Bloomberg reported on September 17 that SpaceXAI has held internal discussions about buying customer and operational records from troubled or bankrupt startups to train Grok. Labs are quickly identifying that records of work, whether clean or messy, are going to be more valuable as broad internet data is commoditized and labs continue to move deeper into the enterprise.
A dead startup is leaving behind years of Slack threads where real decisions got made, Jira tickets showing how engineers triaged bugs, email chains where deals moved or died, support tickets with the messy back-and-forth of actual customers. That's the closest thing in existence to a recording of human judgment at scale, and until recently nobody priced had a use case for it.
SimpleClosure told Forbes it facilitated nearly 100 workplace-data transactions in a year, with payouts generally running $10,000 to $100,000 per company. Cielo24 reportedly sold a bundle of Slack messages, internal emails, and Jira tickets for hundreds of thousands of dollars. Small numbers, but real ones and the scale will only get larger as enterprises identify opportunities to sell de-anonymized information that preserves their customer records but provide labs with key model training data.
The auction that made the market visible was Spirit Airlines'. I covered the details in Trove's first issue: Google's $10 million winning bid for the airline's operational corpus, Mercor's $7.5 million backup bid, Micro1's reportedly late $12.5 million offer. This was the first visible example of the market taking shape, both bankrupt and active companies data is now for sale.
Biotech is running a similar play, with the OpenAI Foundation recently giving $500,000 to 1Day Sooner for an effort called CTD Commons, which buys regulatory dossiers from failed biotech companies: toxicology data, manufacturing records, FDA correspondence. The group's president figures nonexclusive copies could go for a few tens of thousands of dollars each.
The labs don’t need another scrape of the web. They needy to see how people actually decided things, and the cheapest place to buy that may be companies that no longer have a use case for their emails and slack messages.
Partnerships & Deals
Compute, Energy & Infrastructure
Crusoe raised $3.9 billion in a Series F at a $30.9 billion post-money valuation. The company says its vertically integrated AI infrastructure platform has more than $140 billion in total contracted value.
Crusoe signed a multi-year deal to run Perplexity’s full model lifecycle on Crusoe Cloud, training frontier models on dedicated NVIDIA GB300 NVL72 clusters and serving them through managed inference, while Crusoe rolls out Perplexity Enterprise to its 1,800 employees.
Euclyd raised more than €200 million in a Series A co-led by Samsung, Somerset Capital Partners, the Scaleup Europe Fund, and Innovation Industries to develop energy-efficient AI inference chips, memory architecture, and data-center systems.
Emerald AI, Google, and NVIDIA launched the AI Energy Management Alliance with Anthropic, utilities, power producers, and infrastructure companies. The group will develop standards for data centers that adjust electricity use in response to grid conditions.
Data, Knowledge & Retrieval
Infillion agreed to acquire Foursquare, adding a location-data business with more than 100 million points of interest, 16 billion human-verified check-ins, and aggregated coverage of 250 million U.S. devices to its advertising platform.
Village Media and OpenAI partnered to build Open Door, an AI-powered community health and care navigation service. Village Media will own the product, while OpenAI is providing funding, API credits, and technical support.
Google is piloting a “pay per value” AI licensing model with publishers, paying based on how valuable their content was for generating responses across Gemini, AI Overviews, and AI Mode. Dozens of publishers have been approached; early payouts are reportedly tiny next to ad revenue, but most would rather be inside the tent than out.
OpenText and Cohere announced a strategic partnership at the ALL IN AI conference in Montreal, pairing Cohere’s agentic AI platform with OpenText’s enterprise data layer for governments and regulated industries that need private or sovereign deployment options.
Mozilla Data Collective raised $5 million from Mozilla to scale its consented, provenance-tracked training datasets, now spanning more than 450 languages with 350 approved contributing organizations.
Keewano launched with $12 million in seed funding led by Hetz Ventures, with a16z speedrun participating, for a database built for AI agent reasoning that keeps data in sequence and context.
Antfly raised $2 million led by Heavybit to build a unified retrieval engine combining lexical, semantic, and graph retrieval with fine-grained access controls, deployable self-hosted or in Antfly Cloud.
Models, Training & Developer Tools
Cohere and Aleph Alpha signed a definitive business combination agreement, advancing a plan first announced in April. The combined sovereign-AI company will operate as Cohere, employ more than 1,000 people, and be dual-headquartered in Toronto and Berlin, subject to regulatory approval.
TypeSafe AI released Jev, its first System One model, after two years in stealth. The company built a new architecture and a training method called Reinforcement Learning for Calibrated Decisions so the model returns typed, probability-scored decisions rather than generated text for uses such as classification, routing, and agent monitoring.
Arcee AI raised a Series B led by Vista Equity Partners, Cambium Capital, and Emergence Capital at a valuation above $1 billion. The company will use the funding to develop open-weight foundation models and enterprise deployment tools.
AIUC raised a $40 million Series A led by Ribbit Capital to expand its auditing, standards, and insurance system from AI agents to frontier models.
MIND raised a $72 million Series B led by Crosspoint Capital Partners to expand its AI-native data-loss-prevention platform. The round brings its total funding to $112 million.
Accenture and Anthropic will build a team of evaluators embedded inside Anthropic to red-team models, assess alignment, and test safeguards. Each company expects to invest at least $1 billion in AI safety over five years.
Enterprise Deployment & Distribution
Salesforce and Google Cloud expanded their partnership so agents on Agentforce and Gemini Enterprise can reason and act across a shared data foundation. Salesforce workloads will also run on Google Cloud through Hyperforce.
Mozilla and Mistral partnered to bring Mistral Small 4 to Firefox Smart Window beta in the United States and Canada while expanding the product to France with French-language support.
Nokia and Microsoft integrated Nokia Data Suite with Microsoft Fabric to create a governed data foundation for agent-driven telecom operations, including network assurance, root-cause analysis, and predictive maintenance.
Siemens and Salesforce connected Siemens Teamcenter service data with Agentforce so industrial sales and service teams can retrieve engineering information inside customer workflows. Siemens is also using Agentforce to qualify inbound leads for its 18,000 sellers.
Industry Applications & Workflows
TotalEnergies and Mistral launched a three-year, €100 million-plus program to train frontier models on nearly a century of geoscience and reservoir-engineering knowledge and almost 10 petaflops of subsurface data.
Magentic raised an $18 million Series A led by Felicis to expand AI agents that work through email, Microsoft Teams, and enterprise systems for manufacturers.
EnforceShield raised €1.7 million led by Vendep Capital to automate the IP enforcement lifecycle across marketplaces, social media, and ad channels, already removing more than 33,500 violations a month for 20+ enterprise clients.
Physical AI & Robotics
Treble Technologies raised $18 million in a Series A-2 led by Paladin Capital Group to expand its acoustic simulation and synthetic-data platform for robots, wearables, and other audio-enabled systems.
Arrive AI and DXC Technology expanded their partnership to deploy secure exchange points for drones, autonomous mobile robots, vehicles, and human couriers across large manufacturing campuses.
D-Robotics closed a $400 million Series C reportedly led by Mirae Assetto to own the picks-and-shovels layer of physical AI, with 8 million robot chips shipped and the S600 adopted by more than 20 robotics companies.
Shutu Technology raised nearly RMB 100 million in a pre-Series A co-led by CDH Baifu and Kunpeng Fund for infrastructure that converts human behavior into structured robot-training data. The company says its data products serve more than 60% of China’s embodied-AI unicorns valued above RMB 10 billion; the figure is company-reported.
Embra AI raised $1 million in pre-seed funding from investors with Boston Dynamics, Agility Robotics, and NVIDIA backgrounds to build data infrastructure for robotics teams: dataset discovery, evaluation, and sourcing across multimodal data.
Science & Healthcare
Novo Nordisk and Anthropic partnered to use Claude Science and Anthropic models for biological reasoning, drug discovery, and agentic software development inside Novo’s research organization.