Databricks raises $5B at a $190B valuation, up from $134B six months ago
Databricks closed a $5 billion strategic funding round led by Coatue, Blackstone, MGX, and T. Rowe Price-advised accounts, with new investors including BOND, Clearlake Capital, and Point72. The company has surpassed a $7 billion annualized revenue run-rate with over 80% year-over-year growth, and will direct the funds toward Lakebase (a database for AI agents), Genie (an AI business-data assistant), and Unity AI Gateway (a model-usage and cost-control platform).
Why it matters: A 42% valuation jump in six months, funded by traditional finance players rather than pure AI VCs, signals institutional capital treating AI-native data infrastructure as a durable category, not a bubble bet.
Z.ai ships GLM-5.3, tops open-weights coding leaderboard without retraining the base model
Z.ai released GLM-5.3, reusing the same 743-billion-parameter base model as GLM-5.2 and deriving every capability gain from scaled-up post-training alone. Terminal-Bench 3.0 scores jumped from 4.6 to 28.3, and the model unexpectedly gained strong cybersecurity capability (84.5% on CyberGym) that Z.ai says it did not specifically train for. The model is live now via API and Z.ai's coding plan; open weights follow in about two weeks after safety hardening.
Why it matters: Getting frontier-level coding and unplanned cyber capability gains from post-training alone, without a new base model, changes the cost calculus for how open-weights labs compete with closed frontier labs.
NVIDIA to invest up to $3B in Blackstone-backed power developer Lancium
NVIDIA will invest up to $3 billion in Lancium, a Texas-based power infrastructure developer with roughly 4 gigawatts under contract and sites in development for 15 more, tied to the Stargate data center initiative. The initial $2 billion secures a 20% stake, valuing Lancium near $10 billion; a further $1 billion is contingent on grid-connectivity milestones.
Why it matters: NVIDIA taking equity in the power supply chain, not just selling chips into it, is supply-chain insurance against the real bottleneck for AI buildout: electricity, not silicon.
Anthropic details how Claude's new invisible text watermark works
Claude models launched from August 2 onward embed an invisible, non-opt-out watermark into every piece of text they generate, subtly biasing word choices in ways that become statistically detectable over enough content and that survive copy-paste. Anthropic says the policy is largely driven by the EU AI Act's Article 50 transparency requirement, and it plans to ship a public detection API, though technical implementation details remain unpublished.
Why it matters: A default, non-opt-out watermark on all Claude text output — with no independently verifiable proof it doesn't affect quality — puts a live test case in front of every writer and developer using Claude for anything they don't want tagged as AI-generated.
OpenAI's "Ultrafast" mode runs GPT-5.6 Sol at 14x standard speed
OpenAI introduced Ultrafast, a new processing mode for GPT-5.6 Sol built on a partnership with chipmaker Cerebras, delivering roughly 750 output tokens per second, about 14 times the model's standard speed. It's in limited preview for a small group of enterprise customers, targeted at latency-sensitive use cases like incident response, customer service, and financial analysis.
Why it matters: Sub-second responses at frontier-model quality close the gap between "chatbot" and "real-time system," which is the difference that determines whether AI can sit inside live operational workflows rather than beside them.
OpenAI Foundation commits $100M to public-health AI implementation, starting with hepatitis C
The OpenAI Foundation is funding the Breakthroughs to Follow-Through (B2F) Initiative, a $100 million partnership with the Common Health Coalition to help states and local jurisdictions apply AI tools to medical treatment delivery, beginning with hepatitis C care in Alabama, Illinois, Louisiana, and Massachusetts. Grants aren't tied to a specific AI platform, giving participating jurisdictions flexibility in which tools they use. Additional states are expected to join over the next six months.
Why it matters: Funding tied to a specific disease outcome and multiple states rather than a single pilot org is a scale-first approach uncommon in nonprofit AI grantmaking so far.
Community Foundation for Greater Atlanta places 4 college interns to build AI tools for local nonprofits
The Community Foundation for Greater Atlanta's PIN program placed four college interns with Atlanta nonprofits (The Scholarship Academy, Latin American Association, Soul Food Cypher, Georgia ACT) to implement AI solutions this summer, building on CFGA's 2025 $1M Google AI Opportunity Accelerator investment.
Why it matters: A funder converting a general AI grant into a recurring, hands-on internship pipeline is a reusable model for turning one-off AI capacity funding into a standing capability inside grantee organizations.
AI Impact Hub