Databricks lines up funding at $188B to scale AI governance and agents
The data-and-AI platform signs a term sheet led by Coatue and says the cash is for Unity AI Gateway, Genie, and Lakebase. Meta also expands its Louisiana supercluster past $50B, and Microsoft adds model choice to Copilot.
One-Line Summary
Enterprise AI is consolidating around control and compute: Databricks lines up funding at a $188B valuation to push multi-AI governance, Meta scales a $50B data center supercluster, and Microsoft gives Copilot users explicit model choice.
Big Tech
Microsoft brings model choice to Copilot Cowork and previews enterprise agents
Microsoft 365 Copilot’s Cowork feature now includes a model picker so teams can match the model to the task, with options such as Auto, Claude Sonnet 5, Claude Opus 4.8, GPT 5.5 (Frontier), Claude Fable 5 (Preview), and a Sonnet + Opus Advisor mode, according to Microsoft’s documentation. This makes performance, speed, and cost trade-offs visible to end users instead of hiding them behind a single default. 1
Microsoft notes that some models require data retention — for example, Claude Fable 5 is off by default, shows a banner when selected, and retains prompts and responses — while models hosted and operated by Microsoft keep data within Microsoft. Admins can also disable the Anthropic model family in tenant settings. 1
Separately, Microsoft’s Agent 365 (preview) introduces agents with their own identity in Teams, including a Sales Development agent and pre-integrated partner agents from Genspark, Zensai, Egnyte, Zendesk, Manus, Kore, Kasisto, and n8n. Availability is limited under Microsoft’s Frontier/Targeted release programs, signaling early-enterprise testing rather than broad rollout. 2
Industry & Biz
Databricks lines up strategic funding at $188B to push multi‑AI strategy
Databricks, whose Data + AI Platform is used by more than 20,000 organizations and 70% of the Fortune 500, signs a term sheet for a strategic round at a $188 billion valuation led by existing investor Coatue, which it expects to close later this summer. 3
The company says the new capital accelerates three priorities: Unity AI Gateway (multi‑AI governance to control cost and access), Genie (an AI coworker that turns business data into trusted answers and actions), and Lakebase (a serverless Postgres database built for AI agents). 3
CEO Ali Ghodsi frames the shift as moving from “tokenmaxxing to valuemaxxing,” arguing enterprises want the best outcome per dollar and the freedom to choose the right AI for each job; the company adds the capital is also expected to support future AI acquisitions and deeper research. 3
Databricks pitches this as closing the “enterprise context gap” by unifying data and AI on agent‑ready infrastructure so teams can govern cost, security, and reliability while tying AI to business outcomes. 3
Meta’s Louisiana AI supercluster grows to 5GW, topping $50B
CNBC reports that Meta says its Hyperion data center in Richland Parish, Louisiana, will be a 5GW facility costing over $50 billion, up from a previously revealed $27 billion plan tied to a 2GW build with Blue Owl. The expansion comes alongside a 20‑year sales tax exemption for data centers built before 2029; Meta cites $1.6 billion in local contracts since construction began and over $1 billion earmarked for roads and water/wastewater improvements, while stating it pays the full costs of energy, water, and related infrastructure. 4
Meta’s infrastructure brief details a bespoke stack: MTIA in‑house accelerators for inference, an AI‑optimized data center design with liquid‑cooled hardware, and a high‑performance AI network connecting thousands of chips for large‑scale training. 5
Meta’s Research SuperCluster (RSC) features 16,000 NVIDIA A100 GPUs and approaches almost 5 exaflops, enabling fast training for projects like LLaMA. Meta notes LLaMA 65B trained on 2,048 A100s at about 380 tokens per second per GPU in 21 days after ingesting 1.4 trillion tokens. 6
Sheryl Sandberg leads $10M into AI vehicle inspection startup
TechCrunch reports that Sheryl Sandberg leads a $10 million round into Self Inspection, a San Diego startup that uses smartphone photos to assess vehicle damage and generate a detailed repair‑cost report; the company says it has completed more than 1 million inspections with customers including Stellantis’ financial services arm. 7
Self Inspection tells TechCrunch its software has helped customers reduce costs by over $80 million and save more than 300,000 operational hours; the funding will go toward new products, enterprise expansion, and a move into Europe. For operators in automotive, insurance, or fleet, the pitch is lower cycle times without specialized hardware. 7
New Tools
Dynamics 365 Sales Order Agent: automate email‑to‑quote‑to‑order
Sales Order Agent in Dynamics 365 Business Central automates capturing sales order details from customer emails: it identifies the customer, clarifies missing information through multi‑turn email, checks inventory, and produces a quote or order, while keeping designated users in the loop to review outbound messages. The agent runs in the background once you configure the monitored mailbox, roles, and approvals. 8
It navigates Business Central like a user by reading UI metadata, supports capable‑to‑promise to calculate the earliest ship date when stock is short, filters irrelevant messages with a relevance model, and logs telemetry. Admins can fine‑tune permissions and profiles to constrain what the agent can access. 8
What This Means for You
Multi‑model is moving from theory to settings screens. Databricks is funding a multi‑AI governance approach (Unity AI Gateway) while Copilot Cowork exposes model choice directly to knowledge workers, enabling “best outcome per dollar” workflows instead of defaulting to one model for everything. Build small comparisons across models on your core tasks to see where speed, depth, or citation quality matters most. 3
Make data and privacy posture explicit before scaling. Copilot flags when a model requires data retention and allows admins to turn specific model families on or off; use those guardrails to align with legal and regional policies as you evaluate specialized models. 1
Automate where oversight is built‑in. Microsoft’s Agent 365 templates and the Dynamics 365 Sales Order Agent package routine outreach and email‑to‑order work without code, but keep human approvals before messages go out — a good fit for teams that need traceability and control. 2
Compute capacity is a planning constraint, not background noise. Meta’s 5GW, $50B+ Hyperion build underscores the scale behind AI features your teams and customers touch; treat AI capacity and hosting choices as dependencies when scoping timelines and budgets for AI‑heavy projects. 4
Action Items
- Use the Copilot model picker on a real deliverable: For your next proposal or report, compare Auto, GPT 5.5, and Claude Sonnet 5 on the same prompt, then document speed, depth, and citation quality for your team.
- Pilot Dynamics 365 Sales Order Agent: Point it at a test inbox with 10 customer requests, let it draft quotes, and review approvals to measure accuracy and time saved.
- Draft a one‑page model‑choice rubric: Map common tasks (drafting, deep analysis, citations) to a default model, and note when to switch.
- Trial phone‑based inspections in operations: If you manage fleet, field work, or returns, run a small photo‑upload pilot and compare cycle time and error rates to your current process.
Comments (0)