Artificial intelligence software is coming to a device near you.
That’s the broad conclusion of a new report from a group of Morgan Stanley analysts led by Edward Stanley, head of thematic research for the firm in Europe.
“To power the killer apps of the AI age, more compute will be pushed to ‘edge’ consumer devices,” Stanley writes. “2024 should be a catalyst year for this theme.”
The idea here is while cloud computing will maintain a central role in training large language models and handling “inference” work, there is a considerable portion of AI-related computing that can take place on user devices. It’s a twist on the idea of the Internet of Things, connecting everything to the network—but this time, making the devices smarter.
Barron’s most recent cover story made a related point, detailing the imminent launch of PCs capable of running artificial intelligence workloads from Dell Technologies (DELL),
HPInc.
(HPQ), Lenovo (LNVGY), and others—powered by processors from the chip makers Intel (INTC), Advanced Micro Devices (AMD), and Qualcomm (QCOM). That story called out HP and Dell as appealing investment plays on the trend.
The Morgan Stanley report takes a wider view, anticipating the emergence of AI capabilities not only in PCs, but also in smartphones, wearable devices, drones, autonomous cars, and virtual and augmented reality headsets. The report notes that the key benefits of AI computation on PCs and other devices at the edge of the network include lower cost, lower latency, personalization, and heightened security and privacy.
Cost is a big factor here. The current cost per query for data centers is high, the report notes—estimates suggest queries to large language models are 10 times as expensive in computing costs as conventional internet search. Clouds offer “unrivaled computing power,” Stanley writes, but with higher costs. By contrast, the marginal cost of running inference software on the edge of the network is “close to zero,” the report says.
“The compute power edge devices need to process these workloads is already sunk cost embedded in the original device purchase price,” Stanley says. “And because less data needs to be transferred, the cost of bandwidth is reduced.”
The report calls out six companies—all of which Morgan Stanley rates Overweight—as beneficiaries from the “edge AI” trend. The list includes hardware vendors Dell,
Apple
(AAPL), and
Xiaomi
(XIACF), along with chip makers Qualcomm, STMicro (STM), and
MediaTek
(2454.Taiwan).
Qualcomm is expanding from processors for handsets to AI-capable PC processors. STMicro makes microcontrollers that help boost device power efficiency, which Stanley sees as key to successful edge devices. MediaTek is Asia’s largest chip design house, with a focus on multimedia and wireless technologies. The company in August announced a partnership with
Meta Platforms
(META) to run its open-source Llama 2 model on edge devices.
The report asserts all six companies will be key beneficiaries of the trend and are “likely to outperform” in 2024 and 2025 “as AI proliferates” to the edge.
“Everyday examples of edge AI include facial recognition on a smartphone, voice recognition in smart speakers and anomaly detection in industrial equipment,” Stanley writes. “With the advent of generative AI, the impetus for device upgrades to enable greater computational power natively on consumer hardware is accelerating and spanning beyond often narrowly used smart speakers, for example.”
AI applications will grow in breadth and power as more devices with the requisite computing power come to market, the team adds.
“AI-driven consumer use cases will become integrated in devices we interact with every day,” Stanley writes.
Of these six beneficiaries of the trend, Apple’s positioning is the most complex. The company has been including “neural engine” technology in its homegrown microprocessors since 2017, for tasks like facial and voice recognition, but the company has yet to articulate a clear strategy to leverage AI, which makes it stand out from other tech giants.
But Morgan Stanley analyst Erik Woodring, in a separate research note, says Apple will be a key beneficiary of the firm’s AI-at-the-edge thesis. Apple can leverage its installed base of more than 2 billion devices and over 1.2 billion users to “to generate more meaningful data than most platforms we can think of,” he writes.
Woodring asserts that Apple’s Siri smart assistant could be the company’s killer AI app, a point Barron’s made in a recent Tech Trader column.
“When paired together with Apple’s industry-leading focus on data privacy, and some of the most powerful smartphone and PC hardware/silicon in the market today, vertically integrated across hardware, silicon, software and services, we believe Apple can become a true ‘AI Enabler’ in ways many others can’t,” Woodring writes.
He sees at least five ways for Apple to monetize AI, including hardware market share gains, new ways to generate search traffic, better services monetization, higher App Store purchases and potential for a premium subscription version of Siri.
“We estimate that if Apple charged just $5 per month for Siri, every 10% penetration of the iPhone installed base would drive an incremental $7 billion+ of high-margin services revenue per year,” Woodring writes.
Write to Eric J. Savitz at eric.savitz@barrons.com
Read the full article here