CES 2024 in Las Vegas ran January 9-12. The theme, if you could call it that, was AI in everything. Televisions with AI upscaling. Laptops with AI chips. Smart home devices with LLM integration. The AI feature became the headline regardless of the product category.
The NPU moment
The most substantive technical development at CES 2024 was the NPU, the neural processing unit, appearing in mainstream laptop processors. Intel Meteor Lake, AMD Ryzen 8040, and Qualcomm Snapdragon X Elite all ship with dedicated NPU silicon. The NPU handles AI inference tasks without routing to the GPU or the CPU, which means lower power consumption and lower latency for on-device AI workloads. Microsoft's Copilot+ PC category, announced later in May, formalised the NPU as a requirement for AI PC labelling.
Samsung and LG's AI TV war
Both Samsung and LG led their TV announcements with AI processing: upscaling standard video to 8K resolution, optimising colour and contrast scene by scene, removing reflections from glossy screens using computational processing. The underlying technology is convolutional neural networks running on chips embedded in the televisions. The marketing positions this as 'AI' because that is the current attention-capturing term. The technology is real even if the framing is stretched.
Rabbit R1 and AI hardware assistants
The device that generated the most conversation at CES 2024 was not from a major manufacturer. Rabbit's R1, a $199 AI assistant device with an action model that can operate apps on your behalf, generated significant interest. The concept of a device built around an AI that does things rather than tells you things resonated with the post-ChatGPT framing of what AI devices should do. Whether the R1's specific implementation delivered on that promise was a question for reviews after shipping.
The software gap in AI hardware
The consistent pattern across AI-labelled hardware at CES 2024 was impressive hardware capability with software still catching up. NPUs in laptops are available before the operating system and application ecosystem are optimised to use them. AI televisions run neural network inference but the scenes where it makes a visible difference are narrower than the marketing suggests. The hardware inflection point in 2024 is real. The software that makes it valuable follows on a slower schedule.