Your browser is not supported. Please update it.

28 March 2023

AT&T gets in bed with Nvidia to pump AI into its networks

Everybody wants Nvidia’s company these days as the Californian chip vendor soars on the back of its chips optimized for machine learning algorithms in the latest round of AI hype. This time there are more tangible deployments across multiple sectors, in addition to the high profile chatbots amid claims that some pass the Turing Test.

Such claims depend on who is doing the test, bearing in mind that many humans would fail it, such as politicians dodging questions they do not want to answer.

The current wave of AI has propelled Nvidia’s stock price to dazzling levels, raising the company’s value to $640bn, six times that of Intel which has failed to exploit this trend anywhere near as effectively. Nvidia was admittedly well placed with its pedigree in GPUs, which can readily be adapted as AI engines with their optimization for manipulating large data arrays.

This has left other companies in other sectors, including telcos, scrambling around for applications to exploit the powerful capabilities, a case of a technology seeking outlets. Now Nvidia has acknowledged AT&T as the first telco to adopt its latest full suite of products under the AI umbrella.

This includes components for improving energy efficiency of its infrastructure, and products aimed at customers, such as conversational AI using Nvidia Riva for consumer services, and real time vehicle routing on the back of NVIDIA cuOpt.

AT&T is also using the NVIDIA AI Enterprise software suite, a key feature of which is the Apache Spark open source analytics engine designed to optimize training algorithms for machine learning. Such training algorithms are iterative in the sense that they involve repeated visits to the same data sets to converge towards an optimum solution to a problem, such as predicting an outcome on the basis of live data.

AT&T though preferred to accentuate energy efficiency with a quote that could have been made by any one of numerous vendors seeking to be greener than each other. “We strive each day to deliver the most efficient global network, as we drive towards net zero emissions in our operations,” said Andy Markus, AT&T’s chief data officer. “Working with NVIDIA to drive AI solutions across our business will help enhance experiences for both our employees and customers.”

At least AT&T, like many major telcos, does face a genuine challenge reconciling a highly competitive 5G market underpinned by frequent benchmarks from mobile analytics firms like Opensignal, with demands for energy efficiency. AT&T has a self-imposed pledge to be carbon neutral by 2035 and has identified broad initiatives to make its operations more efficient involving machine learning, where Nvidia fits.

From Nvidia’s perspective, deals like the one with AT&T are part of a strategy to diversify and move up the layers to become less dependent on the core silicon. We have seen Broadcom’s protracted and ongoing attempt to buy VMWare, and now we have Nvidia Chief Executive Jensen Huang admitting the collaboration with AT&T was motivated partly by this desire to diversify. He alluded to work with AT&T on dispatching trucks more efficiently, and also work with quantum computing researchers to accelerate software development.

Nvidia has also just stepped into provision of cloud services for AI related processing. This was announced at its annual GPU Technology Conference, allowing any business customer to build and run ML models trained on custom data and created for tasks specific to domains, such as writing ad copy.

This comes under the brand Nvidia AI Foundations, broken down into specific domain cloud services, such as Nvidia NeMo for language models and Nvidia Picasso for image, video and 3D content generation. The latter should not be confused with Picasso AI, an app enabling generation of supposedly unique images from text, aimed more at consumers and available from various sources, such as the Apple Appstore.

Picasso AI is in effect a marriage between AI and Adobe’s Photoshop, which is already widely used by artists as part of their preparation for painting, experimenting with variations in color and image. Picasso AI takes this further by enabling more radical extrapolation from sketches or notes already creates and as such could be a valuable tool for artists, but not yet a substitute for them.

As it happens, Nvidia has been collaborating with Adobe over its Picasso package, which is oriented instead towards commercial enterprise applications. This follows launch of Adobe’s own AI art generator called Firefly, muddying the waters further.

We are seeing a trend towards provision of computationally intensive services around AI or otherwise from the cloud, avoiding the need for on-premises hardware. That is drawing in vendors from different quarters, Nvidia as a chip maker, and also Microsoft on the hyper scaling and productivity software front. On this basis it was Microsoft that won the contract in 2021 to supply the UK’s next supercomputer facility rather than a traditional vendor from that camp, such as Cray Research, which is now part of Hewlett Packard Enterprise (HPE) after the latter’s acquisition for $1.3 bn in September 2019.