On June 21, OpenAI announced its acquisition of Rockset, a move that has garnered significant attention in both the database and the broader data and AI fields. Upon hearing this news, many are likely to be curious about three key questions:
- Why did OpenAI acquire Rockset?
- Is this acquisition bullish or bearish for the database industry?
- What should Rockset’s existing customers do now?
Over the past two weeks, I have discussed this acquisition with colleagues at Rockset, venture capitalists, and other professionals in the database industry. Here, I’d like to share my views on this significant development.
What is Rockset?
Rockset is a database company founded by core members of Facebook’s open-source projects such as RocksDB and TAO. Back in 2017–2018, when the Rockset team was just starting, I had extensive discussions with them. At that time, Rockset was pioneering innovative work: focusing on the cloud, real-time processing, and semi-structured data. Their core technology is known as the converged index, an indexing method based on RocksDB. In simple terms, when a JSON file is imported into Rockset, it creates various indexes for all fields in the file, enabling diverse operations such as point access, search, analytics, and more.
The advantage of this indexing method is its versatility in supporting a wide range of use cases. However, the downside is the potential high CPU and storage costs associated with creating and storing these indexes. For Rockset’s business, this versatility means they can reach a broader market, though it also requires significant engineering effort to optimize for various scenarios. Evidently, after eight years of effort, Rockset has developed a highly successful product, securing hundreds of customers.
Why did OpenAI acquire Rockset?
As a world-leading AI company, OpenAI’s every acquisition naturally draws market attention. OpenAI’s acquisition of Rockset has sparked strong interest: what value can a cloud database provide to an AI company in the AI era? I believe Rockset offers at least two types of value to OpenAI.
The first is enhancing Retrieval-Augmented Generation (RAG). As mentioned in the official press release, OpenAI acquired Rockset to “enhance our retrieval infrastructure.” Retrieval infrastructure is a crucial component in building RAG. Rockset already supports various search functions, including vector search, which aligns well with OpenAI’s needs. Although many specialized vector databases are emerging in the market, few can handle the data volumes OpenAI requires. Rockset’s cloud-native architecture is likely one of the few suitable solutions.
The second value lies in data processing and analytics. Regardless of whether a company is AI-focused or not, data processing and analytics services are indispensable. From displaying billing information to users to generating internal reports, OpenAI can utilize Rockset’s services. Without data, there is no AI. Clearly, Rockset can be used to create various services within OpenAI.
Why did OpenAI choose to acquire Rockset instead of purchasing Rockset’s cloud product? I believe OpenAI must have considered directly purchasing Rockset or similar products. However, OpenAI’s traffic is enormous; its website traffic surpassed many top global search engines in 2023. Few database products can withstand such high traffic. Even if there are, OpenAI would likely be the largest customer, potentially paying billions annually in software service fees. Building a product from scratch might take years and require significant effort for system maintenance and optimization. Instead of paying billions annually or spending years developing a new system, acquiring a company to create custom services is a better choice for OpenAI.
Is the acquisition bullish or bearish for the database industry?
I believe Rockset’s acquisition is highly positive for the database industry. We must acknowledge two facts: first, the global economy is still in a slump, with some European countries in recession; second, much of the hot money is flowing into the AI field, while other “traditional” fields like databases receive less attention.
In this external environment, OpenAI’s acquisition of Rockset will undoubtedly spark speculation.
From my discussions with industry insiders and analysis of Rockset’s product, I believe this is a highly positive acquisition for the industry. Using public data, we can analyze from several dimensions.
Firstly, Rockset just completed a \$44M Series B funding round last August. It’s likely that Rockset had substantial cash on hand at the time of the acquisition, giving them many development options. In this case, it’s clear that the decision to accept the acquisition was in Rockset’s hands.
Secondly, as a system developed over eight years, Rockset is mature and has over a hundred enterprise customers. Clearly, the product has achieved a certain level of Product-Market Fit (PMF). With more financial support, Rockset should be able to accelerate its development.
Finally, we can predict that serving a customer like OpenAI should generate significant annual profits for Rockset. Given this scenario, choosing to be acquired likely means OpenAI offered very favorable terms, making it hard to refuse.
What should Rockset’s existing customers do?
Following OpenAI’s acquisition, Rockset will shut down its cloud service within three months. This sounds harsh but is understandable: OpenAI likely has no interest in entering the database market. Rather than letting the acquired company continue its business, it’s better to shut it down and focus on serving its own needs.
For Rockset’s users, they face the inconvenience of system migration. What system should they choose? Following the acquisition news, many database companies have started promoting their ability to replace Rockset on social media. I believe Rockset’s features are unique and comprehensive, making it difficult for any single database to perfectly replace Rockset. Therefore, when choosing a Rockset replacement, users should focus on their specific use cases.
Analyzing Rockset’s product, we can break down its capabilities into several categories: streaming, text search, analytics, and vector search.
While many vendors claim to offer various functions, each company has its strengths. Below, I list some of the best vendors in these fields based on these four categories.
Of course, in many cases, these systems can all meet users’ requirements. When this happens, the responsiveness of system on-calls and the willingness of vendors to listen to customer feedback become key differentiators.
OpenAI acquired Rockset to build its internal search and analytics platform. This acquisition is a significant positive for the database field. If you are a Rockset user needing to migrate services, I recommend choosing the best product based on your specific use case. When different systems can all meet your business needs, lean towards the one with the best customer service.
###
