Hyundai and Kia bought a stake in VAST Data, the AI storage vendor CoreWeave put $1.17 billion behind

Hyundai Motor and Kia have bought a piece of the layer that feeds AI its data, and shareholders of Hyundai, Kia, CoreWeave▼−15.54% — CoreWeave, down 15.54 percent today and DigitalOcean▼−14.35% — DigitalOcean, down 14.35 percent today now share one exposure to a private vendor. On October 13 the Seoul-listed automakers disclosed a strategic investment↗ in VAST DataPVT~$9.1B — VAST Data, private, latest valuation ~$9.1B, the US software company whose AI Operating System will now carry their driver-assistance, language-model and robotics work ¹. The amount was not disclosed, and matters less than the signal: AI's binding constraint has moved off the GPU and into the data layer.

The bottleneck moved into the data layer

Inference changed what infrastructure must do. A model answering a question needs the right context, with permissions intact, at query time. Pipelines stitched from object stores, vector databases and ETL tools (software that cleans and moves data between systems) tend to drift out of sync, so the source data, its embeddings and its access controls stop matching. Vaughn Stewart, vice president of systems engineering at VAST, makes that drift his case against best-of-breed stacks ⁴.

VAST's answer collapses storage, database and vector search into one system, with data indexed and queryable the moment it lands ³. The results are concrete. 42dot, Hyundai's self-driving affiliate, halved its sensor and video data volumes in research on the platform ¹. DigitalOcean cut boot times from two minutes to fifteen seconds↗ ³.

One private vendor under four listed companies

The AI cloud CoreWeave committed $1.17 billion to VAST in November 2025↗ and calls it the primary data foundation of its AI cloud ². DigitalOcean, the listed cloud for smaller builders, is deploying the AI OS across its AI-native cloud, with the agent data services it launched this week built on VAST ³⁷.

Hyundai's reason mirrors the cloud providers'. The Group says Hyundai and Kia sell more than seven million vehicles a year, and its Data Flywheel now runs a continuous training loop off that fleet↗ ⁸, so consolidating onto one architecture is an efficiency claim ¹. The concentration invites the lock-in charge. Stewart's defense is that AI builders have little patience for software that underperforms and will wipe and replace it, a sharp break from the storage industry's era of lock-in ⁴.

Nvidia gave away the layer above it

On the same day as the Hyundai disclosure, Nvidia▼−3.18% — Nvidia, down 3.18 percent today released Run:ai, its GPU orchestration platform, as open source↗, no subscription required ⁵. Nvidia also partners with VAST one layer down, on DataEnclave, a confidential runtime announced in September with Nvidia and Cohere as partners: models run in secured virtual machines, and the keys are released only after remote cryptographic attestation ⁴.

Rivals WEKAPVT~$1.2B — WEKA, private, latest valuation ~$1.2B and DDN compete on load-time speed. WEKA claims three and a half seconds into GPU memory, by its own benchmark ⁵, and this week signed a Dubai distributor to carry its software across the Middle East ⁶.

The data that legally cannot leave

Nvidia can give Run:ai away because orchestration is code. The data layer cannot be given away: about half the world's data sits in corporate data centers and colocation because it is sensitive and regulated, and much of it legally cannot leave ⁴. Whoever makes that data usable to AI collects the reward. A private company valued at $30 billion in its April round↗ ⁴ is collecting it, and the listed companies leaning on it are the exposure investors can actually price.