Originally reported by Network World. Original publication date: August 1, 2019.
VMware Acquires Bitfusion to Virtualize Hardware Accelerators
In a move that attracted less attention than it deserved at the time, VMware acquired Bitfusion, a startup specializing in virtualization software for accelerated computing. Bitfusion’s core technology improves virtual machine performance by offloading processing to hardware accelerator chips — including GPUs, FPGAs, and custom ASICs — and making those resources available as a shared, network-accessible pool rather than as isolated, per-server hardware.
What Bitfusion Actually Does
Traditionally, GPU-based workloads are tightly coupled to the physical server on which the GPU resides. This bare-metal deployment model creates several practical problems for enterprise infrastructure teams:
- Poor GPU utilization: A GPU assigned to one workload sits idle when that workload is not actively using it.
- Resource silos: Organizations cannot easily share accelerator capacity across teams or applications.
- Limited scalability: Scaling a workload typically means provisioning additional dedicated hardware rather than drawing from a shared pool.
Bitfusion addresses these problems by decoupling GPU and accelerator resources from individual servers. Workloads across the network can draw from a common pool of GPU capacity. When a job requires more compute power, multiple GPUs can be brought to bear simultaneously, regardless of which physical host they reside on.
Integration into vSphere
VMware planned to integrate Bitfusion technology directly into vSphere following the close of the acquisition. Krish Prasad, then Senior Vice President and General Manager of VMware’s Cloud Platform Business Unit, framed the deal explicitly in terms of VMware’s AI and ML infrastructure strategy:
“In many ways, Bitfusion offers for hardware acceleration what VMware offered to the compute landscape several years ago. Bitfusion also aligns well with VMware’s ‘Any Cloud, Any App, Any Device’ vision with its ability to work across AI frameworks, clouds, networks, and formats such as virtual machines and containers.”
Krish Prasad, SVP & GM, Cloud Platform Business Unit, VMware
Prasad also noted that the platform was designed to be extensible beyond GPUs, with the ability to support FPGAs and other custom ASICs as the accelerator ecosystem continues to mature.
Why GPU Virtualization Matters for AI and ML Workloads
AI and machine learning workloads are among the most compute-intensive tasks enterprises run today, and GPUs have become the de facto hardware for training and inference. However, the economics of GPU deployment at scale are challenging:
- High-end GPUs are expensive to purchase and operate.
- Individual AI training jobs may not fully saturate a GPU’s compute capacity.
- Multi-tenant environments struggle to share GPU resources safely and efficiently.
By treating GPUs as pooled, virtualized infrastructure — analogous to how vSphere already treats CPU and memory — Bitfusion’s approach enables higher utilization, better cost efficiency, and more flexible resource allocation across workloads and teams.
This parallels VMware’s earlier expansions into storage virtualization and software-defined networking (SDN), both of which extended vSphere’s value proposition well beyond basic server virtualization. Adding hardware accelerator virtualization to that stack represents a similar architectural shift for GPU-heavy workloads.
Competitive Implications
At the time of the acquisition, neither Microsoft Hyper-V nor KVM offered comparable native GPU pooling and virtualization capabilities at the platform level. By embedding Bitfusion into vSphere, VMware positioned itself ahead of competing hypervisor platforms for organizations looking to run AI and ML workloads in virtualized environments on-premises or in hybrid cloud configurations.
VMware Also Acquires Uhana
Bitfusion was not VMware’s only acquisition announced around this period. The company also acquired Uhana, an AI engine built specifically for telecommunications carriers and network operators. Uhana’s platform is designed to:
- Detect anomalies in network or application behavior in real time.
- Prioritize those anomalies based on their potential operational impact.
- Automatically recommend optimization strategies to network operations teams.
Together, the Bitfusion and Uhana acquisitions reflected VMware’s broader push to embed AI and ML capabilities across its infrastructure portfolio — both at the hardware acceleration layer (Bitfusion) and at the network intelligence layer (Uhana).
Summary
The Bitfusion acquisition gave VMware a concrete technical answer to one of the most pressing infrastructure challenges in AI and ML deployments: how to share, pool, and efficiently utilize expensive GPU hardware across a virtualized environment. By integrating Bitfusion into vSphere, VMware extended its hypervisor platform into accelerated computing territory — a move with lasting significance as GPU-based workloads continue to grow in enterprise data centers.