Databricks has overtaken Snowflake in revenue scale and growth rate, reaching $5.4B annualized revenue versus Snowflake's $4.5B product revenue , while Microsoft Fabric has emerged as the fastest-growing analytics product in Microsoft's history with 28,000+ paying customers in just two years . The competitive picture has shifted dramatically since 2021. AI workloads now drive platform differentiation and revenue growth. Apache Iceberg has decisively won the table format wars, commoditizing the storage layer and shifting competition to compute, governance, and AI capabilities . For Norwegian enterprises, the combination of Schrems III uncertainty , the EU AI Act, and local data sovereignty requirements makes platform selection more consequential than ever.
| Databricks ARR | Snowflake Revenue | Fabric Customers | Iceberg Adoption | Schrems III | DuckDB Downloads |
|---|---|---|---|---|---|
| $5.4B | $4.5B | 28K+ | 78.6% | Pending | 25M |
| +65% YoY · $134B val. | +29% YoY · public | +55% rev growth | exclusive format share | ECJ appeal Oct 2025 | monthly PyPI |
Platform Deep Dives
Snowflake: the SQL-first data cloud matures into AI
Financial trajectory and growth metrics
remains the largest pure-play cloud data platform by public-market presence. Its fiscal year ends January 31, so FY2026 covers the period through January 2026. All figures below come from Snowflake's SEC filings and quarterly earnings releases .
| Metric | FY2022 | FY2023 | FY2024 | FY2025 | FY2026 |
|---|---|---|---|---|---|
| Product revenue | $1,141M | $1,939M | $2,670M | $3,462M | $4,472M |
| Product revenue YoY growth | ~106% | ~70% | ~38% | ~30% | ~29% |
| Total customers | 5,944 | 7,828 | ~9,400 | ~11,000 | ~13,300 |
| Customers >$1M TTM rev. | 184 | 330 | 461 | 580 | 733 |
| Customers >$10M TTM rev. | N/R | N/R | N/R | N/R | 56 |
| Net revenue retention | 178% | 158% | 131% | 126% | 125% |
| RPO | $2.6B | $3.7B | $5.2B | $6.9B | $9.8B |
| Non-GAAP product gross margin | ~75% | ~75% | ~75% | ~76% | ~76% |
| Non-GAAP operating margin | ~1% | ~5% | ~6% | ~6% | 10.5% |
Sources: Snowflake SEC filings and quarterly earnings releases (FY2022–FY2026).
NRR compression
Net revenue retention fell from 178% to 125% over four years. While 125% remains healthy by SaaS standards, the era of explosive within-account expansion is moderating. Average annual spend per customer has risen from ~$192K (FY2022) to ~$336K (FY2026), driven by workload expansion into AI, unstructured data, and streaming, not by price inflation.
For FY2027 (ending January 2027), Snowflake guided to $5.66B in product revenue (27% growth), with non- operating margin expanding to 12.5% . Analyst consensus points to roughly $6.7–7.1B in total revenue by FY2028. The company's original $10B product revenue target for FY2029 appears ambitious but within range if AI monetization accelerates.
Snowflake remains GAAP unprofitable with an estimated FY2026 operating loss exceeding $1B, though the trajectory is improving. adjusted free cash flow margin held at approximately 25%.
Price/performance and value analysis
Snowflake's consumption-based credit model charges $2–6+ per credit depending on edition and cloud provider, with compute typically representing 80% of a customer's bill. Key pricing moves in 2024–2025 include simplified Snowpipe pricing (flat credit-per-GB model), native table support (allowing customers to store data in open formats on their own cloud storage, reducing storage costs by 80–90%), and the launch of Cortex Code as Snowflake's first standalone subscription product untied to compute consumption.
Customer sentiment on cost is mixed. Snowflake is widely perceived as premium-priced, and a cottage industry of tools (Select.dev, Keebo, ChaosGenius) exists specifically to optimize Snowflake spend, with customers reporting 20–30% savings. Time Travel and Fail-safe features silently duplicate storage, creating hidden cost drivers. However, Snowflake's zero-management simplicity means total cost of ownership comparisons depend heavily on workload type. For sporadic / queries, Snowflake often delivers better economics than alternatives.
Platform strategy and AI pivot
Under CEO Sridhar Ramaswamy (appointed February 2024, formerly of Google), Snowflake has executed a decisive AI-first pivot. The Cortex AI platform provides access to from , , Meta, and , callable directly from SQL. Key capabilities include AI functions (classify, transcribe, translate, extract), Cortex Search for over unstructured data, and for natural-language-to-SQL, which Snowflake reports at over 90% SQL accuracy on its internal .
Snowflake Intelligence, launched in public preview in August 2025, is an agentic AI framework for conversational analytics. Snowflake's Q4 FY2026 commentary describes thousands of accounts adopting Intelligence and its broader AI offerings . The company also signed separate $200M multi-year deals with both OpenAI and Anthropic for model integration; the Anthropic agreement makes available to more than 12,600 Snowflake customers.
On the open data front, Snowflake's Iceberg support reached GA in May 2024, and the company open-sourced the (now an Apache incubating project) as a vendor-neutral catalog for Iceberg. Snowflake's January 2026 agreement to acquire Observe (reported at roughly $1B, its largest acquisition to date) targets the IT operations and AI-observability market, extending the platform beyond analytics .
| Strengths | Weaknesses |
|---|---|
| Best-in-class simplicity (zero infrastructure management), true multi-cloud portability (AWS/Azure/GCP), industry-leading data sharing, rapidly maturing AI capabilities, large installed base of 13,300+ customers. | Still GAAP unprofitable, widely perceived as expensive, AI/ML capabilities less mature than Databricks, NRR compression continues, and the mid-2024 customer-account breach campaign (tracked by Mandiant as UNC5537) hit roughly 165 organizations whose Snowflake accounts lacked MFA. |
Nordic presence
Snowflake serves Nordic customers through EU North (Stockholm), Sweden Central, and GCP Finland. No dedicated Norwegian region exists, so Norwegian organizations typically use Stockholm-based regions. Known Nordic customers include Tele2 (Sweden). The partner ecosystem includes Solita (Finland, Snowflake Elite Partner and 2024 Nordic Growth Partner of the Year), Crayon/Inmeta (Norway, one of the first Snowflake partners in the Nordics), Avito (part of View Group), and Capgemini Norway. An active Snowflake User Group operates in Oslo.
Databricks: the lakehouse juggernaut accelerates
Financial trajectory and growth metrics
is private, so financials come from company press releases (typically tied to funding rounds) and analyst estimates. The trajectory is extraordinary. Growth has accelerated at scale, a rare feat in enterprise software.
| Period | Annualized Revenue | YoY Growth | Key Event |
|---|---|---|---|
| End 2020 | ~$425M | — | — |
| End 2021 | ~$800M | ~80% | Series H ($38B valuation) |
| Mid-2023 | $1.5B | ~50% | Series I ($43B valuation) |
| Jan 2024 | $1.6B trailing | >50% | — |
| Mid-2024 | $2.4B run rate | 60% | Accelerating growth |
| Dec 2024 | $3.0B run rate | >60% | Series J ($62B, $10B raised) |
| Mid-2025 | $3.7B run rate | 50% | — |
| Sep 2025 | $4.0B+ run rate | >50% | Series K ($100B+) |
| Dec 2025 | $4.8B run rate | >55% | Series L ($134B, $5B raised) |
| Jan 2026 | $5.4B annualized | >65% | Latest disclosure |
Sources: Databricks press releases (Series J Dec 2024, Series L Dec 2025, Feb 2026); CNBC; TechCrunch; Sacra (Feb 2026). Company-disclosed annualized run rates, not audited annual revenue.
| Customers | >$1M ARR | NRR | AI Revenue |
|---|---|---|---|
| 20K+ | 700+ | >140% | $1.4B |
| Dec 2025 | Dec 2025 | sustained 2024–25 | ~26% of total |
The most striking figure is the >140% , which significantly outpaces Snowflake's 125%. Databricks has surpassed Snowflake in revenue scale, $5.4B annualized versus Snowflake's $4.5B product revenue, while growing at more than double the rate . Databricks reported it crossed the $4.8B run-rate growing >55% YoY at its Series L in December 2025, then disclosed a $5.4B run-rate at >65% YoY in February 2026 . Non-GAAP subscription gross margin exceeds 80%.
Analyst projections (speculative) suggest Databricks could reach $10B by 2028. The December 2025 Series L raised >$4B at a $134B valuation . CEO Ali Ghodsi has kept an on the table, but no S-1 has been filed as of March 2026.
Gross margins have declined from an estimated ~85% (mid-2023) to ~74–80% (late 2025), attributed to AI compute cost absorption. GPU-intensive workloads carry different cost structures than SQL analytics.
Price/performance and value analysis
Databricks charges in (DBUs): Jobs Compute at $0.15–0.30/DBU (cheapest), All-Purpose Compute at $0.40–0.55/DBU, and Serverless SQL at $0.70/DBU (includes infrastructure). Unlike Snowflake's all-inclusive credit model, Databricks traditionally billed DBUs separately from the underlying cloud VMs, creating dual-bill complexity that complicates analysis. Serverless compute, rolled out broadly in 2024–2025, bundles infrastructure costs into a single price at a premium.
Key pricing moves include the phase-out of the Standard tier (auto-upgraded to Premium by October 2025 on AWS/GCP), 30% promotional discounts on serverless (May 2024), and commitment-based discounts of up to 37% for 3-year Azure commitments.
Market consensus: Databricks is cheaper for heavy data engineering, streaming, and workloads (with spot instance savings up to 90%), while Snowflake is more cost-effective for sporadic interactive SQL and BI. Databricks' engine delivers multi-x speedups over open-source for SQL workloads, and Databricks SQL set an official TPC-DS 100TB world record: 2.2x the prior Alibaba result .
Platform strategy and AI dominance
Databricks' strategic position rests on three pillars:
The architecture ( flexibility + warehouse reliability) is now industry-accepted as the default. provides ACID transactions on cloud object storage, and is the centralized governance layer.
The AI/ML platform is Databricks' most significant competitive advantage. After acquiring MosaicML for ~$1.3B in 2023, Databricks disclosed a $1B+ from AI products (alongside a $1B+ data-warehousing run rate) at its December 2025 Series L ; analyst estimates put AI nearer $1.4B (~26% of total) against the later $5.4B run rate . The suite covers training, (Composer), serving, vector search, AI Gateway, and Agent Bricks. MLflow, created by Databricks, is the industry-standard open-source ML lifecycle tool.
The platform expansion into operational databases via the acquisition (~$1B, May 2025) brought serverless Postgres (branded Lakebase), targeting and 's $100B+ market . Databricks also acquired Tabular (founded by the original creators of Apache Iceberg) in 2024 for a reported $1–2B, giving it ownership of expertise behind both major open table formats .
| Strengths | Weaknesses |
|---|---|
| Fastest-growing infrastructure company at scale, AI-native positioning with $1.4B AI revenue, open-source philosophy (Delta Lake, MLflow, Unity Catalog), 140%+ NRR, multi-cloud flexibility, aggressive acquisition strategy expanding TAM. | Still private with no audited financials, declining gross margins (AI compute costs), complex dual-billing pricing, steeper learning curve for non-technical users, $134B valuation demands sustained exceptional execution, some AI features not yet in all European regions. |
Nordic presence
Databricks opened a Stockholm office (November 2023) and an Aarhus, Denmark office (July 2024), with 100+ employees across the Nordics. Eva Fors (former Nordic MD) was appointed VP Benelux & Nordics in September 2025. Named Nordic customers include Swedbank, IKEA, KONE, Nokia, Pandora, and DSV. No Norwegian customers are publicly named, but the active Oslo Databricks User Group (run by Avanade Norway) suggests local adoption, and and are natural fits for Azure Databricks. A 2022 Validio survey of Nordic scale-ups found dominant, followed by Snowflake, with Databricks growing but lower-penetration.
Microsoft Fabric: the ecosystem play enters full force
Financial trajectory and adoption metrics
Microsoft does not break out Fabric-specific revenue, bundling it within Intelligent Cloud. Azure revenue reached $75B in FY2025 (ending June 2025), growing 34% YoY, the first time Microsoft disclosed absolute Azure revenue . AI services contributed 10+ percentage points of Azure growth.
| Date | Paying Customers | Milestone |
|---|---|---|
| Oct 2023 (GA+1 month) | 16,000 | 50%+ of Fortune 500 |
| May 2025 (Build) | 21,000 | 70% of Fortune 500 |
| Mid-2025 (FabCon) | 25,000+ | 80% of Fortune 500 |
| Dec 2025 | 28,000+ | "Fastest-growing analytics product in Microsoft history" |
Sources [R6, R7]: Satya Nadella earnings calls; Microsoft Fabric Blog. Microsoft cites Fabric as its fastest-growing analytics product; by its Q2 FY2026 earnings the count had reached 31,000+. Absolute Fabric revenue not disclosed.
| Revenue Growth | Certified Consultants | Certifications |
|---|---|---|
| 55% | 135K | 50K+ |
| FY2025 YoY | 6,000+ partner orgs | in under 18 months |
The ecosystem scale is massive: 135,000+ trained consultants, 6,000+ partner organizations, and 50,000+ certifications in under 18 months. Accenture alone has 3,000+ certified Fabric specialists. Projecting forward, Fabric could reach 40,000–50,000+ paying customers by 2027–2028.
Pricing and value analysis
Fabric uses capacity-based pricing measured in Capacity Units (CUs): a single compute pool shared across all workloads. This differs from Snowflake's per-workload credit model.
| SKU | CUs | ~Monthly PAYG (US West) | Equivalent |
|---|---|---|---|
| F2 | 2 | ~$262 | Entry level |
| F8 | 8 | ~$1,050 | Small team |
| F64 | 64 | ~$8,396 | = P1 (Power BI Premium) |
| F128 | 128 | ~$16,793 | = P2 |
| F1024 | 1024 | ~$134,000 | Large enterprise |
Source: Microsoft Fabric pricing documentation. Reserved capacity saves ~40.5% vs. PAYG.
per-capacity P-SKUs were retired in early 2025 (new sales stopped July 1, 2024; non-EA renewals ended February 1, 2025), migrating customers to Fabric F-SKUs, a powerful adoption funnel . Organizations that previously paid for Power BI Premium now receive Data Engineering, Data Science, Data Warehouse, Real-Time Intelligence, and Data Activator as part of the same capacity. For organizations consolidating from separate Power BI Premium + Synapse + Data Factory, moving to a Fabric F- almost always reduces total license spend.
Fabric pricing is more opaque than Snowflake's. Predicting CU consumption across mixed workloads is harder. However, the "smoothing" mechanism (averaging burst compute over 24-hour windows for background workloads) provides some cost predictability.
Platform strategy and the all-in-one bet
The Synapse-to-Fabric transition is largely complete. Data Explorer was retired in October 2025; Synapse still receives security patches but Fabric is the clear strategic direction.
Migration case studies
OBOS (Norway) migrated 600+ notebooks and 4,500+ data objects into 87 Fabric lakehouses, reporting 20% lower operating costs and 30% faster processing for 450+ users . Kantar cut costs 50% versus Synapse while 4x-ing development throughput .
Fabric's core architectural bet is : a unified data lake built on using Delta Lake/ natively. OneLake supports "shortcuts" to external storage (, ) without duplication. lets Power BI read directly from OneLake's Delta Parquet files without imports.
For AI, Microsoft's strategy is the broadest of the three: Copilot in Fabric (powered by /-5) is embedded across all workloads. (preview February 2026) understand enterprise data schemas and enforce governance. Integration with and Studio enables multi-agent orchestration across Fabric, M365, and Azure AI, an integration surface unmatched by Snowflake or Databricks.
Fabric's open format support includes native Delta Lake and Iceberg interoperability via metadata virtualization. OneLake auto-translates between Delta and Iceberg metadata bidirectionally using Apache XTable. A Snowflake–Fabric bridge reached GA in early 2026, enabling Snowflake to write Iceberg tables directly into OneLake.
| Strengths | Weaknesses |
|---|---|
| Unified SaaS platform eliminating tool sprawl, tight Microsoft ecosystem integration (Power BI, M365, Teams, Azure), OneLake as single source of truth, Power BI natively included, simplified pricing, AI Copilot deeply integrated, strong governance via Purview, 28,000+ customers in two years. | Azure-centric (not multi-cloud), less proven at very large-scale data engineering versus Databricks, Spark performance not matching Photon, pricing opacity for mixed workloads, platform still maturing (many features in preview), potential overpayment for data-processing-only use cases. |
Nordic presence: Microsoft's home-field advantage
Microsoft has the strongest Nordic infrastructure position. Azure Norway East (Oslo) launched in 2019 and now offers 3 Availability Zones. Microsoft was the first global cloud provider with enterprise-grade services in Norway . Azure Norway West (Stavanger) is a DR region (restricted access since 2022). Major Norwegian customers include DNB Group (Norway's largest bank, "data residency was a decisive factor"), Equinor, Lånekassen, Posten Norge, and OBOS. Norway has one of the strongest Microsoft partner communities per capita globally. Microsoft Cloud for Sovereignty (GA late 2023) provides additional sovereignty controls for public sector.
Comparative Financial Data
Snowflake figures from SEC filings (FY = Jan 31). Databricks figures are company-disclosed annualized run rates (not audited). Fabric revenue is not disclosed by Microsoft.
Revenue / ARR trajectory
| Year | Snowflake Product Rev. | Databricks ARR (est.) | Fabric Revenue | Notes |
|---|---|---|---|---|
| 2021 | ~$1.1 (FY2022) | ~$0.8 | N/A (pre-launch) | Snowflake FY = Jan 31 |
| 2022 | ~$1.9 (FY2023) | ~$1.0–1.1 | N/A (pre-launch) | |
| 2023 | ~$2.7 (FY2024) | ~$1.5–1.6 | Launched Nov 2023 | |
| 2024 | ~$3.5 (FY2025) | ~$2.4–3.0 | Not disclosed | |
| 2025 | ~$4.5 (FY2026) | ~$4.0–5.4 | Not disclosed (55% growth) | Databricks surpasses Snowflake |
| 2026E | ~$5.7 (FY2027 guided) | ~$7–8 (est.) | Not disclosed | |
| 2027E | ~$6.7–7.1 (analyst est.) | ~$9–11 (est.) | Not disclosed | |
| 2028E | ~$8–9 (extrapolated) | ~$10+ (analyst est.) | Not disclosed |
Total customer count
| Year (approx.) | Snowflake Total | Snowflake >$1M | Databricks Total | Databricks >$1M |
|---|---|---|---|---|
| 2021 | 5,944 | 184 | ~5,000 | N/A |
| 2022 | 7,828 | 330 | ~7,000 (est.) | N/A |
| 2023 | ~9,400 | 461 | ~9,000 (est.) | ~220 |
| 2024 | ~11,000 | 580 | ~15,000 | N/A |
| 2025 | ~13,300 | 733 | 20,000+ | 700+ |
Net revenue retention
| Year | Snowflake NRR | Databricks NRR (est.) |
|---|---|---|
| 2021 | 178% | >150% (est.) |
| 2022 | 158% | ~150% (est.) |
| 2023 | 131% | >140% |
| 2024 | 126% | >140% |
| 2025 | 125% | >140% |
Average annual spend per customer
| Year | Snowflake (derived) | Databricks (derived) |
|---|---|---|
| 2021 | ~$192K | ~$160K |
| 2022 | ~$248K | ~$150K (est.) |
| 2023 | ~$284K | ~$170K (est.) |
| 2024 | ~$315K | ~$200K |
| 2025 | ~$336K | ~$240–270K |
Snowflake derived from product revenue ÷ customer count. Databricks estimates from Sacra (ACV ~$209K early 2024) and derived data.
Year-over-year revenue growth
Competitive Scorecard: 2025
| Snowflake | Databricks | Microsoft Fabric | |
|---|---|---|---|
| Revenue / ARR | $4.5B | $5.4B | Not disclosed |
| YoY Growth | 29% | 65% | 55% |
| Total Customers | 13,300 | 20,000+ | 28,000+ |
| >$1M Customers | 733 | 700+ | N/A |
| Net Revenue Retention | 125% | >140% | N/A |
| Avg Annual Spend | ~$336K | ~$255K | N/A |
| AI Revenue | Growing (early) | $1.4B (26%) | Bundled in Azure AI |
| Multi-Cloud | AWS · Azure · GCP | AWS · Azure · GCP | Azure only |
| Norway Data Region | No (Stockholm) | No (Stockholm) | Yes (Oslo, 3 AZs) |
| Open Format Strategy | Iceberg native + Polaris | Delta + UniForm + Iceberg | Delta + Iceberg (XTable) |
| Verdict | Holding | Winning | Gaining fast |
Market positioning: growth vs. revenue scale
Emerging Disruptors
DuckDB and MotherDuck: eroding cloud warehouse economics from below
(an in-process database often described as " for analytics") has become the most consequential new database technology since the cloud data warehouse itself. It runs as an embedded library inside , R, Java, or a browser (via WebAssembly), requires zero infrastructure, and speaks -compatible SQL.
| PyPI Downloads | GitHub Stars | Stack Overflow | Fortune 100 |
|---|---|---|---|
| 25M+/mo | 30K+ | 3.3% (up from 1.4%) | 20+ |
PyPI downloads passed ~20M/month by mid-2025 and ~37M/month by 2026; GitHub crossed 30K stars in June 2025 [R23].
DuckDB handles most analytical queries on datasets under 100GB (the vast majority of enterprise workloads) for a fraction of cloud warehouse costs. Definite migrated its data warehouse entirely from Snowflake to DuckDB, reporting a >70% cost reduction . Headset cut its Snowflake BI costs by 83% while supporting 2,500+ users by pushing query workloads onto DuckDB .
Cost comparison, 1TB data, 10–20 analysts
Snowflake + BI stack costs $3,500–10,000/month versus DuckDB + BI at $250–500/month. DuckDB queries return in 200–400ms where cloud warehouses take 2–5 seconds including cold start.
(founded 2022 by Jordan Tigani, founding BigQuery engineer) commercializes DuckDB as a serverless cloud service with a unique "hybrid execution" model: a dual-execution planner routes queries between local and cloud DuckDB automatically. MotherDuck has raised roughly $133M in total (including a $52.5M Series B at a $400M valuation in September 2023 ). Its integration ecosystem has expanded to , , and .
ClickHouse Cloud: capturing the real-time analytics layer
, the open-source columnar OLAP database originally developed at Yandex, has carved out a distinct position in real-time analytics that cloud data warehouses struggle to match.
| ARR (est.) | ARR Growth | Valuation | Customers |
|---|---|---|---|
| $88–96M | 250%+ | ~$15B | 3,000+ |
ClickHouse wins for sub-second queries on streaming/event data (observability, security analytics, clickstream, IoT), use cases where Snowflake or BigQuery introduce minutes of latency. Funding accelerated: a $350M Series C (2025, $6.35B valuation) followed by a $400M Series D led by Dragoneer (January 16, 2026, ~$15B valuation) . Its acquisition strategy signals expansion into AI observability: it acquired , the open-source LLM-observability project, alongside the Series D . It positions itself as "analytics for the AI era," arguing AI generate far higher query volumes than humans.
Other players worth tracking
| Player | Funding / Valuation | Position |
|---|---|---|
| Dremio | $400M+ raised, $2B val. | Lakehouse on Apache Arrow; #1 engine for Iceberg; ranked #1 in Dresner's 2025 study |
| Starburst | $414M raised, $3.35B peak | Built on Trino; federated queries across cloud and on-prem |
| StarRocks | Linux Foundation project | MPP analytics; traction at Pinterest, Airbnb, Coinbase; complex joins & real-time upserts |
| Apache Doris | 4,000+ enterprises | Dominates China (80% of top 50 internet cos.); limited Western presence |
| Firebolt | $269M raised, $1.4B val. | Quiet since Jan 2022: 3+ year funding gap raises questions |
Open Table Formats: Iceberg Has Won
Apache Iceberg has decisively won the table format wars. The State of the Apache Iceberg Ecosystem survey (fielded January 2026) found 78.6% of respondents use Iceberg exclusively among open table formats, with Delta Lake showing meaningful overlap at 39.3% . Every major platform now supports it: Snowflake natively, Databricks via UniForm and native Iceberg managed tables, via metadata virtualization, plus AWS (Athena, Glue, S3 Tables), , Cloudera, and others.
The evidence is structural. Databricks (creator of Delta Lake) acquired Tabular (founded by Iceberg's creators) for a reported $1–2B in 2024 and now offers native Iceberg managed tables alongside Delta . UniForm auto-generates Iceberg metadata from Delta tables. Fabric auto-translates bidirectionally. Even Snowflake has made Iceberg its primary open format strategy, with Polaris Catalog (now Apache incubating) as its open-source REST catalog.
| Iceberg Exclusive Use | On-Prem Iceberg | Redundant Data Cost |
|---|---|---|
| 78.6% | 28.6% | $2–7M |
| Jan 2026 survey | deploying on-premises | avg Fortune 500 annual waste |
Why this matters
Iceberg commoditizes the storage layer. Data stored once in Iceberg on S3, ADLS, or GCS can be queried by any compatible engine: Snowflake, Databricks, Trino, DuckDB, Flink, or others. The average Fortune 500 company reportedly spends $2–7M annually on redundant data storage and transfer; Iceberg standardization could reduce data engineering workloads by ~30% (AnalyticsWeek).
Competition is shifting to the compute and governance layers. The new battleground is catalogs: who controls metadata and access policies. remains the common default catalog on AWS, while Apache Polaris (Snowflake) and Unity Catalog (Databricks) both position as vendor-neutral, REST-compatible options .
Practical advice for Norwegian enterprises
Stop storing data in proprietary formats, adopt Iceberg as the primary storage format, and choose compute engines based on workload fit rather than data lock-in.
AI/ML Convergence: the Data Platform Becomes the AI Platform
The competitive picture has transformed since 2023. Three years ago, AI/ML workloads were largely separate from . Today, all three platforms have made AI a first-class capability. AI is now the primary driver of revenue growth and differentiation.
- Databricks has the strongest end-to-end AI story for technical teams. $1.4B annualized AI revenue (~26% of total) dwarfs competitors. The Spark foundation provides native support for , a capability Snowflake lacks.
- Snowflake has the strongest SQL-first AI adoption. Cortex functions callable from SQL enable analysts without Python skills, broadening AI access to a much larger user base than Databricks' code-first model.
- Microsoft has the broadest enterprise AI integration surface. Only Microsoft connects the data platform (Fabric), AI platform (Azure AI ), developer platform (/), and productivity suite (M365 Copilot) into one stack.
The "data platform is the AI platform" narrative is winning for , analytics-adjacent AI, and governed enterprise applications: RAG, NL2SQL, classification, summarization. Menlo Ventures' 2025 enterprise survey put annual enterprise AI spend at $37B, with the data and model-serving layers a major share of that growth . For model training and frontier research, separate specialized infrastructure remains common.
MIT's 2025 State of AI in Business study found that 95% of enterprise generative-AI pilots delivered no measurable P&L impact , a signal the market is still early. Platform selection for AI should weigh production-readiness and governance, not just model access.
Data Sovereignty, Regulation & the Norwegian Context
Cloud repatriation is selective, not wholesale
Flexera's 2025 State of the Cloud report found 21% of workloads have already been repatriated , while a Barclays CIO Survey found 83% of CIOs planned to move some public cloud workloads back, up from 43% in 2021 . However, net cloud migration continues to outpace exits; most repatriation is partial/hybrid. Gartner frames the European trend as "geopatriation": moving from a US hyperscaler to a local/regional provider, not back to . 61% of Western European CIOs say geopolitical factors will increase reliance on local providers; 53% say geopolitics will restrict future use of global cloud providers.
Schrems III looms over every US cloud decision
The (DPF), established July 2023, survived its first legal challenge when the EU General Court dismissed the Latombe action on September 3, 2025. But Latombe appealed to the Court of Justice on October 31, 2025 (Case C-703/25 P), so the framework is not yet settled . Multiple factors could still trigger a "" invalidation: the Trump administration's dismantling of the PCLOB, removal of Democrats from the FTC, and potential revocation of Biden's Executive Order 14086 that underpins the DPF.
The US remains the structural concern. Microsoft cannot guarantee that EU data will never be requested by US authorities, as acknowledged at a French parliamentary hearing in June 2025.
Sovereign cloud offerings: expanding but incomplete
All three hyperscalers now offer European sovereign products. Microsoft's Sovereign Public Cloud (GA 2025) keeps data in Europe, under European law, operated by European-resident personnel, with customer-controlled keys. AWS European Sovereign Cloud (announced June 2025) features independent European governance and a dedicated EU parent company; Snowflake plans to deploy on it. Databricks runs on the customer's own cloud account, inherently supporting sovereign configurations.
Sovereignty simulated, not guaranteed
Critics note US providers' sovereign offerings are operational compliance measures that cannot overcome structural US CLOUD Act jurisdiction. National partner clouds like Bleu (France, Orange/Capgemini JV) and Delos Cloud (Germany, SAP subsidiary) create legally independent European entities.
Self-hosted open-source: viable but demanding
A self-hosted stack ( + Spark + + Iceberg + /dbt) is technically viable, with 28.6% of Iceberg respondents deploying on-premises. Proponents claim 3–5x TCO savings for predictable workloads. But the operational burden is significant. And no European cloud provider offers a fully managed data platform on par with Databricks, Snowflake, or BigQuery, the strongest argument for continued US hyperscaler use despite sovereignty concerns.
Norwegian regulatory context
Norway implements through membership, enforced by Datatilsynet (whose NOK 65M Grindr penalty, roughly €5.5–6.5M, was upheld by the Borgarting Court of Appeal in October 2025) . Key regulations: Digital Security Act (effective October 2025, first cross-sector digital security law), (via Norwegian DORA Act, May 2025, covering financial entities), National Security Act (2018). Norway still operates under NIS1 (not yet NIS2) as of late 2025.
Bottom line for Norwegian regulated sectors (finance, health, energy, public)
The combination of CLOUD Act exposure, Schrems III uncertainty, and sector-specific regulations means platform selection must be evaluated through a legal and sovereignty lens first, before capability comparison. Azure Norway East (Oslo) with 3 AZs provides the strongest in-country data residency. A hybrid approach (on-prem / Norwegian-hosted for the most sensitive data, cloud for less regulated workloads) is the pragmatic path.
The DuckDB Phenomenon & Cloud Economics
DuckDB's rise signals a structural challenge to the cloud data warehouse business model. The core insight: the median analytical dataset is measured in gigabytes, not terabytes, and modern hardware can process these workloads locally in milliseconds.
Enterprise adoption pattern: development and testing (eliminates credit burn during iteration); BI caching layer (sub-200ms vs 2–5s cold starts, Headset's 83% BI cost reduction ); embedded customer-facing analytics via WebAssembly; direct replacement for sub-100GB datasets (Definite's full migration ).
Business model implications: (1) bottom-of-market erosion: DuckDB/MotherDuck capture sub-TB workloads that currently pay cloud prices, the bulk of the market by customer count; (2) development tier elimination: cloud warehouses lose testing revenue; (3) pricing pressure: cloud warehouses must compete on per-query economics. Because DuckDB reads Parquet and Iceberg natively from object storage, it reinforces multi-engine architectures on shared open data.
What the 2026 numbers add up to
Who is winning and who is losing
Databricks: Winning
Overtaken Snowflake in revenue ($5.4B vs $4.5B), growing 2x+ faster (65% vs 29%), higher NRR (>140% vs 125%), strongest AI position ($1.4B). Open-source philosophy and lakehouse architecture are industry standard. Primary risk: execution at $134B valuation with declining gross margins.
Snowflake: Holding but losing relative position
Growth stabilized at ~29%, respectable but pedestrian vs. its 106% in FY2022 or Databricks' 65%. NRR compressed 178%→125%. AI pivot is promising (9,100+ accounts) but playing catch-up on ML. Core strengths (simplicity, multi-cloud, data sharing) remain valuable but no longer differentiating.
Fabric: Gaining share rapidly
16K→28K customers in two years via ecosystem gravity, not technical superiority. Compelling bundled value for Microsoft-centric organizations (most Norwegian enterprises). Weakness: Azure lock-in and relative immaturity.
ClickHouse is winning the real-time analytics layer (250%+ growth, $15B valuation).
DuckDB is eroding the bottom of the market by proving most analytical queries don't need cloud infrastructure at all.
The deliberately neutral counterpart, on whether your workload actually needs a platform at all and what fits each operating model:
Three structural shifts
- AI has become the primary axis of platform competition. Platform selection in 2026 is increasingly an AI infrastructure decision disguised as a data platform decision.
- Open table formats (Iceberg) have commoditized storage and reduced switching costs. Vendor lock-in now depends on compute features, governance, and ecosystem integrations. Multi-engine architectures become viable, weakening the case for a single platform.
- Data sovereignty has moved from a compliance checkbox to a strategic constraint. Schrems III, the , DORA, and EUCS requirements mean where data is processed and by whom may override all other technical considerations.
What a Norwegian enterprise should prioritize in 2026
Recommendations
- Evaluate sovereignty requirements first. Determine which data categories require Norwegian/EEA residency, which can use hyperscaler sovereign clouds, and which must remain on-premises. Azure Norway East gives Microsoft an inherent advantage for the most sensitive workloads.
- Adopt Iceberg as the storage standard. Regardless of compute platform, Iceberg on cloud object storage (or on-prem S3-compatible storage) preserves optionality and eliminates lock-in.
- Choose compute by team profile. SQL/BI-heavy → Snowflake or Fabric. Data science/ML → Databricks. Microsoft-standardized → Fabric. Many large enterprises run two platforms (Databricks for engineering/ML + Snowflake or Fabric for analytics), reasonable given Iceberg interoperability.
- Budget for DuckDB in the dev workflow. Even cloud-committed orgs should evaluate DuckDB for development, testing, and exploration to reduce spend and improve velocity.
- Monitor Schrems III closely. If the DPF is invalidated, every US cloud deployment in the EEA needs reassessment. Open formats with clear residency controls provide essential flexibility.
Trends worth watching (not fully covered here)
- Streaming-first architectures (, , Redpanda, RisingWave) merging with lakehouse patterns. 2026 is being called the year of "streaming-first lakehouses."
- Vector-native storage becoming an embedded requirement for AI-ready platforms.
- The European data skills gap: an estimated 450,000 ICT professional shortfall by 2030 (SQL skills abundant; Spark/Python expertise scarce in Norway).
- Cost management (FinOps) becoming a core competency as AI/GPU workloads drive cloud bills higher.
- Databricks' anticipated IPO (likely H2 2026) will expose audited financials and potentially reset market expectations for the entire sector.
Big Data & Cloud Analytics Platforms: A 2026 Market Landscape · March 2026 · Prepared for internal team review. Databricks figures are unaudited company disclosures. Projections are analyst consensus estimates. Full source records are in the References section above.
References34
- 1Snowflake, "Snowflake Reports Financial Results for the Fourth Quarter and Full Year of Fiscal 2026" (Form 8-K), February 25, 2026. sec.gov ↗
- 2Snowflake, "Snowflake Reports Financial Results for the First Quarter of Fiscal 2027" (Form 8-K). sec.gov ↗ Accessed 2026-06-13
- 3Databricks, "Databricks Grows >55% YoY, Surpasses $4.8B Revenue Run-Rate, and is Raising >$4B Series L at $134B Valuation", December 2025. databricks.com ↗
- 4Databricks, "Databricks Grows >65% YoY, Surpasses $5.4 Billion Revenue Run-Rate, Doubles Down on Lakebase and Genie", February 2026. databricks.com ↗
- 5TechCrunch, "Databricks raises $4B at $134B valuation as its AI business heats up", December 16, 2025. techcrunch.com ↗
- 6Microsoft, "Microsoft 2025 Annual Report" (Azure surpassed $75B, up 34% in FY2025). microsoft.com ↗
- 7Microsoft Fabric Blog, "Two Years in: How Fabric Redefines the Modernization Path for Synapse Users" (Fabric as fastest-growing analytics product; customer milestones). blog.fabric.microsoft.com ↗
- 8DataLakeHouse Hub, "The 2026 State of the Apache Iceberg Ecosystem" (78.6% exclusive use; Delta overlap 39.3%; survey fielded January 2026). datalakehousehub.com ↗
- 9Databricks, "Databricks Agrees to Acquire Tabular, the Company Founded by the Original Creators of Apache Iceberg", June 4, 2024. databricks.com ↗
- 10TechCrunch, "Databricks to buy open source database startup Neon for $1B", May 14, 2025. techcrunch.com ↗
- 11ClickHouse, "ClickHouse raises $400M Series D led by Dragoneer to accelerate expansion across analytics and AI infrastructure" (Langfuse acquisition; ~$15B valuation), January 16, 2026. clickhouse.com ↗
- 12ClickHouse, "Our Story" (Series C $350M). clickhouse.com ↗ Accessed 2026-06-13
- 13MotherDuck, "MotherDuck Raises $52.5 Million Series B Funding as DuckDB Adoption Soars", September 2023. prnewswire.com ↗
- 14Sacra, "MotherDuck funding, news & analysis" (total funding ~$133M). sacra.com ↗ Accessed 2026-06-13
- 15Snowflake, "Snowflake Announces Intent to Acquire Observe to Deliver AI-Powered Observability at Enterprise Scale", January 8, 2026. snowflake.com ↗
- 16Benzinga, "Snowflake Mulls $1 Billion Observe Buyout To Sharpen AI Edge", December 2025. benzinga.com ↗
- 17Snowflake, "Snowflake and Anthropic Announce $200 Million Partnership to Bring Agentic AI to Global Enterprises". snowflake.com ↗ Accessed 2026-06-13
- 18Constellation Research, "Snowflake, OpenAI ink $200 million multi-year partnership", February 2026. constellationr.com ↗
- 19Fortune via Yahoo Finance, "MIT report: 95% of generative AI pilots at companies are failing" (MIT *The GenAI Divide: State of AI in Business 2025*), 2025. finance.yahoo.com ↗
- 20IAPP, "European General Court dismisses Latombe challenge, upholds EU-US Data Privacy Framework" (Sept 3, 2025 ruling); WilmerHale, "European Court of Justice to Review Challenge to EU-U.S. Data Privacy Framework" (Latombe appeal Oct 31, 2025, Case C-703/25 P). iapp.org ↗ · wilmerhale.com ↗
- 21Flexera, "The latest cloud computing trends — Flexera 2025 State of the Cloud Report" (21% of workloads repatriated). flexera.com ↗
- 22InfoWorld, "Cloud trends 2025: Repatriation and sustainability make their marks" (Barclays CIO survey: 83% planning repatriation, up from 43% in 2021). infoworld.com ↗
- 23DuckDB, "30,000 Stars on GitHub" (June 6, 2025) and PyPI Stats for the `duckdb` package. duckdb.org ↗ · pypistats.org ↗
- 24Definite, "How We Migrated Our Data Warehouse from Snowflake to DuckDB" (>70% cost reduction). definite.app ↗
- 25Greybeam, "How Headset Cut Snowflake BI Costs by 83% While Supporting 2,500+ Users". greybeam.medium.com ↗
- 26Data Center Dynamics, "Microsoft launches Availability Zones in Norway East Azure Cloud region". datacenterdynamics.com ↗
- 27Microsoft, "Important update to Power BI Premium licensing" (P-SKU retirement). microsoft.com ↗
- 28Databricks, "Databricks Sets Official Data Warehousing Performance Record" (TPC-DS 100TB, 2.2x prior record), November 2, 2021. databricks.com ↗
- 29Datatilsynet (Norwegian DPA), "The NO DPA imposes fine against Grindr LLC" (NOK 65M, 2021) and "The Court of Appeal upholds the fine against Grindr" (October 2025). datatilsynet.no ↗ · datatilsynet.no ↗
- 30Cloud Security Alliance, "Unpacking the 2024 Snowflake Data Breach" (UNC5537; ~165 organizations targeted via accounts without MFA), May 7, 2025. cloudsecurityalliance.org ↗
- 31Microsoft Customer Stories, "OBOS BBL unifies analytics with Fabric and cuts costs by 20%". microsoft.com ↗
- 32Microsoft Fabric Blog, "Two Years in: How Fabric Redefines the Modernization Path for Synapse Users" (Kantar: 50% lower cost vs Synapse, 4x throughput). blog.fabric.microsoft.com ↗
- 33Snowflake Engineering Blog, "Snowflake Cortex Analyst: Behind the Scenes" (text-to-SQL accuracy over 90%). snowflake.com ↗
- 34Menlo Ventures, "2025: The State of Generative AI in the Enterprise" ($37B enterprise AI spend). menlovc.com ↗