DatawhaleShare
Latest: Big Model Business , Editorial: Qubits
150 million yuan per employee.
Just now, the biggest acquisition of AIGC this year was born!
Big data giant Databricks bought MosaicML, an AI startup with only 62 employees, for $1.3 billion (about 9.3 billion yuan).
The acquisition package includes that all employees will be retained.
This start-up company focuses on generative AI platforms. It was established less than two years ago and has open sourced two large models of the MPT series, among which MPT-7B has been downloaded more than 3 million times . Clients include Oracle, Allen AI Lab, etc.
In the last round of financing, MosaicML was still valued at US$220 million. As a result, this acquisition directly increased the valuation by nearly 6 times .
The official announcement blog stated that the combination of the two will help any organization build and use a generative AI model, and the cost will only cost thousands of dollars, not millions of dollars.
Biggest Acquisition in AIGC Field
First of all, let's understand the two sides of the acquisition.
Databricks originated from UC Berkeley and participated in the development of the Apache Spark project. As a data storage and analysis giant, as of 2022, it will be valued at $31 billion, helping large companies such as AT&T, Shell, and Walgreens process data.
Some time ago, I just opened up my own large model Dolly , aiming to achieve a similar effect to ChatGPT with fewer parameters.
As for MosaicML , it was founded in 2021 by two former Intel executives, Naveen Rao and Hanlin Tang, and has raised $64 million so far, with investment institutions including DCVC, Lux Capital, and Future Ventures.
They are committed to helping enterprises train and reason large models at a low cost. They have previously trained Stable Diffusion at a cost of less than $50,000. They have also open sourced the PyTorch library Composer for efficient neural network training. There are as many as 20 types Acceleration method, training speed up to 7 times.
Just a few days ago, MosaicML released the large model MPT-30B as an extension of the MPT-7B. The latter has received an enthusiastic response from the community, with more than 3 million downloads.
MPT-30B has completed 8k context length training on H100, and its performance is better than the original GPT-3. In addition, there are two fine-tuning variants, which are good at multi-round instruction tracking MPT-30B-Instruct and multi-round dialogue The MPT-30B-Chat.
Based on this model, MosaicML provides various services such as training and inference (entry version and enterprise version) for enterprise customers.
Their cooperative clients include Allen AI Institute, Cloudflare, Oracle, 3D design tool Dream3D, large model startup Replit, image generation company Natural Synthetics, video understanding AI company Twelve Labs, etc.
This kind of large-scale model commercialization ability is reported to be what Databricks is looking for.
According to Wall Street sources, the transaction will be completed by July 31 . After the acquisition, MosaicML will become a standalone service under the umbrella of Databricks.
Netizens: 150 million yuan per employee on average
As soon as the news of the acquisition came out, netizens inevitably mentioned another data giant, Snowflake .
They are also in full swing to complete the layout of AIGC. Last month, they just completed the acquisition of AI search company Neeva for US$150 million. Last year, they also acquired Streamlit for US$800 million.
Insiders ridiculed:
The competitive game between the two is getting more and more interesting.
But this is just a microcosm of the whole AIGC wave. Technology giants are making frequent moves, and they are deploying AIGC through independent research and development, cooperation, acquisitions and other means.
Just like Databricks, in addition to the open source Dolly large model, it also cooperates with the enterprise AI platform Dataiku, and has made multiple acquisitions, such as the no-code company 8080 Labs in October last year, and the acquisition of data security startup Okera last month.
This trend is unstoppable.
According to the analysis of relevant institutions, by the end of this year, the global generative AI market expenditure will reach 42.6 billion US dollars, and will grow at a compound annual growth rate of 32%, and will reach 98.1 billion US dollars in 2026.
It is worth mentioning that this time some netizens also calculated the payment price of each employee with great interest:
That's an average of $21 million per person .
Although it cannot be equated with acquisitions just for the sake of talents, it is also a current evaluation system for startups.
Twice on time
Others focused on Naveen Rao , the co-founder and CEO of MosaicML , saying that his vision and timing were so precise .
He founded Nervana Systems in 2014, and launched Neon, the underlying deep learning framework, which exploded out of the circle. As a result, he just caught up with AlphaGo and was directly acquired by Intel for $400 million. He also joined Intel as the head of AI products.
Subsequently, Neon had a mediocre response in the industry and was abandoned by Intel.
So he chose to leave in 2021 to start MosaicML , which has now been acquired for $1.3 billion.
Prior to this, he graduated from Duke University with a bachelor's degree in electrical engineering and computer science. Ten years later (2007), he went to Brown University to study for a Ph.D. processor research.
In his personal blog introduction, he introduced himself like this,
I like to build stuff that will eventually learn how to build other stuff.
Learn to grasp the timing and direction.
参考链接：
[1]https://mattturck.com/mosaic/
[2]https://www.mosaicml.com/blog/mpt-30b
[3]https://twitter.com/mattturck/status/1673339558379888643
[4]https://twitter.com/fi56622380/status/1672137540281974784
[5]https://twitter.com/NaveenGRao
[6]https://www.databricks.com/company/newsroom/press-releases/databricks-signs-definitive-agreement-acquire-mosaicml-leading-generative-ai-platform
[7]https://www.wsj.com/articles/databricks-strikes-1-3-billion-deal-for-generative-ai-startup-mosaicml-fdcefc06
Dry goods learning, like three times ↓