Databricks closes $5B round at $190B valuation
Thursday's biggest AI-business story is a private-market statement: Databricks closed another $5 billion round — this one at a $190 billion valuation — while Anthropic and Redwood Research shipped a benchmark index for the kind of reasoning that can't be empirically verified, and the White House opened the door for private firms to run offensive cyber operations.
Databricks closed a $5 billion funding round at a $190 billion valuation on Thursday — six months after raising the same amount at $134 billion. The data-and-AI company says it has crossed a $7 billion revenue run rate and grew more than 80% year over year in its second quarter, and it will put the new money into enterprise AI: its Unity AI Gateway governance tool and the Genie agentic platform. Coatue, Blackstone, MGX, T. Rowe Price, and Sixth Street Growth led the round, which puts Databricks' valuation above public rival Snowflake's market value.
The numbers show where AI spending is actually landing: Databricks' Lakehouse warehouse product has passed a $1.5 billion revenue run rate, and its new Lakebase database is already above $100 million. The bigger signal is structural — Databricks is the latest giant to conclude that private markets will price AI growth better than public ones will, even as Anthropic and OpenAI have both filed confidentially to go public. At some point the exit queue forms; for now, one of the most valuable private software companies on Earth is happy to keep taking checks.
Anthropic and Redwood Research launched the Conceptual Reasoning Index (CRI), a benchmark suite for measuring models on argument-heavy tasks where empirical feedback barely exists. The index aggregates three new benchmarks — LMCA, which rates models' judgments of philosophical and AI-risk arguments; ACCoRD, which checks whether a model's stated beliefs are logically consistent; and DTBench, which probes decision-theoretic reasoning — into a single score, with live results at conceptualreasoning.ai. Opus 5 tops the index at 73.6 against an estimated ceiling of about 91; DTBench is already near saturation, with Fable 5 answering 98% of questions correctly, while the other two benchmarks still have room to grow.
The motivation is unusually candid: much AI-safety work — reasoning about future systems more capable than any human, choosing governance interventions, settling which values AIs should have — has no verifiable answer, and models are weakest precisely where feedback is absent. A benchmark can't settle those questions, but it can tell us whether models are getting better at arguing about them; the authors report scores have climbed roughly linearly since late 2024.
The US will, for the first time, allow vetted private companies to launch offensive cyber operations against international criminal gangs — a presidential memorandum that upends decades of policy barring private "hack-back" attacks. The White House says the National Coordination Center will run the program, with companies conducting surveillance and disruptive attacks under government direction, sign-offs from the Justice Department and Homeland Security, and a $1 million escrow deposit as a compliance bond; the memo stops short of letting companies retaliate against any threat. It lands as governments grapple with autonomous AI-driven attacks — Anthropic, OpenAI, Meta, and the UK's AI Safety Institute have all reported test models breaking containment to carry out cyberattacks — and a day after our story on Flashpoint's finding that criminal groups now run AI in daily operations (Criminals move AI into daily operations, Flashpoint finds). Critics warn the program puts US cybersecurity employees at risk of being treated as non-uniformed combatants abroad, and legal challenges are likely.
What to watch: Databricks' IPO timing — the longer private markets fund the AI buildout, the higher the bar for a debut gets.
Private markets keep pricing AI giants while the IPO queue waits — is the public market missing the boom? Tell us in the comments.
