Data Infrastructure

Proprietary data itself is becoming a bought, priced asset class

micro1 has committed $1B over twelve months to buy operational data from companies, financed by Citi and Hercules Capital, with published tiers from $100K+ up to $1M+ for highly unique proprietary data. It bid $12.5M for Spirit Airlines' bankruptcy data and lost to Google's $10M offer. Meta's roughly $15B stake in Scale AI reinforces that training-data supply is now strategic. Data infrastructure is shifting from storing data to sourcing and licensing it, with debt financing entering the picture.

Reducto trained its own post-trained model (R1) for documents, claiming frontier-grade accuracy at far lower cost, and is shipping a ChatGPT plug-in. It positions itself as a headless Word for agents, with Rogo as an early customer. A $100M Series E with a16z and Tau Ventures fits the pattern of capital flowing to structured-data pipelines for AI. Voxel51 (robotics multimodal data) and Exa 3.0 (agent search API) show the same pull toward AI-ready data layers.

Why it matters · Narrow, domain-specific models for data ingestion can undercut frontier-model pricing, creating durable infrastructure businesses beneath the agent layer.

Basedash (scheduled CSV exports, view-only dashboards), Teable Mobile and Gretel (Greylock seed) show the long tail of AI-first data products, mostly with no disclosed round size. Accel's pre-seed in a team-knowledge startup points the same way. Seed rounds account for 14 deals but only $565M of the 90-day stage mix, so this layer is crowded and early.

Why it matters · Product velocity is high but capital per deal is small, so winners will have to differentiate on trusted data or distribution rather than interface.

Boomi acquired Elementary Data, CBRE bought Onnec Group for $1.1B, and Salesforce is reportedly weighing its fifth-largest acquisition. Hyperscalers such as AWS are moving to bring OpenAI models to where customer data already sits. Incumbents are buying observability, quality and physical-layer assets rather than building them.

Why it matters · Exits are happening via strategic buyers, which favors point-solution startups in data quality and observability as acquisition targets.

Weekly funding swung from $5.2B and $5.1B in August and $4.2B in early September to $364M, $118M and near zero by late October. Large unknown-stage and Series C rounds carry most of the dollars ($6.7B and $6.0B respectively), while debt adds $3.1B. Velocity is negative, and Elice's 32% drop on its Korean debut tempers the IPO window.

Why it matters · Headline capital is driven by a handful of infrastructure-scale deals, so underlying venture appetite for software data startups is steadier but far smaller.