Databricks in May 2026: what a lakehouse company became, and why Japan is where it shows up first

A lakehouse company is how Databricks described itself five years ago.

That has changed.

The story did not start with lakehouses. It started in 2009 in Soda Hall, the computer science building at UC Berkeley, where a PhD student called Matei Zaharia was watching a machine learning job that should have finished in minutes take hours. MapReduce was the dominant big-data framework at the time, and its limitations on iterative algorithms were obvious to anyone trying to do anything interesting with them. Zaharia built Spark to solve the problem. It used in-memory processing where MapReduce wrote everything to disk, and it gave the framework a clean general-purpose API on top. Spark was open-sourced in 2010 and donated to the Apache Software Foundation in 2013, the same year seven AMPLab researchers, Ali Ghodsi, Ion Stoica and Zaharia among them, founded Databricks. The original product was unglamorous: a managed Spark service. None of the founders had run a company before.

The next decade did two things at once. Spark went from a research artefact to the de facto compute layer for big data inside thousands of enterprises. And Databricks, the company built around it, started arguing for an architectural thesis: that the historical split between data warehouses for analytics and data lakes for everything else was a constraint imposed by old infrastructure, not by the work itself. The lakehouse was the answer. One open-format storage layer for structured and unstructured data, with warehouse-grade governance and performance on top. By 2020 the thesis had a name in the market and a credible product behind it. By 2021 it had Series G money.

February 2021: Series G at a $28 billion valuation. Annual recurring revenue $425 million.

February 2026: completion of more than $7 billion in combined equity and debt at the $134 billion valuation set the previous December, and a run-rate of $5.4 billion growing more than 65 per cent year on year. AI alone runs at $1.4 billion. More than 800 customers now spend over $1 million a year, 70 of them spend over $10 million and net revenue retention sits above 140 per cent. Over 60 per cent of the Fortune 500 are inside the platform.

What Databricks built between those funding moments reshaped the platform. Unity Catalog made governance a first-class concern across the stack, with a single layer covering tables, files, models, dashboards and AI assets. Databricks SQL took on the warehouse and crossed a billion in run-rate during 2025, on Snowflake’s territory. Delta Lake hardened the open-table format underneath. MLflow became the de facto standard for managing the machine learning lifecycle. Lakebase brought operational databases inside the platform, a serverless Postgres designed for AI agents that in its first six months grew revenue at twice the pace of the data warehousing product. Agent Bricks turned generative AI from a demo into a production pattern, with reusable agents stitched into enterprise workflows. Genie put natural language across the stack so employees with the right access could chat with their data. And the label kept moving with the work. Lakehouse became a Data Intelligence Platform, with AI now leading the framing across the product line. None of it abandoned the open-format premise the whole architecture was built on.

Spring 2026 added two announcements worth pausing on.

AI Runtime launched on 19 March. Serverless NVIDIA GPU compute for training and fine-tuning, delivered as a utility on the same platform that already runs the analytics and the inference. It removes the procurement friction around dedicated GPU clusters and lets data and ML teams pull GPU capacity the way they already pull CPU capacity, in a single workspace, with the same governance.

Lakewatch followed five days later. The argument behind it sits on a single number. According to ZeroDayClock data cited in the Databricks launch material, the mean time to exploit a disclosed vulnerability collapsed from 23.2 days in 2025 to 1.6 days in 2026. Purely manual security workflows struggle with that velocity. Lakewatch is an open agentic SIEM that pulls security, IT and business telemetry into a single governed environment, with AI agents handling detection, triage and response at machine speed. Up to 80 per cent lower total cost of ownership against legacy SIEMs, with petabyte-scale retention that traditional architectures penalise on every byte ingested. Security telemetry treated as a data problem the lakehouse already knows how to solve. Adobe and Dropbox are listed among early customers.

A platform that began as a faster Spark now runs compute, governance, analytics, AI and security on a single open-format substrate. For procurement teams in Tokyo, Singapore and Seoul, the comparison is no longer Databricks versus a warehouse. It is Databricks versus a stack: warehouse, ML platform, governance layer, operational database and now SIEM.

Nowhere has that platform consolidation landed harder than in Japan, where regulatory rigour, the scale of incumbents and ISMAP gating make platform choices structural rather than experimental.

MUFG Bank adopted the platform in April 2025 to serve 30,000 employees. Toyota’s vista platform followed on 8 January 2026, addressing the months-long data-delivery lags that had constrained earlier work. ISMAP certification arrived three weeks later on 28 January, removing a major public-sector procurement barrier. Simon Davies took over as Senior Vice President and General Manager for Asia Pacific & Japan in April 2026, with more than 1,500 people across the region and Q4 growth running above 85 per cent year on year. New APJ customer disclosures since the Davies handover include Singapore Customs, Samsung Life and SingTel.

The Japanese systems integrator commitment is on the same trajectory. NTT DATA signed a strategic partnership with Databricks on 22 January 2025, with a $200 million revenue target over five years and a plan to train 500 NTT DATA engineers to proficiency on the Data Intelligence Platform by FY2027. For a Japanese IT giant of NTT DATA’s standing to size a single-platform commitment at that level tells you where regulated enterprise data work in Japan is heading. The wider partner evidence points the same way: Yuki Saito of NTT DATA Japan won the APJ Partner Champion of the Year award at the 2025 Global Partner Awards on the back of fourteen Solutions Architect Champions and a hundred certified practitioners cultivated inside the firm.

The same procurement signal shows up in Singapore. The Government Technology Agency of Singapore (GovTech) is the public-sector tech agency driving Singapore’s digital transformation across recruitment, finance, procurement and beyond. Inside its first year on the Databricks Data Intelligence Platform, GovTech’s Chief Data Office cut dashboard creation from 90 days to 30 days, saved 8,000 hours of labour annually, democratised data across 50 per cent of corporate divisions and delivered six-figure cost savings through BI consolidation. Sachin Tonk, GovTech’s Deputy Chief Data Officer, is presenting the journey at the Summit in June. Public-sector references with hard outcomes attached are how procurement in regulated industries actually moves.

Funding rounds and product launches are the visible part of the story. What separates platform claim from platform reality is production at scale inside regulated environments where disrupting live operations is not an option.

The European Central Bank’s SPACE platform supports more than 1,000 data scientists across 96 migrated processes on HBase, Kafka, Spark and Kubernetes. SPACE underpins the analytical work that informs monetary policy decisions across the eurosystem. HUK-COBURG, Germany’s biggest motor insurer, has run its telematics platform with zero downtime since 2016. That is a decade of continuous operation at insurance scale, with pricing models tuned in production against live driving data. Vattenfall’s central Kafka ecosystem operates to the same standards, with real-time event streams feeding generation systems that cannot afford a bad batch. These are the operational backbones of central banking, motor insurance at population scale and electricity generation across northern Europe. The standard is not a successful proof of concept. It is years of ordinary Tuesdays without an outage.

Ultra Tendency architected and implemented all three. Its engineers hold Apache committer status on HBase and other open-source projects in the stack, meaning people who have been trusted, over years and thousands of commits, with the trunk of the projects the entire ecosystem depends on. That is the operational discipline you want around a 24/7 telematics platform or a digital euro pilot. The firm ended Q1 2026 as the biggest Databricks Professional Services Partner by revenue in EMEA, with continuous regulated production work running from Magdeburg across to Berlin, Tokyo, Singapore and Kathmandu.

San Francisco in June is where these conversations move next. Moscone, 15 to 18 June. The Summit drew 22,000 attendees in person last year and a further 65,000 virtually. If you are working on lakehouse, governance, security or cross-regional AI at production scale, the architecture for the next phase will be settled in that room.

Ultra Tendency will be there.