Intel Corp. today said that it’s buying Andreessen Horowitz-backed artificial intelligence startup SigOpt Inc. for an undisclosed sum.

SigOpt, based in San Francisco, develops a software platform that is used by private firms and research groups such as OpenAI to boost the performance of their AI models. Its platform improves model performance through a method known as hyperparameter optimization.

Hyperparameters are settings defined by developers that influence how an AI processes data. These settings can include, among others, the number of artificial neurons in a model and the way those neurons interact with each other. Hyperparameter optimization, SigOpt’s specialty, is the process of fine-tuning these settings to maximize performance.

The task has historically been quite time-consuming because of the sheer number of possible setting combinations that may exist for a large AI. Traditional approaches to finding the optimal configuration often involve searching for hyperparameters in a semirandomized manner. SigOpt says that its platform can perform the process considerably faster by employing a statistical method called Bayesian optimization.

The startup’s platform allows developers to specify the property of their AI that they wish to improve, for instance its accuracy, and then create an “optimization loop” to fine-tune it automatically. In complex projects, multiple properties can be fine-tuned at once. SigOpt says that its platform works with a wide variety of AI types including machine learning models and more complicated deep learning models.

Using SigOpt’s performance-optimization capabilities, Intel could help enterprise customers running AI models on its chips achieve higher processing speeds. The company said that the startup’s technology will be integrated “across Intel’s AI hardware products” in the wake of the deal.

The chipmaker reported more than $3.8 billion in AI-driven revenue last year and this part of its business is only set to become a bigger focus. Intel is currently preparing to launch its first graphic processing unit line in two decades, the Xe series, which will among others include AI-optimized chips optimized for the enterprise. The company previously spent $2 billion last year to buy machine learning processor startup Habana Labs Ltd.

Intel’s AI investment is part of a broader strategy to create new revenue sources beyond the central processing units for which it’s best known. At an investor event late last year, Intel Chief Executive Officer Bob Swan outlined a vision of the company capturing 30% market share in “all silicon.” Intel is pursuing this effort in a time when rival Advanced Micro Devices Inc. is putting more pressure on its core CPU business. 

Photo: Intel