Can Intel Finally Master AI Through SambaNova? Eyeing a Second Chance in Custom Chip Strategy

The Deal in Focus

Reports indicate that Intel is eyeing a potential acquisition of SambaNova Systems, an AI infrastructure specialist that has seen its valuation collapse from $5 billion to a rumored purchase price of just $1.6 billion. Unlike its previous venture into AI training accelerators, this acquisition targets a fundamentally different market segment.

SambaNova specializes in AI inference infrastructure—the computational workload that processes trained models rather than creating them. The company’s custom chips, branded as Reconfigurable Dataflow Units (RDUs), form the backbone of its SambaRack system. This integrated solution bundles hardware, networking software, and management layers into complete rack-scale packages designed for seamless datacenter deployment.

Why Intel’s Last AI Bet Flopped

The semiconductor giant’s acquisition history tells a cautionary tale. In 2019, Intel paid approximately $2 billion for Habana Labs, a company developing the Gaudi AI training processor. At the time, this seemed strategic—Gaudi was gaining traction among hyperscalers interested in alternatives to Nvidia’s dominant GPUs.

However, Intel’s execution faltered. Subsequent generations like Gaudi 2 and Gaudi 3 delivered respectable performance metrics but couldn’t overcome two structural barriers. First, Gaudi employed an unfamiliar architecture paired with an underdeveloped software ecosystem. Second, and more critically, Nvidia’s proprietary CUDA platform had already crystallized into industry standard over nearly two decades. This created a “moat” that transcended raw chip performance—CUDA’s mature software libraries and developer familiarity gave Nvidia an insurmountable advantage.

Intel simultaneously pursued a scatterbrained dual approach: developing Gaudi training chips while launching competing data center GPUs. Neither gained meaningful market share. When Intel ultimately cancelled Falcon Shores—a data center GPU incorporating Gaudi technology—it effectively surrendered the AI training battlefield to Nvidia.

A Structurally Different Opportunity

The SambaNova acquisition could represent a strategic pivot rather than a repeat mistake. Several factors distinguish this scenario.

First, the inference market operates under different competitive dynamics than training. While training demands maximum raw computational throughput, inference prioritizes efficiency—getting accurate predictions from existing models with minimal power consumption and latency. Custom silicon like SambaNova’s excels in this constrained optimization problem.

Second, SambaNova already possesses tangible market momentum. The company secured deployment agreements for sovereign AI inference clouds across Australia, Europe, and the United Kingdom in October. Subsequently, OVHcloud, a major European computing provider, selected SambaNova’s systems to complement its AI infrastructure portfolio. These wins suggest the market recognizes value in rack-scale inference solutions beyond commodity GPU deployments.

Third, Intel Capital already owns SambaNova as a portfolio company, and Intel CEO Lip-Bu Tan chairs its board. This reduces acquisition risk through existing familiarity and operational alignment.

Strategic Alignment With Intel’s Pivot

Intel’s cancelled Falcon Shores announcement signaled a strategic reorientation: the company would abandon chip-only strategies in favor of integrated rack-scale AI systems. An acquisition accelerates this transition by providing immediately deployable technology rather than requiring ground-up development of Jaguar Shores, Falcon Shores’ successor.

For a company that has stumbled repeatedly in specialty semiconductor markets, acquiring proven inference infrastructure represents a more tractable path than competing directly against Nvidia in training acceleration or building redundant GPU alternatives.

Whether such a deal materializes remains uncertain. Nevertheless, targeting AI inference rather than training, acquiring validated commercial solutions rather than speculative R&D projects, and leveraging existing portfolio relationships suggest Intel might have learned from its Habana Labs experience—and this time chosen a fundamentally different battleground.