Today’s apps are operating at a much larger scale and there’s no off button: Data never stops coming in, nor do questions about that.

Because of this, enterprises are burning through compute and storage 24/7 — but compute is orders of magnitude more expensive than storage. Ultimately, compute efficiency can make or break whether an organization can build an Application.

Real-time analytics database Rockset says it can help solve this problem: The company today rolled out a new instance class resulting in a 30% reduction in compute costs. It is also announcing a new strategic investment from Hewlett Packard Enterprise’s venture capital arm Hewlett Packard Pathfinder.

“Compute efficiency is the name of the game,” said Venkat Venkataramani, Rockset cofounder and CEO. “Eliminating cost barriers associated with powerful artificial intelligence (AI) [AI] and real-time analytical applications is our primary goal.”

Vector databases are increasingly critical for AI and machine learning (ML) apps, as they store and query data with large amounts of features (such as deep learning embeddings).

In fact, the vector database market size is expected to nearly triple from $1.5 billion in 2023 to $4.3 billion in 2028, with Pinecone, MongoDB, Qdrant, Milvus and others competing in the space.

Rockset, which counts among its customers JetBlue, Meta and Windward, says it sets itself apart with its search and analytics database supporting real-time indexing and SQL on vector embeddings as well as JSON, text, time series and geospatial data.

The company’s new general purpose instance has different memory-to-CPU ratios and autoscaling compute that allows users to scale based on workload, Venkataramani said, and developers can begin building apps for as low as $232 per month.

“It is always price performance,” said Venkataramani. “A 30% win is a massive update here.”

Rockset recently expanded into vector search powered by approximate nearest neighbor (ANN), which affirms that the closest neighbor is almost as good as the correct one. This supports cost-effective, “billion-scale similarity search” in the cloud, the company says.

Venkataramani said that traditional vector embeddings convert data into numbers to understand meaning and relationship — but this process is both time and resource intensive, he contends.

“Indexing and searching across high dimensional vector embeddings can get very expensive

very quickly,” he said. Or, that process is “going to take a really long time. You can’t build effective applications on top of that.”

This is driving movement toward more unified search and retrieval systems, or what Venkataramani calls “hybrid search.”

“Very quickly, the world is moving to hybrid search from pure vector search,” he said. This is “rather than putting together multiple disparate systems that were never designed to work together.”

With this capability, users can “combine these data and ask really powerful questions.”

Rockset’s platform also offers compute-compute separation, allowing users to decouple indexing compute from search query compute so that different clusters can use the same real-time data.

The company’s database allows for low-latency, real-time injection, and supports high concurrent applications. Venkataramani reiterated that compute efficiency is critical to low latency.

“It’s even more important for AI applications; if you get it wrong, the cost could make the Application you want to build completely infeasible,” he said.

Top use cases for Rockset’s platform are customer-facing analytics and search, ecommerce recommendation engines and dynamic pricing engines, Venkataramani said.

For instance, video auction ecommerce marketplace Whatnot faces a unique visibility problem. He likened the platform to “Twitch meets eBay,” where anyone can start a live auction, whether that be shoes or Magic the Gathering cards.

This means that they don’t know beforehand what auctioneers are going to sell — yet they need to provide relevant feeds so that users find what they want and bid on items.

If the company were to look at video streams and compare those to user preferences without real-time data context, “the first 99 out of 100 auctions that matched would be auctions that have already ended,” said Venkataramani.

Modern, real-time recommendation engines provide much more relevant insights so that companies like Whatnot can scale to millions of users, he noted.

“Of all the ongoing auctions that are not spammy, that are still alive, which are relevant to the user when they’re logged in?,” Venkataramani posited.

Meanwhile, Allianz, one of Europe’s top insurance firms, uses Rockset’s database to more quickly generate policies. In just milliseconds, users can put up hundreds of risk factors, Venkataramani said.

Insurance is a highly competitive market, he pointed out, and insurers don’t want to underwrite more risk that they need to. Policy engines help keep this process in check.

“If they’re not incorporating real-time data, insurance quotes could be too expensive,” he said. “If they lose a customer, it’s for good reason.”

JetBlue also uses Rockset for its chatbot and real-time feature store.

As Venkataramani noted, “none of these applications have an off button. They’re running 2/47, data is always coming in that needs to be processed.”

Rockset has raised $105 million to date and today also announces a strategic investment from HPE.

Venkataramani called this a “great sign of validation for what we’re doing in the market.”

Paul Glaser, VP of Hewlett Packard Enterprise and head of Hewlett Packard Pathfinder, said that the importance of real-time data analytics is “intensifying,” driven by demand for accurate and timely insights.

“Rockset is addressing the evolving demands with their advanced capabilities,” he said,  “extending search and analytics into vector search to provide AI and ML developers with rapid access to real-time data.”