K-AI semiconductor pavilion organized at IFA 2026 under NIPA's lead
Cuckoo demos AI barista water purifier powered by DeepX chip
FuriosaAI, Rebellions, HyperAccel join the show
Korean chipmakers eye global data center customers in Europe
Nvidia, AMD also expand into IFA as show broadens beyond home appliances
"Chip Chip Fab Fab" is a series that explores the semiconductor industry powering the AI era, breaking down complex technology in an accessible yet in-depth way. The name connects the "chip" at the heart of the AI revolution with the "fab" that makes it, and echoes the rhythm of a train in motion, capturing the relentless advance of the semiconductor industry. The series reads the present and future of semiconductors alongside its readers, making the technology easy to understand while going deep on the stories behind the news and the forces shaping the industry.
FuriosaAI, Rebellions and HyperAccel — three of South Korea's leading AI fabless chipmakers — made a surprise appearance at IFA 2026, a show long dominated by global consumer electronics giants. The companies traveled to Europe to open new business-to-business sales channels overseas.
Even Cuckoo, best known for its electric rice cookers, drew attention by setting up in the K-AI semiconductor pavilion alongside the chip companies. It turned out that the "AI barista water purifier" Cuckoo showcased at the event runs on a chip designed by domestic fabless firm DeepX.
Order coffee by voice: the AI water purifier powered by a DeepX NPU
South Korea's AI fabless companies gathered under one roof at the K-AI semiconductor pavilion at IFA 2026, Europe's largest consumer electronics show, held in Berlin from Sept. 4 to Sept. 8. The National IT Industry Promotion Agency led South Korea's participation to raise the profile of the country's AI ecosystem.
The most eye-catching exhibit was a real-world example of AI semiconductors applied to home appliances. Cuckoo showcased an AI barista water purifier running on-device AI.
![Cuckoo's AI barista water purifier on display at IFA 2026 in Berlin. The device runs on DeepX's on-device AI chip.
[[/IMAGE id="71fc5bl9025d4a96923accf28e6fa9d7"]]
[[IMAGE id="71fc5bf9025d4a96923accf28e6fa9d7"]]
Cuckoo's AI barista water purifier on display at IFA 2026 in Berlin. The device runs on DeepX's on-device AI chip.](https://wimg.heraldcorp.com/news/cms/2026/09/11/news-p.v1.20260911.1e56b8d8d00449bda997c87585dce362_P1.jpg)
According to the company, the purifier calculates the optimal water temperature, extraction volume and brew time for a given coffee bean and then drips the coffee automatically. Built on a large language model, it also lets users place voice requests — asking for a stronger or lighter brew, for instance.
The AI barista water purifier runs on DeepX's DX-M1 chip. Cuckoo began developing the combined water purifier and AI drip-coffee device late last year in partnership with DeepX and Able AI.
The DX-M1 runs AI computations entirely on the device without a cloud connection, making it a natural fit for products like the AI barista water purifier.
Its low power consumption and cost efficiency have helped demand grow quickly. DeepX began mass production in the third quarter of last year and has since secured more than $13 million in commercial purchase orders across more than 10 countries worldwide.
Low power, high efficiency: K-fabless firms target European data centers
FuriosaAI, Rebellions and HyperAccel — South Korea's flagship AI fabless chipmakers — each set up booths side by side. FuriosaAI brought its second-generation AI inference accelerator, the Renegade (RNGD), built on TSMC's 5-nanometer process and equipped with 48 gigabytes of fourth-generation HBM3 memory offering 1.5 terabytes per second of bandwidth.
At the show, FuriosaAI put Samsung SDS's Samsung Cloud Platform — built on the RNGD — front and center. The platform is available on a subscription basis as NPU as a Service, meaning customers do not need to purchase AI servers outright.
Rebellions exhibited its next-generation AI chip, the Rebel 100, set for release in the second half of this year. The product is considered comparable in performance to Nvidia's H200 GPU, used in high-performance server applications.
The company is using the Rebel 100 as its calling card to break into the global AI data center market, with a strategy focused on improving return on investment for customers in the AI inference segment.
"A GPU does everything — training, inference, even bitcoin mining — but the Rebel 100 is focused entirely on inference," a Rebellions official said. "In that domain, it can deliver far greater performance at much lower power consumption than a GPU."
HyperAccel similarly brought its data center AI inference accelerator, Bertha, to the show. Bertha's main selling point is cost efficiency.
Bertha uses LPDDR — low-power DRAM — rather than HBM, which gives it lower memory bandwidth than other NPUs but cuts power consumption to one-third. Its per-capacity price is also significantly lower than HBM, positioning it as an affordable AI accelerator.
"This is our first time at a European trade show, so it is still uncharted territory," a HyperAccel official said. "We sensed that the European market is hitting the limits of its dependence on Nvidia and is looking for alternative supply chains — that is what brought us here."
Beyond home appliances: Nvidia and AMD stake out territory at a broader IFA
Korean companies were not the only AI chip firms to show up at IFA. The show's organizers have been actively recruiting semiconductor and robotics companies as they move away from a strategy centered purely on consumer electronics, aiming to sharpen the event's competitive edge.
Against that backdrop, Nvidia's presence stood out. At IFA 2026, Nvidia announced new technologies to improve the accessibility and performance of local AI and said it would work with a range of partners to expand the local AI ecosystem.
The company made the world premiere of PAIR, a personal AI router. More than half of American households own two or more PCs, yet much of that computing power sits idle. PAIR puts those unused resources to work for local AI.
When multiple requests hit the same GPU simultaneously in an agentic AI environment, performance inevitably degrades. PAIR automatically scans a local network for compatible PCs and routes individual inference requests to whichever system has spare capacity.
Nvidia also announced plans to launch the Nvidia RTX Spark Windows PC next month. Its partners unveiled new hardware at IFA as well — Acer revealed a compact desktop RTX Spark concept, while Lenovo introduced the Yoga Pro 9n.
AMD, another major AI chip company, set up a booth this year at CityCube Berlin, the venue Samsung Electronics vacated. In his keynote address, Jack Huynh, AMD senior vice president of the computing and graphics group, said PCs would become personal AI partners and showcased a laptop running Ryzen AI Halo, AMD's local AI platform.
The exhibition floor, however, leaned heavily toward gaming. AMD primarily displayed PCs running its central processing units and GPUs, and its pavilion was located in the Retail Innovation Zone.
"AMD is well known as an AI company, but we have also built deep expertise in gaming over many years," an AMD official said. "AMD chips power Xbox and PlayStation consoles, and we wanted to highlight that here."
jeongwan@heraldcorp.com
![Popular mukbang influencer, 24, dies suddenly: 'ate 70 eggs at a time' [China Pick]](https://wimg.heraldcorp.com/news/cms/2026/09/08/news-p.v1.20260908.581b7189494849c3a9f1ade47eb6596d_T1.jpg?type=h&h=120)
