Ion Stoica
Last reviewed
May 4, 2026
Sources
18 citations
Review status
Source-backed
Revision
v1 ยท 3,730 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 4, 2026
Sources
18 citations
Review status
Source-backed
Revision
v1 ยท 3,730 words
Add missing citations, update stale details, or suggest a clearer explanation.
Ion Stoica is a Romanian-American computer scientist, professor of electrical engineering and computer sciences at the University of California, Berkeley, and a serial entrepreneur whose academic and commercial work has shaped much of the open-source infrastructure that runs modern artificial intelligence and large-scale data systems. He is the co-founder of Databricks (2013) and Anyscale (2019), and a co-creator of Apache Spark, Apache Mesos, and the Ray distributed computing framework. He is also an advisor to the LMSYS organization, which produced the Chatbot Arena leaderboard and the Vicuna model, and his research group at Berkeley produced vLLM, one of the most widely used open-source large language model serving systems.
Stoica earned his PhD in electrical and computer engineering from Carnegie Mellon University in 2000 under the supervision of Hui Zhang, and joined the Berkeley EECS faculty the same year. Across roughly two and a half decades there, he has co-directed a sequence of influential systems research labs: AMPLab from 2011 through 2017, the RISELab from 2017 through 2022, and the Sky Computing Lab launched in 2022. The pipeline of open-source projects produced by these labs, including Spark, Mesos, Tachyon and Alluxio, Spark Streaming, Drizzle, Ray, vLLM and SGLang, has been adopted by virtually every major technology company and now forms a backbone of cloud-based AI workloads.
His early systems work on the Chord distributed hash table, co-authored at MIT in 2001, was recognized with the ACM SIGCOMM Test of Time Award in 2011 and the ACM SIGOPS Hall of Fame Award in 2015. He received the ACM SIGOPS Mark Weiser Award in 2019, was elected an ACM Fellow in 2012, and was elected to the United States National Academy of Engineering in 2024. As of 2025 his commercial work includes Conviva, the streaming analytics company he co-founded in 2006, and Databricks, the AI and data lakehouse company that completed a $5 billion funding round at a $134 billion valuation in early 2026.
| Field | Detail |
|---|---|
| Born | 1964 or 1965, Romania |
| Nationality | Romanian-American |
| Education | Polytechnic University of Bucharest (MS, 1989); Carnegie Mellon University (PhD in Electrical and Computer Engineering, 2000) |
| Doctoral advisor | Hui Zhang (CMU) |
| Known for | Chord DHT, Apache Spark, Apache Mesos, Tachyon/Alluxio, Ray, vLLM, Chatbot Arena, Sky Computing |
| Institutions | University of California, Berkeley (since 2000) |
| Labs led | AMPLab (co-director, 2011 to 2017); RISELab (co-director, 2017 to 2022); Sky Computing Lab (director, since 2022) |
| Companies co-founded | Conviva (2006); Databricks (2013); Anyscale (2019) |
| Notable awards | ACM Doctoral Dissertation Award (2001); NSF PECASE (2002); Sloan Research Fellowship (2003); ACM SIGCOMM Test of Time (2011); ACM Fellow (2012); ACM SIGOPS Hall of Fame (2015); ACM SIGOPS Mark Weiser Award (2019); IEEE Koji Kobayashi Computers and Communications Award (2023); National Academy of Engineering (2024) |
Ion Stoica was born in Romania around 1964 or 1965. He completed both his undergraduate and graduate engineering studies at the Polytechnic University of Bucharest, receiving a Master of Science degree in computer science and control engineering in 1989. He moved to the United States in 1994 to pursue doctoral studies, beginning a PhD program at Old Dominion University with Hussein Abdel-Wahab. In 1996 he transferred to Carnegie Mellon University (CMU), where he joined the systems research group of Hui Zhang.
Under Zhang's supervision, Stoica completed his PhD in 2000 with a dissertation titled "Stateless Core: A Scalable Approach for Quality of Service in the Internet." The thesis introduced Dynamic Packet State, a technique that approximates the per-flow guarantees of stateful routers without requiring routers to maintain per-flow state, and became one of the foundations of his later interest in scalable distributed systems. The work received the 2001 ACM Doctoral Dissertation Award, the most prestigious recognition for a computer science PhD thesis. The collaboration with Zhang continued well past graduation: the two would co-found Conviva together six years later.
Stoica joined the EECS Department at the University of California, Berkeley as an assistant professor in 2000 and has remained on the Berkeley faculty since. He was promoted to associate professor in 2005 and to full professor several years later. He holds the Xu Bao Chancellor's Chair in Computer Science. Throughout his Berkeley career he has been most visible as the co-founder and director of a sequence of five-year industrial-affiliate research labs that have produced an unusually large fraction of the world's most widely deployed open-source data systems.
In 2011 he co-founded the AMPLab, an acronym for Algorithms, Machines and People, with Scott Shenker, Michael Franklin, Randy Katz, Michael Jordan, Armando Fox, Anthony Joseph and David Patterson. AMPLab incubated Spark, Mesos, Tachyon and other systems that came to define the post-Hadoop big data stack. When AMPLab concluded its planned five-year run in 2017, Stoica became principal investigator and director of its successor, the RISELab, an acronym for Real-time Intelligent Secure Execution. RISELab focused on systems for low-latency machine learning serving and secure data analytics, and it was the home of the Ray and Clipper projects, among others.
In 2022, Stoica launched the Sky Computing Lab as the next chapter in this lineage. Sky Computing is a research vision in which workloads run across multiple cloud providers through an intermediation layer of "intercloud brokers" that select the best mix of clouds for a given job. The Sky Computing Lab has hosted the development of vLLM, SGLang, the LMSYS Chatbot Arena, and the Vicuna model series. Stoica continues to direct it as of 2026.
| Period | Lab | Role |
|---|---|---|
| 2011 to 2017 | AMPLab | Co-founder and co-director |
| 2017 to 2022 | RISELab | Co-founder and director |
| 2022 to present | Sky Computing Lab | Founder and director |
Stoica's research output is unusually broad, spanning networking, peer-to-peer systems, cluster scheduling, in-memory analytics, machine learning systems, and language model serving. The unifying theme is the design of systems abstractions for very large scale, and most of his projects have been released as open source and have spawned commercial spin-outs.
While a postdoctoral collaborator at MIT in 2001, Stoica co-authored "Chord: A Scalable Peer-to-Peer Lookup Service for Internet Applications" with Robert Morris, David Karger, M. Frans Kaashoek and Hari Balakrishnan. Presented at the SIGCOMM 2001 conference, Chord defined a distributed hash table using consistent hashing on an identifier ring, with each node maintaining a logarithmic-sized finger table that enabled key lookup in O(log N) hops. Chord became one of the four canonical first-generation DHT designs alongside CAN, Pastry and Tapestry, and it is among the most cited papers in computer networking. The original paper won the ACM SIGCOMM Test of Time Award in 2011 and the ACM SIGOPS Hall of Fame Award in 2015.
Stoica's networking work in the early 2000s also included the Internet Indirection Infrastructure (i3), an architecture in which packets are addressed to logical identifiers rather than to fixed network destinations, and in which receivers register triggers that route packets to their current location. The work was published at SIGCOMM 2002 and inspired follow-on work on multicast, anycast and mobility.
Around 2009, Matei Zaharia, Benjamin Hindman, Andy Konwinski, Ali Ghodsi, Anthony Joseph, Randy Katz, Scott Shenker and Stoica began a project at Berkeley aimed at letting multiple cluster computing frameworks, such as Hadoop and MPI, share a single physical cluster. The result, Apache Mesos, was published in the technical report "Mesos: A Platform for Fine-Grained Resource Sharing in the Data Center" in 2010 and presented at NSDI 2011. Mesos introduced a two-level scheduling model based on resource offers, in which the platform offers free resources to registered frameworks and the frameworks decide which offers to accept. Mesos was adopted as the foundation for cluster management at Twitter and was later used to run the Apple Siri infrastructure. It is widely regarded as a precursor to Kubernetes.
Apache Spark began in 2009 as the PhD project of Matei Zaharia, advised by Stoica at AMPLab. Spark introduced the Resilient Distributed Dataset, an abstraction for distributed in-memory collections that supports lineage-based fault tolerance. The first paper, "Spark: Cluster Computing with Working Sets," was presented at HotCloud 2010 and was co-authored by Zaharia, Mosharaf Chowdhury, Michael Franklin, Scott Shenker and Stoica. The full RDD paper appeared at NSDI 2012. Spark was open-sourced under a BSD license in 2010 and donated to the Apache Software Foundation in 2013, becoming a top-level Apache project in 2014. Spark replaced Hadoop MapReduce as the dominant general-purpose batch and streaming engine in big data and remains one of the most widely deployed pieces of data infrastructure in industry.
In 2013, Haoyuan Li began the Tachyon project at AMPLab under the joint supervision of Stoica and Scott Shenker. Tachyon, later renamed Alluxio, is an in-memory virtual distributed file system that sits between compute frameworks such as Spark and storage layers such as HDFS or S3. The project was open-sourced in 2014 and commercialized through Alluxio, Inc., founded by Li.
In 2016 and 2017, Stoica's group at the new RISELab began work on Ray, a distributed execution framework designed for the workloads emerging in modern AI: reinforcement learning, hyperparameter tuning, model training and online serving. The project was led by Robert Nishihara and Philipp Moritz, then PhD students working with Stoica and Michael Jordan. The first paper, "Ray: A Distributed Framework for Emerging AI Applications," was posted to arXiv in December 2017 and presented at OSDI 2018. Ray combines a task-parallel model with stateful actors, backed by a distributed scheduler and an in-memory object store, and it has become a widely used substrate for large language model training and inference workflows at companies including OpenAI, Uber, Shopify and Anyscale itself.
In 2023, Stoica's group at the Sky Computing Lab released vLLM, an open-source serving system for large language models. The accompanying paper, "Efficient Memory Management for Large Language Model Serving with PagedAttention," was published at SOSP 2023 and authored by Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying Sheng, Lianmin Zheng, Cody Hao Yu, Joseph Gonzalez, Hao Zhang and Stoica. vLLM introduced PagedAttention, a key-value cache management technique that draws on the operating system idea of virtual memory paging to share and reclaim KV cache blocks across requests. The system reported throughput improvements of two to four times over previous LLM serving systems such as FasterTransformer and Orca, and it became one of the most widely used inference engines in the open-source LLM stack within months of release.
The Large Model Systems Organization (LMSYS) was formed in 2023 as a multi-university collaboration involving researchers at Berkeley, CMU, Stanford, UC San Diego and MBZUAI, with Stoica serving as one of its faculty advisors. In March 2023 the group released Vicuna-13B, an open-source chatbot fine-tuned on conversations collected from ShareGPT. The model was reported to reach roughly ninety percent of the response quality of ChatGPT in GPT-4 graded comparisons and was trained for under three hundred dollars in compute. In May 2023 LMSYS launched the Chatbot Arena, a crowdsourced platform on which users compare anonymized model outputs side by side and from which Elo-style rankings are produced. Chatbot Arena rapidly became a de facto reference benchmark for new frontier models, with vendors including OpenAI, Anthropic, Google and Meta engaging with its leaderboard. Related projects in the LMSYS family include FastChat, an open serving and evaluation platform, and SGLang, a structured generation language for LLM programs.
Beyond specific systems, Stoica is a leading proponent of the Sky Computing vision, articulated in the 2021 white paper "From Cloud Computing to Sky Computing" and elaborated in the 2022 paper "The Sky Above the Clouds." The argument is that the public cloud market has reached a point where workloads benefit from running across multiple providers in a federated manner, and that an intermediation layer of intercloud brokers can route workloads to the best mix of clouds based on price, performance and policy.
| Project | Year | Role |
|---|---|---|
| Chord DHT | 2001 | Co-author with Robert Morris, David Karger, M. Frans Kaashoek, Hari Balakrishnan (MIT) |
| Internet Indirection Infrastructure (i3) | 2002 | Co-author |
| Apache Mesos | 2009 to 2011 | Co-creator, AMPLab |
| Apache Spark | 2009 to 2010 | Co-author and PhD advisor of Matei Zaharia |
| Tachyon / Alluxio | 2013 to 2014 | Co-advisor of Haoyuan Li |
| Ray | 2017 to 2018 | Co-creator, RISELab |
| vLLM | 2023 | Senior author, Sky Computing Lab |
| Vicuna and Chatbot Arena | 2023 | Faculty advisor, LMSYS |
| Sky Computing | 2021 to present | Lead architect of vision |
Stoica has co-founded three operating companies, each of which commercialized open-source research from his Berkeley group.
Conviva was co-founded in 2006 by Hui Zhang, Stoica, Aditya Ganjam and Jibin Zhan. The company, initially named Rinera Networks before being renamed Conviva in 2008, commercialized streaming media analytics work that originated in the End System Multicast project at CMU. Conviva's platform measures and optimizes the quality of digital video delivery for major broadcasters and streaming services and is used to monitor a substantial fraction of the world's premium video traffic. Hui Zhang has served as chief scientist and chairman of the board.
Databricks was incorporated in September 2013 by the original creators of Apache Spark: Stoica, Matei Zaharia, Ali Ghodsi, Patrick Wendell, Reynold Xin, Andy Konwinski and Arsalan Tavakoli-Shiraji. Stoica served as Databricks' first chief executive officer until January 2016, when Ali Ghodsi succeeded him as CEO and Stoica took over as executive chairman. The company began as a commercial sponsor and managed-service provider for Apache Spark on top of the public clouds, then expanded into the data lakehouse architecture combining the openness of data lakes with the transactional and governance properties of data warehouses, anchored by the open-source Delta Lake and Unity Catalog projects. With the rise of generative AI, Databricks moved aggressively into AI tooling, including the 2023 acquisition of MosaicML for $1.3 billion. By the mid-2020s Databricks was reporting revenue growth above sixty percent year over year and was one of the largest privately held software companies. The company raised a $10 billion Series J in late 2024 at a $62 billion valuation, was reported to have raised at a $100 billion valuation in mid-2025, and announced a $5 billion round at a $134 billion valuation in February 2026.
Anyscale was co-founded in 2019 by Stoica, Robert Nishihara and Philipp Moritz to commercialize the Ray distributed computing framework. The company offers a managed Ray platform with developer tooling, optimized runtimes and a hosted control plane, and it organizes the annual Ray Summit. Anyscale raised a $100 million Series C round in December 2021 at a roughly $1 billion valuation, achieving unicorn status, with the round co-led by Andreessen Horowitz and Addition.
| Company | Founded | Co-founders | Stoica's role |
|---|---|---|---|
| Conviva | 2006 | Hui Zhang, Aditya Ganjam, Jibin Zhan | Co-founder, CTO |
| Databricks | September 2013 | Matei Zaharia, Ali Ghodsi, Patrick Wendell, Reynold Xin, Andy Konwinski, Arsalan Tavakoli-Shiraji | Co-founder; first CEO 2013 to 2016; executive chairman thereafter |
| Anyscale | 2019 | Robert Nishihara, Philipp Moritz | Co-founder, executive chairman |
A distinctive feature of Stoica's career is the number of his students and postdoctoral researchers who have gone on to lead companies, large open-source projects, or both. The list below is selective.
| Name | Relationship | Notable role |
|---|---|---|
| Matei Zaharia | PhD student | Creator of Spark; CTO and co-founder of Databricks; later Stanford professor and now back at Berkeley and Databricks |
| Ali Ghodsi | Postdoctoral researcher | Co-founder and CEO of Databricks |
| Robert Nishihara | PhD student | Co-creator of Ray; co-founder and former CEO of Anyscale |
| Philipp Moritz | PhD student | Co-creator of Ray; co-founder and CTO of Anyscale |
| Reynold Xin | PhD student | Co-founder and chief architect of Databricks |
| Patrick Wendell | PhD student | Co-founder of Databricks |
| Andy Konwinski | PhD student | Co-creator of Mesos and co-founder of Databricks |
| Mosharaf Chowdhury | PhD student | Co-author of Spark; professor at University of Michigan |
| Tathagata Das | PhD student | Creator of Spark Streaming; founding engineer at Databricks |
| Haoyuan Li | PhD student (joint with Scott Shenker) | Founder and CEO of Alluxio |
| Ganesh Ananthanarayanan | PhD student | Researcher at Microsoft Research |
| Hui Zhang | Doctoral advisor | Co-founder of Conviva |
| Michael Jordan | Long-time collaborator | Co-author on Ray |
| Joseph Gonzalez | Faculty colleague | Co-author on vLLM and Chatbot Arena |
| Hao Zhang | Postdoctoral researcher | Co-author on Vicuna and Chatbot Arena; UCSD faculty |
| Woosuk Kwon | PhD student | Lead author of vLLM |
| Lianmin Zheng | PhD student | Co-creator of vLLM and Chatbot Arena |
| Wei-Lin Chiang | PhD student | Co-creator of Vicuna and Chatbot Arena |
| Year | Award |
|---|---|
| 2001 | ACM Doctoral Dissertation Award |
| 2001 | Okawa Foundation Research Grant |
| 2002 | NSF Presidential Early Career Award (PECASE) |
| 2003 | Sloan Research Fellowship |
| 2004 | IEEE Communications Society William R. Bennett Prize |
| 2007 | CoNEXT Rising Star Award |
| 2011 | ACM SIGCOMM Test of Time Award (Chord, 2001) |
| 2012 | ACM Fellow |
| 2015 | ACM SIGOPS Hall of Fame Award (Chord, 2001) |
| 2019 | ACM SIGOPS Mark Weiser Award |
| 2022 | Honorary Member of the Romanian Academy |
| 2023 | IEEE Koji Kobayashi Computers and Communications Award |
| 2024 | Member, United States National Academy of Engineering |
| 2025 | Fiat Lux Faculty Award, UC Berkeley |
Stoica has also received several conference best-paper recognitions, including SIGCOMM, NSDI, and USENIX awards across his career.
Few computer scientists in the past quarter century can claim to have shaped as many widely deployed pieces of infrastructure as Ion Stoica has, and fewer still have done so while remaining a full-time academic. The chain of labs he has co-led at Berkeley, AMPLab and then RISELab and then the Sky Computing Lab, has been an unusually productive engine of open-source systems software. The list of artifacts that emerged from these labs while Stoica was directing or co-directing them includes Apache Spark, Apache Mesos, Tachyon and Alluxio, Spark Streaming, Drizzle, GraphX, MLlib, Velox, Ray, Clipper, Modin, vLLM and SGLang. Several of these are now standard components of cloud and AI stacks, and several have given rise to publicly traded or near-public companies.
This pattern of academic-to-industrial transfer reflects a deliberate model that Stoica has described in talks and interviews: pose a problem that matters at industrial scale, build a real open-source artifact that addresses it, and use the academic environment to attract the best students and the most ambitious collaborators. The faculty colleagues with whom he has co-led these labs, including Scott Shenker, Michael Franklin, Randy Katz, Michael Jordan, Joseph Gonzalez and others, have shared in this pattern, but Stoica is the connective figure across the longest run of consecutive labs and the broadest set of subsequent companies.
In the artificial intelligence era his impact has if anything intensified. Ray underlies large fractions of LLM training infrastructure across the industry. vLLM is among the most widely adopted open-source inference engines. Chatbot Arena has become a primary public benchmark for frontier models, and the Vicuna model series helped catalyze the open instruction-tuned LLM ecosystem in 2023. Databricks has emerged as one of the largest AI-data platforms in the world, with revenue in the multi-billion-dollar range and a 2026 valuation in the hundreds of billions. Anyscale has positioned itself as the commercial home of Ray. Stoica's own role has been less that of a single-paper academic and more that of a builder of long-running research programs that give rise to communities, companies and standards.
Stoica's career also illustrates a broader pattern in modern systems research, in which the most consequential work is often produced by tightly knit groups that combine deep academic expertise with the engineering discipline of industrial-grade open source. The students, postdocs and collaborators who have passed through his groups have populated the leadership of much of the contemporary AI infrastructure industry, including the executive teams of Databricks, Anyscale, Alluxio and other companies, as well as the faculties of Stanford, the University of Michigan, the University of California San Diego and elsewhere.