appConfigEntityId

bodyType

customContents

entityId

entityUuid

isVideo

articlePageQuery

article

articleSchemaContent

content

hooksArticlePageAdvertisementConfigArticle

hooksArticlePageFloatingButtonConfigArticle

appConfig

filter

json

liveArticle

Wency Chen

Ann Cao

TikTok owner ByteDance aims to finalise the design of its next-generation in-house central processing unit (CPU) by early next year at the latest, targeting mass production and wider deployment in the second half of 2027 to fuel its expanding AI ambitions, according to three people familiar with the matter.
An early version of the proprietary CPU had been used in-house since late last year, one of the people said. However, given the urgent demand, tape-out of the new CPU could be brought forward, another source added.
Tape-out is the final stage of chip design before physical fabrication begins.
The effort comes as the role of high-performance CPUs is being reassessed in the age of agentic AI, where workloads are shifting from pure matrix computation to more complex task orchestration.
For ByteDance, the CPU is part of a broader push to build its own silicon portfolio and integrate more self-developed hardware into its AI infrastructure. Internal computing demand has surged with the growth of products such as Doubao, its popular AI chatbot, and Seedance, its video-generation model.
The company, which has rapidly expanded in-house chip designs in recent years, has never publicly disclosed progress in this area.

To accelerate development and help secure manufacturing capacity from foundries, ByteDance is collaborating with US chipmaker Qualcomm, which was actively seeking to expand its business in AI data centres, according to two of the sources.
Teaming up with fabless chipmaker Qualcomm – which outsources manufacturing to foundries like Taiwan Semiconductor Manufacturing Company (TSMC) – was intended to help ByteDance navigate a heavily strained advanced chip supply chain, which currently faces bottlenecks in everything from wafer fabrication to advanced packaging, the sources added.
Last week, Qualcomm announced that Meta Platforms would use its data centre CPU, the Dragonfly C1000, while revealing it was also developing custom silicon for two other unnamed “hyperscalers”. Qualcomm management expected its data-centre chip business to generate US$5 billion in revenue in the financial year 2027, including US$1 billion from custom-chip clients.
ByteDance and Qualcomm did not respond to a request for comment.
Morgan Stanley’s “bull case” scenario puts the CPU AI market opportunity at US$238 billion by 2030, implying a 251 per cent compound annual growth rate from 2026.
ByteDance’s heavy investment in self-developed chips mirrors a broader global industry shift. Tech giants and AI developers – including Google, Amazon.com, Microsoft, and OpenAI – have all stepped up investment in custom silicon to lower costs, optimise chips for specific workloads and gain greater control over their supply chains.
However, a severe capacity squeeze remains a universal challenge.
Morgan Stanley recently forecast that TSMC’s capacity for chip-on-wafer-on-substrate (CoWoS) will reach 200,000 wafers per month by the end of 2027. Rival foundries were expected to expand CoWoS-like capacity to 80,000 wafers per month, the bank said.

CoWoS is an advanced packaging technology that acts as a bridge, physically stitching processors and memory modules together on a single piece of silicon. Because it is essential for processing heavy AI workloads, global capacity is severely strained, making it a critical bottleneck for any company designing custom chips.
While Nvidia remains the dominant user of TSMC’s CoWoS capacity, Google’s Tensor Processing Unit – a custom AI accelerator – is the second-largest consumer, securing its capacity through application-specific integrated circuit partners including Broadcom and MediaTek, according to Morgan Stanley.
Amid this global supply crunch, ByteDance and other major Chinese cloud service providers have also been sourcing domestic AI chips from companies including Biren Technology, MetaX Integrated Circuits, Iluvatar CoreX, Moore Threads Technology and Enflame Technology.
ByteDance bought tens of thousands of AI processors from Shanghai-based Iluvatar, the South China Morning Post reported earlier this month.


Top Chinese AI models make 10% of OpenAI, Anthropic revenue despite high valuations: report

ByteDance spin-off Anew Labs raises US$290m as AI drug developers enjoy fundraising boom

Once reliant on Samsung and SK Hynix, Chinese smartphone makers are turning to CXMT

ByteDance’s AI-enhanced short-drama app eclipses China’s Netflix rivals combined

ByteDance

ByteDance targets early next year for new CPU to power own AI infrastructure: sources

ByteDance is seeking wider deployment of its self-developed hardware across its own AI infrastructure. Photo: NurPhoto via Getty Images

The Qualcomm logo is seen outside its headquarters in San Diego, California. Photo: The San Diego Union-Tribune/TNS

The Nvidia logo is seen in this illustration taken June 11, 2026. Photo: Reuters

As AI pushes data centres to breaking point, some Chinese chipmakers bet on SiC

A view of Exchange Square in Central. Photo: Jelly Tse

Basic is expected to debut on public markets with an IPO in Hong Kong. Photo: Jelly Tse

Faster AI, lower costs: DSpark eases inference bottlenecks and chip strain, says DeepSeek

DeepSeek says DSpark delivers faster results and lower costs, marking a breakthough in inference. Photo: Reuters

The new technique could reduce the computing resources needed to serve AI systems, according to a programmer. Shutterstock

Tech

Tech Trends

To accelerate development and help secure foundry capacity, ByteDance is collaborating with US chipmaker Qualcomm


To accelerate development and help secure foundry capacity, ByteDance is collaborating with US chipmaker Qualcomm.


.css-1c6uqr6{color:inherit;font-weight:inherit;font-size:inherit;font-family:inherit;line-height:inherit;overflow-wrap:break-word;}To accelerate development and help secure foundry capacity, ByteDance is collaborating with US chipmaker Qualcomm

To accelerate development and help secure foundry capacity, ByteDance is collaborating with US chipmaker Qualcomm