Spark Capital

AWS partners with Superblocks to embed vibe coding

Superblocks, a vibe-coding startup, announced a multiyear joint marketing agreement with Amazon Web Services (AWS) to allow its tool to be embedded inside customers' private AWS clouds. The integration ensures apps built with Superblocks run inside an enterprise's AWS account (spinning up Amazon Aurora databases) and can integrate with Amazon Bedrock, keeping data inside corporate cloud environments. AWS will also assist in selling Superblocks via its partner channels. Superblocks has about 50 employees and had raised $60 million at Series A (announced May 2025), backed by Spark Capital, Kleiner Perkins, Meritech Capital and Greenoaks. The deal is an example of hyperscalers pushing enterprises to run AI/agentic tooling and orchestration inside their clouds to avoid data leakage and vendor lock-in.

Baseten Nears $1.5B Round at $13B Valuation

TechCrunch reports that AI inference startup Baseten is reportedly close to finalizing a $1.5 billion funding round that would value the company at about $13 billion, according to the Wall Street Journal. Baseten raised a $300 million Series E at a $5 billion valuation five months earlier and a $150 million Series D nine months before that. The WSJ says the new round is structured as a split-priced round, with some investors buying in at $13 billion and others at $11 billion. Sources say the deal is co-led by Spark Capital, Sands Capital, Altimeter Capital and Wellington Management. Baseten, founded in 2019, focuses on routing inference requests to cost-effective models, including open-source alternatives.

XDOF raises $70M to build robot training data

XDOF, a robotics data startup emerging from stealth, announced a $70 million funding round led by Thrive Capital, Spark Capital, a16z, Lux and WndrCo. Founded in October 2024 by Philipp Wu, Fred Shentu and Nemo Jin, XDOF has about 60 employees and says it works with roughly 20 customers, including unnamed frontier AI labs. The company builds data pipelines, collection tools, cleaning and annotation systems for robot training and plans global teleoperator and egocentric data collection operations. In partnership with UC Berkeley AI Research, XDOF is releasing the ABC dataset — described as 130,000 robot manipulation trajectories, 300 hours of simulation and 100 hours of evaluations — and has used such data to train robots on tasks like folding T-shirts and loading AirPods into cases.