Our Story

Xockets was founded in 2012 by Dr. Parin Dalal and a team of network engineers to develop a new class of cloud processors known today as advanced Data Processing Units, or DPUs, that he invented prior to founding the company. DPUs free server processors, including CPUs, GPUs, and hybrids of these host processors, from data-intensive workloads that would otherwise slow down distributed computing and the growth of the cloud industry.

Dr. Dalal invented new virtual switch computing and switching plane architectures that implement programmable hardware acceleration in the network of cloud data centers for processing data-intensive workloads at network speeds—or line rate—to enable a new era of accelerated computing and machine learning / artificial intelligence (ML/AI) in clouds.

The data-intensive tasks include computing operations for moving data between server processors in distributed computing, such as  proprietary security, networking and storage operations, as well as sorting, organizing, and reducing/combining the data streams to ready them for further processing by applications running on server processors in cloud data centers, including in training large AI models across clusters or collectives of GPUs.

The company was funded by visionary investors that include Dr. Greg Lavender, the current CTO of Intel; Robert Cote, one of the nation’s top IP investors and lawyers, also a Board member, who guided the company in protecting its DPU innovations; and Jerry Yang, the co-founder of Yahoo, who invested through the venture capital firm that he founded, AME Cloud Ventures, to invest in breakthrough cloud technologies.

  • In the early 2010s, Dr. Dalal invented a new computing architecture for clouds and a new cloud fabric, or switching plane architecture, for offloading from server processors (CPUs, GPUs, and hybrids), accelerating, and isolating the processing of data-intensive workloads required in running business applications for customers across clusters or collectives of server processors using the distributed computing services of a cloud data center.
  • Dr. Dalal’s inventions address the exponential growth in data that must be brokered in the network of a cloud data center by server processors to accelerate distributed computing of cloud applications, including the training of large language models for generative AI.
  • His inventions went against the conventional wisdom at the time by extending computing into the network fabric while the industry applied conventional approaches to the big data problem for years before switching to Xockets’ technology. The industry continued in the wrong direction until Dr. Dalal’s new cloud processors, or advanced DPUs, with his new virtual switch computing architecture were revealed in published patent applications, industry meetings, and public demonstrations that occurred starting in 2015.
  • The industry’s conventional approach of adding additional server processors or throwing more hardware at the big data problem under control of server processors lacked the versatility and flexibility of a DPU’s virtual switch-enabled programmable logic pipelines and independence from server processors.
  • Xockets’ DPU technology enables accelerated computing and ML/AI in clouds today.

2010
Xockets
In the early 2010s, Dr. Dalal invents advanced Data Processing Units, or DPUs. His DPU inventions include new virtual switch computing and switching plane architectures for cloud offload of data intensive workloads to programmable logic pipelines of hardware accelerators in the network to enable accelerated computing and AI in cloud data centers
2012
Xockets
Xockets files its first of many patent applications in May 2012, disclosing Xockets’ DPU virtual switch computing and switching plane architectures for cloud offload
2014
Xockets
Xockets starts to evangelize the cloud industry and investors to Xockets’ DPU architectures and begins building a StreamSwitch as the commercial embodiment for going to market with its DPUs
2015
Xockets
Xockets demonstrates its StreamSwitch of DPUs to the industry at Strata, the premier conference for big data solutions and new networking products and technologies, showing extraordinary, accelerated computing performance and TCO savings for cloud data centers. Xockets enters into an exclusive license with SMART Modular for the manufacture of its StreamSwitch of DPUs and seeks to raise the capital to build a chip
2016
Mellanox-Microsoft
In June 2016, Mellanox releases its ConnectX-5 SmartNICs with Xockets’ virtual switch computing architecture, turning its SmartNICs into DPUs for cloud offload, including for deployment in Microsoft’s cloud to deliver accelerated computing of cloud applications. Xockets puts Microsoft on notice of its IP portfolio. Microsoft does not disclose that Mellanox had begun implementing Xockets’ patented DPU computing architecture.
2020
Mellanox-NVIDIA
NVIDIA acquires Mellanox for $7B for its ConnectX DPUs and announces the release with Mellanox of new BlueField DPUs using Xockets’ DPU computing architecture to transition Microsoft’s cloud data centers to training large models for AI with NVIDIA’s GPUs
2022
Xockets
After learning of NVIDIA’s release of its ConnectX and Bluefield DPUs for cloud offload in 2021, Dr. Parin Dalal personally puts NVIDIA on notice of their use of Xockets’ DPU inventions in a February 2022 meeting with one of the leaders of NVIDIA’s DPU business who had joined NVIDIA from Mellanox as part of the acquisition.
2022
NVIDIA-Microsoft
NVIDIA expands its infringement of Xockets’ DPU inventions in August 2022 with the release of its NVLink Switch DPUs—a new DPU that uses Xockets DPU switching plane architecture to form a new cloud fabric for connecting together GPUs in NVIDIA’s new Hopper GPU-enabled AI server computer systems—up to eight GPUs in a Hopper GPU system. The NVLink Switch DPUs in this new cloud fabric are designed to accelerate computing in the training of large models for AI for Microsoft and other cloud customers
2024
Xockets
Xockets retains an investment banking firm specializing in IP transactions to run a sales process to sell the company and its IP portfolio or license exclusive IP rights to cloud providers. NVIDIA and Microsoft do not engage with Xockets despite using its patented technology. Instead, RPX approaches Xockets seeking to make a deal on behalf of its members
2024
NVIDIA-Microsoft
NVIDIA announces in March that its Blackwell GPU-enabled AI server computer systems will be released in the fall 2024 also with NVLink Switch DPUs. Blackwell greatly expands the number of GPUs connected together using Xockets’ switching plane architecture to 72 GPUs within a rack and 576 GPUs across a cluster of racks to accelerate the training of large models for AI for Microsoft and other cloud customers, providing a 30x improvement in computing performance and a 25x lower TCO than Hopper GPU systems