Baidu (BIDU +0.31%) recently partnered with Samsung Electronics (OTC: SSNLF) to produce its new Kunlun AI accelerator chips, marking the first collaboration between the Chinese and South Korean tech giants.
Samsung will manufacture the chips with its 14nm process and I-Cube (interposed-cube) package solution, which merges a SoC (system on chip) with HBM (high-bandwidth memory) to create a smaller chipset. It will be powered by Baidu's own XPU neural processor architecture for cloud, edge, and AI tasks.
Baidu and Samsung claim that the new chip will support demanding AI workloads such as search rankings, speech recognition, image processing, deep learning, and autonomous driving. Let's see how this new alliance will benefit both companies.
Image source: Getty Images.
Why is Baidu making AI chips?
Baidu owns the largest search engine in China, so it generates most of its revenue from online ads. However, it's gradually expanding into other markets, including virtual assistants, smart speakers, autonomous cars, and cloud computing.
Its DuerOS voice assistant hit 400 million users in July, up from just 100 million users last August. Last quarter, it stated that monthly voice queries on the platform had more than quadrupled year-over-year to 4.2 billion. Its Xiaodu smart speakers, which are tethered to that ecosystem, now rank third in the global smart speaker market after Amazon (AMZN +2.33%) and Alibaba (BABA +2.98%), according to Canalys.
In the autonomous driving market, Baidu's Apollo software platform has secured over 100 partners since its launch two years ago, and it recently launched its first robotaxi fleet of 45 vehicles in China. In China's cloud platform market, Baidu still ranks fourth behind Alibaba, Tencent, and Amazon -- but its total cloud revenue surged 70% annually last quarter as its expanding ecosystem won over more enterprise customers.
Those newer growth engines don't generate much revenue yet, but they all tether more users to Baidu's core search, advertising, and data mining ecosystem. Crunching all that data requires a lot of processing power, so Baidu previously relied on NVIDIA's (NVDA +1.80%) high-end GPUs to accelerate its AI tasks. However, newer tasks -- especially natural language processing -- require much faster chips.
Image source: Getty Images.
Baidu and Samsung stated that the new Kunlun chip will enable Baidu's natural language processing framework, ERNIE, to process language three times faster than its current GPUs and significantly strengthen DuerOS.
Moreover, switching out NVIDIA's AI accelerators with its own would also reduce Baidu's overall dependence on American technologies, insulate it from trade war headwinds, reduce its data center costs, and widen its moat against AI rivals like Alibaba -- which recently launched its own AI accelerator chip.
How does this partnership help Samsung?
Samsung's semiconductor business -- which generated 28% of its revenue and 39% of its operating profits last quarter -- mainly produces DRAM and NAND memory chips. Samsung leads both markets, but both businesses are struggling with a global supply glut in memory chips and cyclically low prices.
Samsung's semiconductor unit also produces its Exynos application processors for mobile devices, but it only installs those chips in a handful of its higher-end devices and sells them to lesser-known smartphone makers like Meizu.
Samsung's chip unit also owns one of the world's largest chip foundries alongside Intel, TSMC (TSM +2.88%), and GlobalFoundries. Samsung and TSMC, the only two foundries that currently develop 7nm chips, regularly manufacture chips for companies like Apple and NVIDIA.
Gaining Baidu as a major customer will strengthen Samsung's foundry business and possibly offset its slower sales of first-party memory chips. It also keeps its 14nm process occupied as other companies ramp up their orders for new 7nm chips.
A win-win deal which could impact other companies
The deal between Baidu and Samsung clearly benefits both companies. Baidu bolsters its AI hardware, reduces its dependence on American chips, and cuts its data center costs. Samsung gains a new stream of revenue for its foundry business as it awaits a rebound in memory prices.
However, the deal could also highlight risks for other companies. NVIDIA could face slower demand for its data center GPUs as major tech companies develop more in-house AI accelerators. Other Chinese tech companies could choose Samsung over TSMC, which is based in Taiwan, as a preferred foundry due to political concerns.
The partnership won't move the needle for any of these companies right away, but it highlights the growing importance of AI accelerator chips in data centers.
