Magic, a startup focused on generative AI programming, has recently announced the completion of a remarkable funding round, raising a whopping $320 million.
This round was led by former Google CEO Eric Schmidt and attracted a slew of high-profile investors, including Alphabet's CapitalG, Atlassian, Elad Gil, Jane Street, Nat Friedman, Daniel Gross, and Sequoia Capital. This funding brings Magic's total raised to nearly $465 million, quickly propelling it into the ranks of well-funded AI programming startups.
Extended Context Windows
One of Magic's standout technologies is its extended context window, known as the "Long-Term Memory Network" (LTM). This means its models can handle larger volumes of input data, avoiding biases when generating code.
The latest model, LTM-2-mini, can process 100 million tokens, equivalent to about 10 million lines of code, or 750 novels. The company claims it decodes tokens 1,000 times more efficiently than traditional models. Compared to current market leaders like Llama3.1, LTM-2-mini requires significantly less memory, showcasing a strong performance advantage.
Official Blog: https://magic.dev/blog/100m-token-context-windows
Partnership with Google Cloud
Alongside the funding, Magic has also partnered with Google Cloud to build two "supercomputers" on the Google Cloud platform. These machines, Magic-G4 and Magic-G5, will be equipped with Nvidia H100 GPUs and the upcoming Nvidia Blackwell chips, respectively.
Magic plans to scale its computing power to "tens of thousands" of GPUs, achieving a processing capacity of 160 Exaflops per second, equivalent to a quintillion operations per second.
Eric Steinberger, Founder and CEO of Magic, stated: "We are thrilled to collaborate with Google and Nvidia, leveraging Google Cloud to build the next generation of AI supercomputers." He noted that Nvidia's Blackwell system will significantly enhance the inference and training efficiency of their models, while Google Cloud will enable them to scale rapidly and benefit from a rich ecosystem of cloud services.
Founded in 2022 by Steinberger and Sebastian De Ro, Magic has a team of about twenty. Although not yet profitable, their goal is to help software engineers automate coding tasks such as writing, reviewing, debugging, and planning code changes, striving for more efficient coding practices.
Key Points:
🌟 Magic secured $320 million in funding, with investors including former Google CEO Eric Schmidt and other prominent institutions.
💻 The company is partnering with Google Cloud to build two powerful supercomputers to enhance AI model training and inference efficiency.
🚀 Magic is dedicated to helping software engineers achieve automated coding, with its extended context window technology giving it a competitive edge in the market.