In just 4 days, OpenAI exposed 2 acquisitions.
On June 21, a major milestone acquisition was announced, OpenAI acquired Rockset, a real-time analytics database start-up, and in addition to the integration of the team, the technology was also integrated into OpenAI's products. Details of the transaction value were not disclosed, but the industry speculates that the acquisition could be valued at US$500 million (about 3.6 billion yuan) based on Rockset's previous financing valuations.
Since its inception to the end of August last year, Rockset has raised more than $100 million, and its main investors include well-known venture capitals such as Sequoia Capital and Greylock, and according to foreign media reports, the post-investment valuation at that time reached a maximum of $500 million.
It is worth noting that according to LinkedIn, Rockset, which was established in 2016, currently has only 86 employees, which is quite effective based on the acquisition price of 3.6 billion.
In the past four days, OpenAI has announced that it has completed the acquisition of remote collaboration company Multi, and the company's only five employees will join OpenAI to merge with the ChatGPT desktop team to jointly promote innovation in the field of artificial intelligence in desktop applications.
Multi has closed two rounds of funding totaling $13 million, and the latest investor is Musha Ventures, as well as Greylock, whose latest typical statement is that it's the best time for AI vertical software.
Regardless of the technical background and financing ability, the two start-ups are quite interesting. Founded by a former Facebook engineer, Rockset has made a name for itself with its unique technological innovations in real-time search and data analytics. Known for its innovative screen sharing and real-time collaboration features, Multi enables efficient communication for small teams, including voice, video calling, and screen annotation capabilities.
The two acquisitions have also sparked heated discussions about OpenAI's strategic moves, and the consensus is that this is a clear signal that OpenAI is advancing deeper into the ToB market. There has also been speculation that multi-party transactions may hint at the next GPT's "multiplayer mode," which allows users to work with AI in real-time on their own devices.
Whatever the underlying implications, an indisputable fact has emerged: OpenAI will take a big step towards a more integrated AI solution. As the mystery is gradually revealed, every move of OpenAI will quietly change the industry ecology.
OpenAI's "Super Gas Station"
The protagonist of the two acquisitions, the most noteworthy is Rockset, which is highly regarded by the industry for its innovation in real-time data analysis and processing.
Rockset's core strength lies in its innovative "vector search" technology, which is especially important for applications that rely on AI technology such as recommendation engines, voice assistants, and chatbots.
By building a "Converged Index", Rockset's database can achieve high-performance search and analysis on various types of data, support real-time SQL queries, and significantly improve the efficiency of data-dependent application development.
Its database service optimizes the processing of structured, semi-structured, geospatial, and time-series data in a targeted manner, ensuring query latency down to the millisecond level, while simplifying O&M, making it suitable for multiple data-intensive industries such as finance, gaming, e-commerce, and logistics.
In "Dear Data", technology author Tan Jing likened Rockset to an instant response expert in the data community, eliminating waiting for AI services, efficiently integrating query technology, and easily controlling the torrent of complex data, which is a powerful tool for enterprises to control big data.
Talking about OpenAI's acquisition logic, Tan Jing pointed out that just as cars can't do without gasoline, the growth of AI models is also hungry for data nourishment. Rockset is like a super gas station for AI, and it has become inevitable for OpenAI to acquire by fine-tuning its training and RAG strategies, especially its combination of the accuracy of a data warehouse, the flexibility of Elasticsearch, and the complex data processing capabilities of a vector database.
Rockset's team is also well-off, having been founded by a group of senior software engineers from well-known tech companies such as Facebook (now Meta), Yahoo, Google, Oracle, and VMware. These engineers' backgrounds span large-scale data management and distributed systems design, building the foundation for Rockset's real-time data analytics and search technologies.
Four of Rockset's six executives have worked at Facebook, including founder and CEO Venkat Venkataramani, who has a background in computer science at the University of Wisconsin-Madison, 14 years of database experience when he founded Rockset in 2016, and was previously director of engineering on Facebook's infrastructure team, and earlier worked at Oracle Database Technology.
CTO Dhruba Borthakur is an alumnus who preceded Venkat by seven years, also worked on databases at Facebook, and is the original developer of RocksDB, a high-performance embedded key-value store database developed and open-sourced by Facebook.
The services provided by Rockset mainly include real-time analysis and queries for event streams, change data capture (CDC) streams, and vector data, which have been successfully applied to multiple customer scenarios, including Facebook (now Meta), and are said to effectively shorten the query response time.
Since its inception, Rockset has raised a total of $117 million in top-tier venture capital investments, including Icon Ventures, Sequoia, and Greylock, with a final round valuation between $100 million and $500 million, with major investors including well-known venture capitals such as Sequoia Capital and Greylock. The company has also expanded from a small team to a medium-sized enterprise with about 100 employees.
剑指AI+Data infra?
On June 24, Multi, a remote collaboration platform, announced that it had joined OpenAI.
This is OpenAI's second move in a week, and the purchase amount has not yet been announced, but the Multi co-founder has revealed that five members of the company will join OpenAI's ChatGPT desktop team after the deal is completed. At the same time, Multi will be shut down on July 24 and all user data will be deleted.
Founded in 2019, Multi has developed a multiplayer collaboration app for macOS that allows teams of up to 10 people to collaborate in real-time through screen sharing, including voice, video calling, annotation, and more.
Multi has raised a total of $13 million in two funding rounds, most recently back in 2020. According to public information, Multi has 6 investors, the latest round is a Series A financing raised by Musha Ventures and Greylock, and the earlier round is a seed round supported by SV Angel, First Round and other institutions.
Multi is also a highly streamlined company, with part-time staff among the 3 founding team.
CEOHamze Ghalebi studied at emlyon business school and is currently co-founder of Narrator, a startup that integrates voice services. CTO Charley Ho is a Chinese with a master's and bachelor's degree in electrical engineering from Stanford University, and is currently a member of the technical team at OpenAI, having previously worked as a software engineer at Google. Another co-creator and CEO, Alexander Embiricos, graduated from Stanford University and was a former Dropbox product manager.
OpenAI's intensive "marriage" has led to speculation that OpenAI is building its own operating system communication tools, and many speculations point to a common intersection: the future trend of GPT-4o real-time voice and the Sora model.
On the one hand, with the help of Rockset, OpenAI can support more complex real-time application scenarios, such as personalized recommendations, dynamic chatbots, and real-time monitoring systems. On the other hand, Multi's technology is more focused on deep integration with existing operating systems, strengthening the application of ChatGPT in fields such as screen drawing and code editing.
This makes people inevitably think of the function of GPT-4o, it has been 1 month since the press conference, GPT-4o's video and voice functions have not yet been launched, and GPT-4o has recently begun to test its voice and video functions in grayscale, indicating that the upgrade of its product line is not far from being more intelligent and interactive.
In addition to the intersection, the more important news may not stop there. In particular, the integration of Rockset's technology into products such as ChatGPT to transform data into actionable intelligence is too much for imagination, indicating that a data-driven AI application revolution is gaining momentum.
Some industry insiders pointed out that this move shows that its strategy picture of AI+Data infra (data infrastructure) is becoming clearer and clearer. OpenAI is actively recruiting data infra experts from Internet giants, proving once again that data is still an indispensable cornerstone in the exploration of AI.
OpenAI's in-depth exploration of AI+Data infra indicates that the development of AI technology in the future will rely more on the real-time, accuracy and diversity of data, and also emphasizes the importance of interdisciplinary integration.
This intensive acquisition is not only a big move of technology integration, but also a signal light for the industry reshuffle. Previously, unicorn companies such as Adept, Humane, and Stability AI have been planning mergers and acquisitions, indicating that the wave of restructuring in the field of large models has arrived.