative AI collapsed that cost across the entire field inside ten months. What did not get cheaper is the part that actually breaks: schema drift, exception handling, credential rotation, and the reconciliation logic deciding what happens when a downstream system rejects a record.
The 60% and 70% reductions reported in 2024 and 2025 are real, and they are narrow. They measure time to build, not time to run, and a workflow written in a week still takes a quarter to certify inside a regulated environment. Buyers evaluating on demo speed in 2026 are measuring the part of the problem vendors already solved.
The more useful evaluation question is what the platform does at three in the morning when a partner API changes its response format without notice. No vendor demonstrates that.
The Model Context Protocol Turned Connector Sprawl Into a Standards Question
The architectural choice separating durable deployments from stranded ones is whether a platform's connection layer is proprietary or protocol-based. Every vendor in this category maintains a connector catalogue, and every catalogue is a maintenance liability that grows with the number of systems a customer runs. The Model Context Protocol, introduced by Anthropic in November 2024 and moved to Linux Foundation stewardship under the Agentic AI Foundation in December 2025, reframes that liability as a standards problem rather than a vendor problem.
Cross-vendor adoption is what makes this consequential rather than interesting. That foundation was co-founded by Anthropic, Block, and OpenAI with support from Google, Microsoft, AWS, Cloudflare, and Bloomberg, and the protocol's 2026 roadmap prioritises enterprise authentication and registry infrastructure over new surface area. Those are the two things an integration buyer needs before any protocol can carry production traffic.
The counter-argument deserves a hearing. A protocol reduces switching costs, which gives the vendors holding the largest catalogues the weakest incentive to standardise, and a declared implementation is not the same thing as a default path in a regulated deployment. Watch which vendors ship registry and audit support rather than which ones issue a statement of intent.
For buyers, this changes the diligence question. The right thing to ask a vendor in 2026 is not how many connectors ship in the catalogue, but what happens to the customer's investment when a standard connection layer makes catalogue size less valuable. A platform whose defensibility rests entirely on connector count is holding a depreciating asset.
August 2026 Moved EU AI Governance From Documentation to Enforcement
The European Commission gained active enforcement powers over general-purpose AI systems on 2 August 2026, alongside the Article 50 transparency obligations requiring disclosure when a person interacts with an AI system or with AI-generated content. Non-compliance with the general-purpose AI and transparency provisions in force from 2 August 2026 carries penalties up to EUR 15 million or 3% of global annual turnover, while the Article 5 prohibitions reach EUR 35 million or 7%. Legal analysis of the steps required before that date placed governance documentation, incident reporting, and supplier contract terms at the centre of readiness work.
The timeline is not settled, which is itself a planning problem. Amendments approved by the European Parliament in June 2026 moved standalone Annex III high-risk obligations to 2 December 2027, giving deployers additional runway while leaving the transparency and general-purpose duties in force. Integration platforms sit awkwardly across that split, because they are rarely the AI system and almost always the data path into one.
Under the General Data Protection Regulation, that data path is already a controller-processor question with documented obligations. Adding model inference to it does not create a new duty so much as make an existing one much harder to evidence.
IBM Paid USD 2.3 Billion for Integration Plumbing and Set a Price Floor
IBM Corp. completed its acquisition of StreamSets and webMethods from Software AG on 1 July 2024 for USD 2.3 billion, closing a deal announced in December 2023 that brought in products serving more than 1,500 companies. The purchase was not positioned as an AI acquisition. It was positioned as the data and integration substrate the automation and AI portfolios would run on, which is how IBM described it at close.
That framing is the thesis shift worth tracking. Capital in this category is moving toward the unglamorous layer, because the model itself is commoditising while governed connection to enterprise data is not. A USD 2.3 billion price on integration middleware in 2024 set a reference point that makes the remaining independents both more expensive and more likely to transact. Workato Inc., SnapLogic Inc., and Tray.ai Inc. are the names that reference point now prices.
For a CIO, the practical consequence is vendor continuity risk. An integration layer selected in 2026 has a meaningful chance of being owned by a different company before the contract reaches renewal. Write the exit terms first. A migration clause negotiated during procurement costs nothing, while the same clause requested after an acquisition costs whatever the acquirer decides it costs.
Three Failure Modes That Never Appear in a Platform Demo
The three risks that end AI integration programmes are data quality, identity, and organisational ownership, and none of them surface during evaluation. Data quality fails first, because a generated workflow inherits whatever the source system contains, and a model acting on inconsistent master data produces confident wrong decisions at machine speed. Nobody catches it during a pilot, because pilots run on curated data.
Identity fails second. When AI agents act on behalf of a user across six systems, the authorisation question becomes which permissions apply and who is accountable for the resulting action. Enterprise-grade centralised authentication only reached the Model Context Protocol specification in 2026, which means most deployments running today are improvising it. Improvised authorisation is the finding an auditor writes up first.
Ownership fails third and most quietly. Integration platforms are bought by IT, configured by operations, and audited by compliance, and in most organisations no single function owns the failure. In a European bank, that gap is what the European Commission's Article 50 disclosure duty lands on. The skills shortage named in the primary data as a restraint is really an accountability gap wearing a hiring-problem costume.
Hiring an integration architect does not repair a governance model that never assigned the workflow to anyone.
What CIOs and Compliance Leaders Should Settle Before December 2027
Two decisions determine whether a platform bought in 2026 still serves the business in 2029, and both need settling before the December 2027 high-risk deadline arrives. The pattern that emerges from Kaiso Research's primary market data is that neither decision is about the platform.
For CIOs and platform owners, the first decision is sequencing. Retire the integration debt in the two or three systems carrying the most cross-system traffic before buying the platform, because a generated workflow built on an unreliable source produces faster failure rather than faster delivery.
The second decision is contractual: negotiate connector portability and data-residency terms into the initial agreement, while competitive tension still exists. Sequencing is not a technical preference. It is the difference between a platform that pays back inside eighteen months and one that becomes another system nobody trusts.
For compliance and risk leaders, the work is inventory before policy. Most organisations cannot currently produce a list of which automated workflows touch personal data, which of those invoke a model, and which run on tools a department expensed without review. That inventory is the precondition for every Article 50 disclosure obligation the European Commission now enforces and every GDPR record of processing. AI governance tooling can automate the record-keeping, but nothing on the market discovers a workflow that nobody documented.
Start the inventory now. The organisations handling December 2027 comfortably will be the ones that treated 2026 as a discovery year rather than a deployment year.
Where the Market Lands by 2035 if Legacy Retirement Holds Its Current Pace
The path from USD 8.30 billion in 2025 to USD 159.61 billion in 2035 is a demand curve, and it will hold. Enterprises don't get to opt out of connecting systems, and every AI initiative adds connection surface rather than removing it. Agentic AI compounds that, because an agent touches more systems per task than the human workflow it replaces ever did. Asia-Pacific grows fastest across the 2026 to 2035 window while North America holds the largest share, which means the fastest-growing revenue arrives from buyers carrying the least legacy weight and the least tolerance for enterprise pricing.
The figure that will not track the forecast is realised value. A 34.40% CAGR describes what gets bought. Nothing in the data describes what gets retired. That second number is the one worth building an operating plan around, and no vendor in this category has any incentive to produce it.
The Platform Is the Easy Purchase. The Estate Underneath Is Not
Every vendor in this category now sells the same promise: describe the workflow, and the platform builds it. MuleSoft LLC, Talend Inc., Microsoft Corp., Google LLC, and Zapier Inc. each shipped a version of that between July 2024 and April 2025, and the differentiation window on it has already closed. When five competitors ship the same capability inside ten months, it stops being a reason to choose any one of them.
What remains unsolved is the thing that made integration hard originally. Systems disagree about what a customer is, records arrive incomplete, and permissions were granted to people rather than to software acting on their behalf. Generating the connection faster resolves none of that. It reaches the disagreement sooner, at a volume no reconciliation team was sized to absorb.
The 34.40% growth rate through 2035 will be earned. The open question is who earns the operating margin behind it, and on current evidence that is the organisations that spent 2026 retiring systems rather than the ones that spent it evaluating platforms. Buying the platform is a procurement decision. Fixing what it connects to is an engineering programme nobody wants to fund.
---
About Kaiso Research and Consulting
Kaiso Research and Consulting is a global market intelligence firm publishing 5,000+ research reports across 11+ industry verticals.
[email protected] | +1 872 219 0417
Dhwani Sharma, Lead Industry Analyst, Kaiso Research and Consulting | Covering enterprise software and integration markets across North America, Europe, and Asia-Pacific
Published: 2026-08-04 | Report Code: IMSS1497
Market Study: Access the full index or request a complimentary sample directly via the AI Integration Platform Market page
