Are we investing in the wrong AI future? 'Edge computing is beating data centers'

Investors and tech giants are currently pouring hundreds of billions of dollars into massive AI data centers. Yet in doing so, they may be betting on an outdated vision of the future. According to Joachim Klement, managing director at financial advisory firm Panmure Liberum, the real breakthrough of artificial intelligence lies not in the cloud, but on the user's desk and in their pocket.

Klement argues that the market is too heavily focused on so-called "frontier models" like ChatGPT and Claude, which require enormous amounts of computing power in data centers. Research from Stanford University, however, shows that smaller language models, which can run on ordinary consumer hardware, can already correctly handle more than 80 percent of everyday computing tasks.

The economic advantages of local hardware

The economic arguments for this shift to the "edge" are compelling. Running an AI model locally on a PC is roughly 80 percent cheaper per gigabyte of memory in initial investment than doing so in a data center. Electricity costs can also be cut by 70 to 80 percent, even when calculated at consumer power rates.

Large companies are already adjusting their strategies accordingly. AT&T, for instance, deploys different models based on cost and required performance, while JPMorgan expects businesses to use a mix of large and small AI models. Thomson Reuters recently launched its own internal system based on Alibaba's open-source Qwen model, which runs at a fraction of the cost of large cloud systems.

A shift in the chain: from hyperscalers to PC makers

If AI tasks move en masse to local devices, so-called "hyperscalers" risk being left with a surplus of data center capacity. In that scenario, value shifts toward hardware manufacturers and edge-chip suppliers, which investors have so far often viewed as AI laggards.

Klement sees manufacturers like Apple and Dell as the big potential winners of this development. Chip designers such as Arm and Qualcomm also stand to benefit directly from rising demand for local computing power. Among memory manufacturers, demand is expected to shift away from costly high-bandwidth memory (HBM) for data centers toward traditional DRAM for PCs and mobile phones.

Nvidia is preparing for the shift

Although chip giant Nvidia remains heavily dependent on big data center spending, the company appears to already sense the change coming. With the introduction of desktop AI solutions like DGX Spark, Nvidia, according to Klement, is building a safety net in case the data center boom cools off.

The edge segment currently accounts for less than 10 percent of Nvidia's revenue, but it offers the company an important escape route should the investment wave in hyperscale infrastructure peak and businesses start scrutinizing their IT spending more critically.