Fabric's 35,000 paid customers are the tell. Microsoft reported that figure in fiscal Q3 2026, up 60% year over year, while data in Fabric OneLake rose nearly 4x (Microsoft earnings, FY2026 Q3). The lakehouse market has moved beyond architecture debate and into the production control layer. The shift from data lakes to active lakehouses is not simply a storage upgrade. It is a capital allocation decision about where enterprise AI gets its memory, governance, latency, and operating context.
The old data lake promise was cheap storage plus future optionality. That worked when the enterprise question was, what happened last quarter? It breaks down when AI agents need to price risk, adjust supply plans, detect fraud, or summarize customer exposure while the facts are still changing. In 2026, the decisive issue is no longer whether an enterprise has enough data. It is whether the data layer can serve trusted, current, governed context into analytical and operational systems at the speed AI workflows require.
The database has become a control system.
The strategic center of gravity is shifting from passive repositories to active lakehouses that combine open table formats, streaming ingestion, semantic governance, vector and graph context, and AI-native query engines. That makes the category strategically important for cloud hyperscalers, independent data platforms, streaming vendors, and systems integrators. It also makes the category politically sensitive inside large enterprises, because active lakehouses sit between CFO cost controls, CTO architecture choices, chief data officer governance mandates, and business-unit pressure to move AI from pilot to production.
For more MarketIntel coverage of enterprise technology spending, see MarketIntel.
$175 Billion Is Only Entry
Gartner estimated the worldwide data and analytics software market at $175 billion in 2025 and forecast it to reach $358 billion in 2029, implying a 15.4% CAGR at constant currency (Gartner Market Opportunity Map, 2025). IDC separately reported global digital transformation software spend on pace for $640 billion in 2026, equal to 32% of total DX spend, with software growing at a 21.7% CAGR and AI reaching roughly 40% of DX software spend by 2029 (IDC Worldwide Digital Transformation Spending Guide, 2026).
Together, those figures frame the TAM: active lakehouses address the data management, analytics, governance, and AI execution layer inside a much larger enterprise software budget.
The nearer SAM is smaller, but it is growing faster. Global Market Insights valued the data lakehouse market at $14.2 billion in 2025 and $17.09 billion in 2026, with a forecast of $105.9 billion by 2034 at a 25.6% CAGR (Global Market Insights, 2025). The Business Research Company placed the 2026 lakehouse market at $12.58 billion and forecast $27.28 billion by 2030 at a 21.4% CAGR (The Business Research Company via Research and Markets, 2026).
The estimates cluster around a 2026 market of roughly $13 billion to $17 billion, with the gap mostly reflecting whether managed services, data governance modules, and adjacent AI workloads are included.
The budget is broad. The contest is narrower.
Real-time analytics adds a second demand pool. Fortune Business Insights valued the real-time analytics market at $1.10 billion in 2025 and $1.37 billion in 2026, rising to $7.54 billion by 2034 at a 25.1% CAGR (Fortune Business Insights, 2026). The Insight Partners used a wider definition, forecasting $12.32 billion in 2025 and $29.27 billion by 2034 at an 11.42% CAGR (The Insight Partners, 2026). The divergence matters because enterprise buyers do not buy a pure category. They fund use cases: fraud scoring, dynamic inventory, industrial telemetry, customer next-best action, and risk monitoring.
North America still leads adoption, with Fortune Business Insights assigning it 36.41% of the real-time analytics market in 2024 (Fortune Business Insights, 2026). Asia-Pacific is the faster-growth region because e-commerce, digital payments, logistics, and telecom data volumes are still compounding from a lower installed base. Europe is slower on experimental AI spend but stronger on governed adoption, because GDPR and the EU AI Act push buyers toward traceable data lineage, policy controls, and audit logs.
Regulation is turning governance into demand.
Six Platforms Want the Control Layer
Databricks is the pure-play benchmark for the active lakehouse thesis. The company said in December 2025 that it had crossed a $4.8 billion revenue run rate, growing more than 55% year over year, with more than $1 billion in run-rate revenue from AI products and more than $1 billion from data warehousing (Databricks press release, 2025). In March 2026, Databricks documentation showed Lakebase availability for compliance-enabled workspaces, while Accenture launched an Accenture Databricks Business Group with more than 25,000 trained professionals focused on Lakebase, Genie, Agent Bricks, and Lakehouse deployments (Databricks documentation, 2026; Accenture, 2026).
Its strategic position is clear: turn the lakehouse into the operating substrate for agents, not just a place to query historical data.
Snowflake is defending from the governed data cloud side. The company reported FY2026 product revenue of $4.47 billion, up 29%, remaining performance obligations of $9.77 billion, and 733 customers with trailing 12-month product revenue above $1 million (Snowflake FY2026 results). In April 2026, Snowflake expanded Snowflake Intelligence and Cortex Code, positioning the platform as a control plane for business users and builders working with enterprise data and AI agents (Snowflake, 2026). Snowflake's strength is executive trust, cross-cloud neutrality, and a consumption model CFOs already understand, but it must keep narrowing the perceived gap with Databricks in open-format engineering and developer-led AI workloads.
Trust is Snowflake's moat. Engineering intensity is Databricks'.
Microsoft is turning Fabric into the lakehouse layer for the Microsoft enterprise estate. In fiscal Q2 2026, Microsoft said Fabric had an annual revenue run rate above $2 billion, over 31,000 customers, and revenue growth of 60% year over year (Microsoft earnings, FY2026 Q2). By fiscal Q3 2026, paid Fabric customers reached 35,000, while more than 15,000 customers used both Foundry and Fabric, up 60% year over year (Microsoft earnings, FY2026 Q3).
The 2026 product move was Fabric IQ and deeper Real-Time Intelligence integration, including geospatial, graph, and semantic layers tied to OneLake, which makes Microsoft the default active lakehouse contender wherever Power BI, Azure, Microsoft 365, and Copilot are already entrenched.
AWS is approaching the same market through S3, Redshift, SageMaker, Glue, and Iceberg. Amazon reported AWS net sales growth of 37% in Q2 2026, equal to a $169 billion annualized revenue run rate (Amazon Q2 2026 results). The key product move was the general availability of Amazon S3 Tables integration with Amazon SageMaker Lakehouse in 2025, which gave customers access to S3 Tables across Athena, EMR, Redshift, Glue, and Iceberg-compatible engines (AWS News Blog, 2025). AWS has the largest cloud infrastructure base, but its challenge is packaging clarity: buyers must see one active lakehouse offer rather than a set of powerful but separately named services.
AWS has gravity. It still needs a cleaner story.
Google Cloud is using BigQuery and Apache Iceberg to attack cross-cloud data estates. Alphabet said Google Cloud revenue grew 82% year over year in Q2 2026 and cloud backlog reached $514 billion (Alphabet earnings, Q2 2026). At Google Cloud Next 2026, Google announced Cross-Cloud Lakehouse, lakehouse catalog federation across AWS Glue, Databricks, Snowflake, and SAP, plus BigQuery Graph, ObjectRef, hybrid search, and AI functions for unstructured data (Google Cloud Next, 2026). Its position is strongest where BigQuery is already the analytics engine and where enterprises want to connect data across clouds without moving every workload into one vendor's storage layer.
Confluent is the streaming bridge into active lakehouses. The company reported 2025 revenue of $1.17 billion, up from $963.6 million in 2024, and Confluent Cloud revenue of $623.6 million, equal to 54% of total revenue (Confluent 10-K, FY2025). Its strategic move was Tableflow, which materializes Kafka topics directly into Apache Iceberg or Delta Lake tables, and Confluent Intelligence, which feeds real-time context into AI systems and agents (Confluent 10-K, FY2025). Confluent does not need to own the whole lakehouse to win. It can tax the motion from event streams into governed analytical tables.
The toll road may be more valuable than the city.
Share is moving toward platforms that can bind open table formats, streaming freshness, security policy, and AI context into one operating model. Databricks is gaining with developers and AI-native teams, Microsoft with installed enterprise buyers, Snowflake with governed analytics budgets, AWS with infrastructure gravity, Google with cross-cloud analytics, and Confluent with real-time ingestion. The mechanism is workload expansion: each vendor starts with storage, BI, streaming, or cloud compute, then moves into agent context, governance, and operational action.
The AI Act Moves Money
The single most concrete trigger in 2026 is the EU AI Act's 2 August 2026 transparency enforcement date. Article 50 obligations apply from that date, and providers and deployers of covered AI systems must inform users when they interact with AI or when content has been generated or altered by AI; fines can reach EUR 15 million or 3% of worldwide turnover (European Commission, 2026). High-risk AI rules have later dates, but the 2026 transparency clock is enough to shift architecture spending now because AI systems need traceable inputs, logs, permissions, and explainable context.
For active lakehouses, the regulation turns data freshness and lineage from engineering preferences into board-level controls. A data lake that stores raw files without fine-grained lineage, semantic meaning, and access enforcement can support research, but it cannot easily support regulated AI workflows that must prove what data was used, which user had access, when the output was generated, and whether the content needed disclosure. A governed lakehouse, by contrast, can centralize policy enforcement across structured records, unstructured documents, embeddings, event streams, and model outputs.
Compliance is becoming an architecture requirement.
The regulation also changes procurement timing. European banks, insurers, healthcare providers, public-sector agencies, and global firms with EU exposure cannot wait until every high-risk AI obligation applies. They need 2026 architecture decisions that will not be rewritten in 2027. That favors platforms with native logs, catalogs, policy inheritance, lineage, and model-facing controls. It hurts do-it-yourself stacks where every control must be stitched together by engineering teams under audit pressure.
Three Risks Look Underpriced
The first risk is cost opacity, with a 55% probability over the next 12 months by analyst estimate. Active lakehouses can reduce data copying, but they also invite new workloads: vector indexing, streaming ingestion, graph queries, agent traces, and unstructured-data processing. Microsoft said cloud gross margin pressure reflected AI infrastructure investment and Azure sales mix in FY2026 Q2 (Microsoft earnings, FY2026 Q2), which shows that the cost curve is real. The affected players are high-growth vendors with consumption pricing and buyers running agent workloads without chargeback discipline. The timeline is short because CFOs will see AI-linked analytics bills within two budget cycles.
The second risk is open-format fragmentation, with a 40% probability over 18 months by analyst estimate. Apache Iceberg is becoming the common table format across AWS, Google, Snowflake, Databricks, and others, but catalogs, governance layers, performance optimizers, and managed services still differ. A buyer may avoid storage lock-in but still face control-plane lock-in. The most exposed players are enterprises trying to run the same governed tables across multiple clouds while preserving consistent permissions, lineage, and performance. The likely timeline is 2026 to 2027, as early cross-cloud lakehouse deployments leave pilot mode.
Open storage does not guarantee open control.
The third risk is governance theatre, with a 45% probability over 24 months by analyst estimate. Vendors now market policy controls, lineage, and agent governance aggressively, but many enterprise deployments still have weak data ownership, stale business definitions, and unclear accountability for model outputs. The affected players are regulated enterprises buying platform features faster than operating models. This shows up as internal audit findings, delayed AI rollouts, and duplicated semantic layers.
The tail risk is a major AI incident caused by stale or poisoned enterprise data rather than model failure. Most boards still focus on model selection, but an agent that reads outdated credit exposure, incorrect supplier status, or manipulated telemetry can make a bad decision even with a strong model. That risk is underweighted because it sits between data engineering, security, and business process ownership.
The model may be innocent. The data may not be.
Buyers Need Workload Discipline
Enterprise buyers should start with workload classification, not vendor demos. Real-time fraud, trading risk, clinical workflow, industrial maintenance, and customer service agents need different latency, lineage, and retention levels. A CFO should require each lakehouse business case to separate storage savings, data engineering productivity, BI consolidation, AI inference support, and compliance value, because bundling them hides weak economics.
CTOs should insist on open table formats but test the control plane. Apache Iceberg compatibility is necessary, but the harder questions are whether row-level permissions, catalog federation, lineage, and audit logs follow the data across engines. Buyers should run a 90-day proof using one operational stream, one unstructured corpus, one BI dashboard, and one agent workflow. If the platform cannot explain access and cost at that small scale, it will not get cleaner at enterprise scale.
A proof of concept should test control, not theatre.
Investors Should Follow Durable Consumption
Investors should separate revenue quality from AI excitement. Databricks' $4.8 billion run rate and Snowflake's $4.47 billion FY2026 product revenue show large demand pools, but the better metric is expansion into governed AI workloads with durable consumption (Databricks, 2025; Snowflake FY2026 results). Investors should watch net retention, million-dollar customer counts, cloud gross margin pressure, and services attach rates. A platform needing heavy implementation services to make agents work may grow bookings but struggle on margin.
Private equity buyers of vertical software companies should treat active lakehouse readiness as a diligence item. Portfolio companies with fragmented customer data, weak event capture, and manual reporting will pay more to add AI later. A clean lakehouse architecture can become a product margin lever if it supports benchmarking, automated insights, and workflow copilots.
AI readiness is becoming balance-sheet diligence.
Vendors Must Sell Decisions
Vendors should stop selling lakehouse architecture as infrastructure modernization. The winning pitch is active decision support: reduce data copies, prove lineage, serve live context, and let AI act inside controlled workflows. Product teams should package cost controls, policy simulation, and agent observability as first-order features, not admin extras. Systems integrators should build repeatable patterns around regulated use cases because EU AI Act compliance, financial-risk explainability, and supply-chain traceability will fund projects faster than generic data platform renewal.
The Next Two Years Decide
The base case, assigned a 60% probability by analyst estimate, is that active lakehouses become the default architecture for new enterprise AI analytics programs by late 2027. In this scenario, lakehouses do not replace every warehouse, streaming platform, or operational database. They become the governed coordination layer where enterprise data is cataloged, joined, enriched, and exposed to BI, AI agents, and operational applications. Databricks, Microsoft, Snowflake, AWS, and Google all grow, while Confluent benefits from streaming demand into those platforms.
The contrarian view, assigned a 25% probability by analyst estimate, is that open lakehouse adoption accelerates faster than vendor consolidation. In this case, buyers standardize on Iceberg tables and independent catalogs, then force cloud and analytics vendors to compete on engines, governance, and AI tooling rather than data gravity. Google Cloud's 2026 Cross-Cloud Lakehouse announcements and AWS S3 Tables show this path is plausible (Google Cloud, 2026; AWS, 2025).
Vendors want gravity. Buyers want leverage.
The downside scenario, assigned a 15% probability by analyst estimate, is that AI workload costs and governance failures slow production adoption. Enterprises may keep pilots alive but restrict agent access to sensitive data until audit teams approve controls. That would benefit vendors with strong governance reputations and hurt vendors whose growth depends on unconstrained consumption.
Leading indicators are specific. Watch the share of Snowflake and Databricks revenue tied to AI products, Microsoft Fabric customer growth beyond 35,000 paid accounts, AWS and Google attach rates for Iceberg-based lakehouse services, and Confluent Tableflow adoption into lakehouse targets. Also watch EU AI Act enforcement actions after 2 August 2026, because the first fines or public investigations will turn governance features into budget requirements.
The first enforcement action will do more than any webinar.
Seven Takeaways for Executives
- Active lakehouses are becoming the execution layer for AI analytics, not just a cheaper way to store enterprise data.
- Gartner's $358 billion 2029 data and analytics software forecast shows the broad budget pool active lakehouse vendors are trying to capture (Gartner, 2025).
- Lakehouse market estimates cluster around roughly $13 billion to $17 billion in 2026, but forecast CAGRs above 21% show faster growth than general enterprise software (Global Market Insights, 2025; The Business Research Company, 2026).
- Microsoft, Databricks, Snowflake, AWS, Google, and Confluent are attacking the same control layer from different starting points.
- The EU AI Act's 2 August 2026 transparency obligations make lineage, logging, and policy enforcement buying criteria, not optional features.
- Open table formats reduce storage lock-in, but catalog, governance, and performance layers can still lock buyers into one vendor's operating model.
- The next major AI failure may come from stale or corrupted enterprise data rather than the model, which puts data architecture back on the board agenda.
Should CFOs Fund Another Lakehouse?
A CFO should fund it only when the project ties to measurable workload outcomes. A passive data lake may store low-cost raw data, but it often lacks governed tables, live ingestion, semantic definitions, and audit-ready access controls. The financial case improves when the active lakehouse replaces duplicated ETL pipelines, reduces warehouse copies, supports real-time analytics, and enables AI products that create revenue or lower operating cost.
Gartner's forecast that data and analytics software reaches $358 billion by 2029 shows the budget pool is large, but that does not justify every project (Gartner, 2025). The CFO should require baseline costs for storage, compute, data engineering labor, BI tools, and compliance effort before approving a migration.
Databricks or Snowflake for AI?
Databricks looks stronger where developer-led AI, open formats, machine learning, and agent infrastructure drive the purchase. Its late-2025 disclosure of more than $4.8 billion in revenue run rate, more than $1 billion from AI products, and thousands of Lakebase customers supports that view (Databricks, 2025). Snowflake looks stronger where governed analytics, procurement trust, and cross-cloud enterprise data sharing matter more.
Snowflake reported $4.47 billion in FY2026 product revenue and $9.77 billion in remaining performance obligations, which gives it scale and visibility (Snowflake FY2026 results). The answer is workload-specific: Databricks for engineering-heavy AI systems, Snowflake for governed business-data activation, with overlap rising fast.
The winner depends on the workload, not the slogan.
How CTOs Avoid Quiet Lock-In
The CTO should distinguish storage openness from operating openness. Apache Iceberg, Delta Lake, and open APIs can reduce the risk of trapped data, but lock-in can reappear in catalogs, access controls, optimizer behavior, lineage, notebooks, and AI governance tools. AWS S3 Tables, Google Cloud Lakehouse, Databricks Unity Catalog, Microsoft OneLake, and Snowflake Polaris-related moves all point toward openness, but each vendor still wants to own the control plane.
A practical test is to run the same governed table through two query engines, one BI tool, one streaming source, and one AI workflow. If permissions, audit history, and performance break outside the primary vendor, the architecture is not as portable as the sales pitch suggests.
The AI Act Rewrites Architecture
The EU AI Act turns data governance into AI infrastructure. Article 50 transparency obligations apply from 2 August 2026, and fines can reach EUR 15 million or 3% of worldwide turnover for relevant breaches (European Commission, 2026). For active lakehouses, the effect is direct: AI systems need to know which data was used, whether content was AI-generated, what access rights applied, and how outputs can be audited. That pushes enterprises toward unified catalogs, automated lineage, model and agent logs, and policy controls across structured and unstructured data. Companies such as Microsoft, Snowflake, Databricks, and Google are making semantic and governance layers central because regulated AI needs proof, not just performance.
Where Investors Can Still Win
The best returns are likely in control points, not generic storage. Hyperscalers will capture infrastructure spend, and large platforms such as Microsoft, Databricks, Snowflake, AWS, and Google will capture the main lakehouse budgets. Specialist opportunities sit in streaming integration, governance automation, semantic modeling, cost controls, data quality, and agent observability. Confluent is a public example: it reported $1.17 billion in 2025 revenue and positioned Tableflow as a way to materialize Kafka streams into Iceberg or Delta Lake tables (Confluent 10-K, FY2025). Private investors should look for vendors that make active lakehouses safer to run, cheaper to monitor, or easier to connect to real operational systems.
The Winning Architecture Is Active
The shift from data lakes to active lakehouses is a response to a practical constraint: AI cannot act well on stale, fragmented, poorly governed data. Enterprises spent the past decade centralizing data for analytics, then spent the past two years discovering that AI needs a more demanding substrate. It needs fresh events, historical context, unstructured documents, semantic meaning, permissions, lineage, and cost controls in one governed flow.
The market will not resolve into one winner. Microsoft will win inside Microsoft-centered estates, Databricks will win many AI-native and engineering-heavy programs, Snowflake will remain strong in governed enterprise analytics, AWS will pull lakehouse workloads toward S3 and SageMaker, Google will press cross-cloud BigQuery-led architectures, and Confluent will benefit wherever real-time streams feed analytical and AI systems.
The buyer's task is to avoid being impressed by architecture diagrams and instead test whether the platform can answer four questions under load: what changed, who can see it, what did the AI use, and what did it cost?
By December 2027, more than half of new Fortune 1000 enterprise AI analytics programs will specify an active lakehouse architecture with streaming ingestion, open table formats, governed semantic context, and agent audit logs as mandatory design requirements.
--- [✉️ Talk to us](https://marketintel.co.in/contact) · [Subscribe to our Newsletter](https://marketintel.co.in/newsletter) --- ### Related Articles - [Why Corporate Boards Are Wrong About Supply Chain Resilience](why-corporate-boards-are-wrong-about-supply-chain-resilience-baea20a2) - [AlphaSense Hitting $400 Million ARR Will Kill Dashboard Theater](why-market-intelligence-agents-will-kill-dashboard-theater-829f0a82) - [Answer Engines Capture The First B2B AI Search Shortlist](stop-treating-your-website-as-b2b-discoverys-front-door-b6b46820)See our Gartner research for deeper analysis.