Participated · Series A · May 2017

Bitfusion provides an Elastic AI software platform that virtualizes GPU resources so any workload can remotely attach to partial or whole GPUs over Ethernet for the duration of runtime. The company addresses fragmented GPU deployments in enterprises, aiming to improve utilization, performance and reduce CapEx and OpEx. Bitfusion’s software creates a shared GPU pool across heterogeneous clusters, enabling flexible allocation across private and public clouds. CEO Michael Zimmerman said the product helps companies make efficient use of GPU servers and reduce costs. Samsung Ventures’ strategic investment and board seat signal partnership with a major AI industry player and support for wider deployment and optimization of Bitfusion’s AI-attached network software. The company is based in Sunnyvale, Calif. Bitfusion began by offering pre-compiled libraries to make better use of GPUs, FPGAs and other co-processors to speed applications. The company has shifted focus toward training deep learning models and infrastructure management for AI projects. Its core product, Bitfusion Flex, is a framework-agnostic platform that provides a single dashboard to move projects from development to training, testing and deployment. Flex uses containers to simplify scaling and moving experiments between local machines and the cloud, and it also supports bare-metal deployments. The service includes a web-based interface and a command-line interface that can attach remote GPUs to a local laptop. Bitfusion plans to invest the new funding into R&D, bring Flex out of beta in the next few months, and expand its Silicon Valley presence while keeping most R&D in Austin. Bitfusion re-compiles, curates and deploys pre-optimized versions of popular open-source libraries to extract more performance from target CPUs. Its first offering claims up to 4x speed improvements by selecting platform-specific optimized builds. The team plans to extend beyond CPUs to GPUs and FPGAs, targeting up to 100x speedups and exploring binary translation and other acceleration techniques. Bitfusion intends to commercialize via three business lines: software, an on-prem appliance with built-in hardware accelerators, and an accelerated RackSpace cloud. Target markets include pharmaceutical and bioinformatics companies, data analysis, media and virtual reality firms, and SMBs without in-house optimization expertise. The company was founded by three former Intel employees in Austin and has raised a $1.45 million seed round from Data Collective, Resonant VC, and Geekdom.