Anyscale raises $40 million to launch a managed service for distributed computing workloads

Anyscale, the startup behind the open source project Ray, today closed a $40 million funding round. A company spokesperson says the capital will be put toward growing the ecosystem around Ray and promoting Anyscale’s first commercial offering, a managed Ray platform.

The types of distributed computing involved with AI and machine learning are often fraught with challenges. That’s because distributed systems vary in terms of how difficult they are to implement. Engineers working with distributed computing must test for all aspects of device and network failure, considering all the permutations of failures and bugs that a distributed system can encounter.

That’s where Anyscale comes in. Its solution enables developers to share and run distributed projects using a no-frills set of tools. The company was founded by the creators of Ray, which was developed at the University of California, Berkeley’s RISELab, a five-year intensive computer science research lab.

Ion Stoica, a professor of computer science at UC Berkeley, where he co-directs the RISELab, founded Anyscale last June with CEO Robert Nishihara, CTO Philipp Moritz, and UC Berkeley professor Michael Jordan. Stoica played a significant role in the creation of Apache Spark and other big data frameworks and tools. He was among the founding team at Databricks, a company that helped commercialize Apache Spark, which is now valued at $6.2 billion. And he helped to establish the framework for Ray, which targets software developers as opposed to data scientists.

Anyscale provides a serverless experience that allows users to build, deploy, and manage distributed apps. It’s cloud-agnostic and supports both stateless and stateful computations, in addition to a range of hardware including graphics cards and custom chips like Google’s tensor processing units (TPUs). Anyscale’s serverless experience abstracts away servers and clusters, and it provides autoscaling, which ostensibly makes it easier to move from one node to many.

With Anyscale’s managed service (which is in beta) and the latest version of Ray (Ray 1.0), developers can start on one public cloud like Amazon Web Services or Google Cloud Platform and deploy to another. They also get access to a serverless compute API with support for Python libraries tailored to AI model training, as well as workloads like natural language processing and hyperparameter tuning.

“The Anyscale managed Ray platform redefines serverless computing by freeing users from managing servers and enabling them to move seamlessly between their laptops and the cloud with no code changes,” Stoica said. “This dramatically simplifies the process of developing and productionizing distributed applications and machine learning applications.”

Anyscale has competition in Hypernet Labs, which is developing a platform-agnostic app called Galileo that promises to expedite code deployment for compute-intensive work. Like Ray, Galileo can run in parallel across several or many machines, and it’s designed to scale to hundreds or even thousands of runs for simulations.

But Anyscale says it’s seeing increased customer demand ahead of its managed platform’s general availability, coming in the first half of 2021. “These [rival] serverless platforms are notoriously bad at supporting scalable AI,” Stoica told VentureBeat in a previous interview. “We are excelling in that aspect.”

Existing investor NEA led the series B announced today, with participation from Andreessen Horowitz (a16z), Intel Capital, and Foundation Capital. The oversubscribed round follows a $20.6 million series A in December 2019 and brings Anyscale’s total funding to date to $60.6 million.

Source: Read Full Article