Anyscale Raises $40M to Scale Distributed Computing with Ray

The landscape of distributed computing experienced a significant shift this year as Folding@home, a long-standing distributed computing initiative, attracted a surge of new contributors eager to assist COVID-19 research by providing computational resources for protein folding and other calculations essential for identifying potential antiviral drug candidates. Currently, a new company is also capitalizing on the potential within distributed computing and has recently announced a new funding round to further its development.

Anyscale, a company established by a team originating from UC Berkeley – the creators of the Ray open-source Python framework for distributed computing projects – has secured $40 million in funding.

The company intends to allocate these funds to the continued enhancement of Anyscale, a platform constructed on Ray, designed to make Ray accessible not only to experienced developers and computing experts, but also to a broader range of technical professionals seeking to execute projects demanding substantial computing capabilities.

Ion Stoica, Anyscale’s executive chairman and co-founder alongside Robert Nishihara, Philipp Moritz, and UC Berkeley professor Michael I. Jordan, explained in an interview that the company is responding to a trend amplified by the events of 2020, but fundamentally driven by the increasing demand from businesses – fueled by the expansion of cloud computing, significant digital transformations, and the necessity to maintain a competitive edge. Businesses, regardless of their industry, are setting increasingly ambitious technological strategies and objectives.

“Essentially, we’re observing a trend where all applications are distributed and operate on clusters, yet constructing these applications is remarkably complex and necessitates teams with specialized knowledge,” stated Stoica. “Our goal is to simplify the process of building a distributed computing project to the same ease as running a program on a personal computer. This will empower standard developers to create scalable applications comparable to those developed by organizations like Google.”

The initial version of Anyscale – which will enable organizations to construct multi-cloud applications from a single system and utilize serverless architecture that dynamically scales to accommodate application requirements – is currently in private beta testing, with a full public launch planned for the coming year.

Stoica noted that there has been considerable interest from companies in the financial services, retail, and manufacturing sectors, as well as organizations involved in design, informatics, and medical research, including those focused on COVID-19 vaccine development, who are currently participating in the private beta program.

This Series B funding round is spearheaded by existing investor NEA, with participation from Andreessen Horowitz (a16z), Intel Capital, and Foundation Capital. A16z previously led the company’s Series A funding round, a $20 million investment made less than a year ago in December.

Intel is functioning as a strategic investor. Alongside other major technology companies like Microsoft, Intel is leveraging Ray’s distributed computing model for its own projects.

Stoica – who also co-founded Databricks, Conviva, and was an early developer of Apache Spark – and Nishihara refrained from disclosing Anyscale’s valuation during the interview, but Stoica confirmed that the funding round was oversubscribed. The company has now raised a total of slightly over $60 million.

While continuing to refine Anyscale, the company has also made significant progress with Ray itself over the past year.

During the Ray Summit – Anyscale’s developer conference held virtually in late September – Anyscale released Ray 1.0, which, in addition to a universal serverless compute API, includes an expanded library for use with Ray 1.0. Nishihara characterized this release as a “major milestone,” as it represents a step toward realizing their broader vision of Anyscale being utilized by non-technical companies for technical tasks.

A recent example highlighted was a recommendation algorithm developed by Intel for Burger King. “The challenging aspect isn’t generating the recommendations themselves, but rather learning from user interactions and choices, and quickly incorporating that feedback into the system,” he explained. This process can be achieved through other methods, but often results in a less responsive user experience due to delays.

Nishihara stated that Ray has experienced “substantial growth” in interest over the past year, but acknowledged the difficulty in determining whether this is attributable to remote work trends or broader developments in computing.

“It’s evident, at the very least, that the pandemic is accelerating this shift,” said Stoica. “Ray offers strong support for various cloud platforms, including Azure, Google Cloud Platform, and others, making it a particularly attractive option.”

We are witnessing a noteworthy trend in enterprise IT, where startups are identifying opportunities by enabling organizations without extensive technical expertise to overcome the digital divide, by providing improved access to cutting-edge computing advancements to companies that lack the resources to build and maintain such tools independently. Similar to how companies like Element AI are working to democratize advancements in AI, Anyscale aims to achieve the same in the realm of enterprise computing.

Furthermore, the fields of AI and computing are intrinsically linked: modern AI applications require significant computational power, which is often beyond the reach of the average company.

“The demand for distributed computing continues to grow alongside the increasing adoption of AI and machine learning in application development,” stated Pete Sonsini, general partner at NEA, in a press release. “However, scaling applications on clusters remains a significant challenge. Serverless computing is emerging as the preferred platform for developing distributed applications. Unfortunately, current serverless offerings support only a limited range of applications and are often cloud-specific—but not Ray and Anyscale. The company’s progress to date demonstrates the characteristics of a leading technology innovator, and we are excited to support the team through this next phase of bridging their open-source and commercial offerings.”