ByteDance announced Seedance 2.5 at the Volcano Engine FORCE conference, highlighting 30‑second 4K video generation, 3D white‑model rendering, a new AI copyright platform, mixed community reactions, and a massive surge in Doubao model token usage.
Lance, ByteDance’s newly open‑sourced 3‑billion‑parameter multimodal model, runs on a single 40 GB GPU, tops HuggingFace trend charts, and achieves leading scores on DPG Bench, GenEval, and video generation benchmarks while surpassing several state‑of‑the‑art single‑modal models.
The 2026 ByteDance Scholarship opens to worldwide applicants, expands slots and funding, and now accepts stage‑level results; a five‑year review shows 67 awardees—spanning PhDs, masters and undergraduates—from top universities who have entered top AI labs, founded startups, or taken faculty positions, illustrating how early‑stage research often precedes industry trends.
The article examines Volcano Engine’s new Agent Plan, detailing how its bundled flagship models, Seedance image and video generation skills, web‑search and memory capabilities streamline tasks such as browser‑plugin replication, data‑analysis report creation, full‑stack web dashboards, PDF translation, PPT generation, and Three.js visualizations within Claude Code and Hermes Agent, while comparing it to the earlier Coding Plan model.
The article analyzes ByteDance's Cola DLM, a fully open‑source continuous diffusion language model that abandons token‑centric generation in favor of latent semantic representations, detailing its architecture, training strategy, scaling stability, and how it compares with the earlier ELF model.
DreamLite is a compact 0.39 B unified diffusion model open‑sourced by ByteDance that runs on smartphones, delivering text‑to‑image generation and text‑guided editing in about three seconds for 1024×1024 pictures, with performance comparable to Flux, Z‑Image and LongCat‑Image and offering two variants to balance fidelity and latency.
Douyin, along with other Chinese internet giants, operates data centers with T‑level (≈1 TB/s) outbound bandwidth, hosts over 200 000 servers, and relies on dual‑link designs, CDN acceleration, and multi‑node load balancing to deliver seamless video streams to hundreds of millions of concurrent users.
In early April 2024 the AI sector exploded with record‑breaking financing, groundbreaking model releases such as ByteDance's full‑duplex voice model and the open‑source video leader HappyHorse‑1.0, strategic alliances among top AI firms, and ambitious plans for GPT‑6, signaling rapid commercial scaling of artificial intelligence.
This article ranks the eight most difficult Chinese IT firms to get hired by in 2026, detailing each company's interview pressure, technical depth, system‑design expectations, and the specific knowledge areas that candidates must master to succeed.
The article examines Meta’s acquisition of the AI‑only forum Moltbook and ByteDance’s launch of InStreet, detailing their design choices, rapid user growth, security flaws, market hype, and the broader implications for AI‑driven social ecosystems.
ByteDance and Beijing Jiaotong University unveil VideoWorld 2, a visual‑only AI model that learns from massive video data without language mediation, promising richer detail retention, reduced bias, and a potential paradigm shift in how artificial intelligence perceives the world.
ByteDance and Tsinghua University have created the CUDA Agent, an AI compiler that automatically writes and optimizes GPU kernels, delivering up to double the performance, and heralding a shift where AI‑generated low‑level code could reshape the hardware‑software competition landscape.
ByteDance’s newly released Seed 2.0 series, especially the Pro model, outperforms GPT‑5.2 High and Claude Opus on MathVista and MathVision tests, offers competitive coding scores, multimodal capabilities, and a pricing model up to four times cheaper, while still lagging behind in some programming and factual‑accuracy benchmarks.
ByteDance's Seed 2.0 Pro (Doubao 2.0) showcases industry‑leading performance on math, vision, document, long‑video, and code benchmarks, dramatically lowers inference cost, and is now available in the Doubao app and Trae IDE, positioning it as a serious challenger to GPT‑5.2 and other top LLMs.
ByteDance has open‑sourced the 12‑billion‑parameter video generation model Alive, which supports text‑to‑video/audio, image‑to‑video/audio, pure text‑to‑video and text‑to‑audio modes, runs on a 24 GB GPU, outperforms competitors in cross‑modal synchronization, and includes novel TA‑CrossAttn and UniTemp‑RoPE techniques.
The LMSYS Chatbot Arena now shows Qwen 3.5 (codenamed Karp-001/002) alongside ByteDance's Pisces‑llm models and DeepSeek‑V4, with new Transformers configs and hints of an Active‑3B MoE architecture, suggesting a fresh wave of flagship large language models arriving for the Spring Festival.
ByteDance’s StreamShield delivers a three‑layer resiliency framework—engine self‑healing, hybrid replication at the cluster level, and chaos‑tested releases—that enables over 70,000 concurrent Flink jobs on 11 million CPU cores to meet strict SLAs with second‑level startup and robust fault tolerance.
This article compiles the full set of first, second, and third‑round interview questions used by ByteDance for a data warehouse engineer role, covering topics such as window functions, data skew, shuffle mechanisms, data modeling, data quality, governance, and system design, along with interview duration and interviewer details.
This comprehensive manual outlines ByteDance's Agent platform, covering its technical foundations, architecture, development workflow, real‑world application scenarios, operational optimization, security compliance, future innovation paths, case studies, team collaboration, risk mitigation, tooling, and global adaptation.
After Meta’s $30 million acquisition of Manus, ByteDance counters with AnyGen, a free AI tool designed for workplace tasks such as lesson‑plan creation, CSV‑to‑PPT conversion, and sentiment‑analysis reports, emphasizing structured, deliverable outputs over generic chatbot responses.
Major tech firms including ByteDance, Microsoft, and Amazon are banning the AI coding assistant Cursor over concerns that it uploads proprietary code to cloud models, posing data leakage and potential remote control risks, highlighting a broader industry shift from efficiency‑driven AI tools to security‑first policies.
The article traces how ByteDance’s limited‑release Doubao phone assistant reuses SmartisanOS’s deep‑customized Android code, revealing three layers of inherited assets, the system‑level AI‑assistant architecture, security concerns, and the broader implications for the future of mobile operating systems.
A recent graduate compares offers from Meituan's Beijing marketing platform and ByteDance's Shanghai Douyin e‑commerce team, weighing base location, title prestige, salary, work pressure, team atmosphere, and career growth across transaction, product, instant delivery, and national subsidy domains.
Douyin (ByteDance) operates over 170,000 servers across multiple self‑built data centers with total outbound bandwidth estimated between 7 TB and 10 TB, leveraging dual‑link designs, CDN acceleration, and multi‑node load balancing to support hundreds of millions of concurrent users worldwide.
The article examines how breakthroughs in large language models and high‑fidelity TTS are powering AI‑generated podcasts, analyzes the technical advances behind the "human‑like" sound, surveys major players such as Google, ByteDance, Xiaohongshu and startups, and evaluates the market potential of this rapidly expanding AIGC niche.
This article examines ByteDance’s enormous server inventory and data‑center export bandwidth, explaining how T‑level (terabit) connections, dual‑link designs, CDN acceleration, and global data‑center deployments enable billions of users to stream content simultaneously.
This article compiles the complete set of interview questions used in ByteDance's data warehouse engineer hiring process, covering three interview rounds with topics ranging from self‑introduction and window functions to data skew, shuffle mechanisms, warehouse architecture, data quality, and governance, plus interviewers' feedback and recommended preparation tips.
In a three‑day sprint, ByteDance’s VP Hong Dingkun built an English‑learning app using the AI‑coding platform TRAE, illustrating how large‑model‑driven code completion, natural‑language programming, and AI‑enhanced development can dramatically boost productivity, democratize coding, and push the limits of software intelligence.
Volcano Engine, ByteDance’s cloud arm, is rapidly reshaping China’s cloud computing landscape by achieving record revenues, capturing a dominant share of large‑model token usage, and leveraging aggressive low‑price strategies to challenge Alibaba, Huawei, Tencent and Baidu.
The article explains ByteDance's enormous data‑center bandwidth, server counts, and CDN architecture that enable hundreds of millions of concurrent users on Douyin and TikTok, detailing estimates of total outbound capacity, multi‑link designs, and the role of cloud and IDC resources.
The article analyzes ByteDance's MegaScale‑Infer paper, detailing micro‑batching, M:N Attn‑MoE ratios, cost‑driven constraint search, communication redesign with Mesh All‑2‑All, network latency challenges, and innovative NIC and routing solutions for large‑scale mixture‑of‑experts inference.
This article shares the ByteDance Java backend internship first‑round interview experience, detailing the company's massive hiring drive, interview structure, and deep technical questions covering URL parsing, DNS resolution, TCP handshake, TLS, HTTP request handling, and related networking concepts.
This article explains how ByteDance’s TrafficRoute GTM, a DNS‑based global traffic routing service, uses GEO‑based routing, health‑check orchestration, and intelligent load‑balancing to achieve high stability, performance, and cost efficiency for ultra‑large‑scale traffic across multiple regions and CDN providers.
An intern at ByteDance accidentally removed all sub‑GB machine‑learning models by deleting a parent directory with skip‑trash, prompting a P0 incident that sparked a massive internal discussion about model impact, flat‑management permissions, and the broader implications for AI operations.
This issue presents a curated overview of recent AI developments, including Sam Altman's 2025 technology vision poll, LeCun's interview on future AI directions, ByteDance's hierarchical large language model for recommendation, and the performance and cost advantages of the open‑source DeepSeek‑V3 model.
ByteDance and Kuaishou have launched an “anti‑jargon” campaign, posting restroom cartoons that urge employees to abandon buzzwords like “赋能” and “抓手,” arguing that such jargon hampers clear communication, reduces efficiency, and fuels overtime, while also serving as a status symbol within the tech industry.
The article details ByteDance's use of Ray and RayData to construct scalable audio and video data processing pipelines for multimodal AI models, addressing challenges of massive data volume, resource constraints, and fault tolerance through pipeline design, RayCore enhancements, and custom scheduling optimizations.
AnimateDiff-Lightning, an open‑source diffusion model released by ByteDance on Hugging Face, delivers high‑resolution image and video generation with versatile integration, showcasing how community‑driven AI tools can accelerate creative and commercial applications.
This article details the evolution of ByteDance's TLS (Tinder Log Service) from a Loki‑based prototype to an EB‑scale, cloud‑native log system, covering its core properties, data organization, architecture, caching, hybrid storage, private codec, ecosystem compatibility, intelligent features, and real‑world case studies.
The article explains ByteDance’s massive data‑center fleet, detailing server counts from 2017 to 2020, estimating total outbound bandwidth in the multi‑terabit range, and describing how CDN, load‑balancing, and multi‑link designs enable billions of users to stream videos simultaneously.
This article explains how ByteDance's Douyin platform achieves massive concurrent user capacity by operating data centers with multi‑terabit outbound bandwidth, extensive server fleets, CDN acceleration, and dual‑link designs, providing a technical overview of its infrastructure and bandwidth estimates.
This article details ByteDance's extensive migration of Spark Shuffle to a cloud‑native architecture, describing the massive data volumes, the underlying ESS and CSS services, the challenges of resource isolation, monitoring, throttling, spill‑splitting, and the performance gains achieved across stable and mixed‑resource clusters.
This article outlines ByteDance's exploration of WebAssembly in serverless FaaS, highlighting a conference talk that covers the technology's benefits, runtime architecture, edge integration, and future micro‑service orchestration, while offering attendees practical insights and engineering details.
The article summarizes ByteDance's DataOps journey, detailing its mid‑platform tool and Data BP model, core performance metrics, quality, hardware and human efficiency challenges, concrete DataOps implementation, productization through DataLeap, best‑practice promotion, and future outlook for data‑driven business value.
This article explains why A/B testing is considered the gold standard for causal inference, shares ByteDance’s extensive internal experimentation practices, describes the Volcano Engine platform architecture, outlines how to design and run experiments, and provides real case studies and Q&A for product teams.
This article explains why A/B testing is considered the gold standard for causal inference, shares ByteDance’s extensive internal experimentation practices and case studies, describes the Volcano Engine experiment platform architecture, and outlines the step‑by‑step process for launching reliable A/B experiments.
The article examines how OKR, once hailed as a revolutionary performance‑management tool by companies like Google and ByteDance, has become a source of misalignment, double‑counted KPIs, and cultural friction, and outlines which types of enterprises can still benefit from it while warning that blind adoption may backfire.
The article explains ByteDance's server fleet growth from a few thousand in 2017 to hundreds of thousands today, estimates its data‑center total outbound bandwidth at several terabits per second, and describes the CDN and load‑balancing technologies that enable billions of concurrent users.
This article explains ByteDance’s server fleet size, data‑center export bandwidth, and the CDN technologies that enable hundreds of millions of users to stream videos simultaneously, providing estimates of tens of terabits of capacity and insights into their infrastructure design.
This article explains the evolution, architecture, and high‑availability solutions of ByteDance’s Abase KV storage system, detailing its multi‑write design, leader‑less approach, multi‑region deployment, consistency mechanisms, performance optimizations, and real‑world metrics that support billions of requests per second.
This article introduces Abase, ByteDance’s massive online KV storage system, detailing its evolution from a single‑cluster KV service to a multi‑region, multi‑tenant platform, and explains the high‑availability challenges and the leaderless multi‑write architecture, hybrid logical clocks, quorum settings, and performance optimizations that enable hundred‑billion QPS and sub‑10 ms latency.
This article describes ByteDance's comprehensive approach to managing the massive volume of event‑tracking (埋点) data, detailing the background, cost‑reduction strategies, experience review, future plans, and a Q&A session that together illustrate how systematic data governance can dramatically cut storage and processing expenses.
This article explains ByteDance's large‑scale adoption of profile‑guided optimization (PGO), covering its principles, instrumentation and sampling methods, the automated platform built for data collection and compilation, and the resulting performance gains across dozens of critical services.
Chrome 107 Enables HEVC Hardware Decoding: Impact and Contributions from ByteDance Engineer, enabling lower deployment costs, better user experience for high‑resolution video, and detailing the developer’s journey, challenges, and future plans across platforms.
This article reviews ByteDance’s user‑growth framework, explaining the lifecycle‑based model, MAU formula, and how the company boosts new‑user retention in the critical 0‑to‑1 stage through incentive systems, onboarding, targeted outreach, and fine‑grained analytics.
This article shares a detailed, step‑by‑step account of three interview rounds at ByteDance, covering the background, technical and algorithm questions on Java, Redis, MySQL, Kafka, and more, along with personal reflections and practical tips for future candidates.
The article recaps the online Flutter Festival Hangzhou session where Bytedance engineer Sun Kai presented the Flutter offline R&D tool experience, introduced the flutter_ume debugging platform, announced a plugin competition, and invited developers to join the community and view the full PPT.
This article details ByteDance's wallet asset middle platform designed for the 2022 Spring Festival, covering eight‑app reward interoperability, high‑QPS challenges, token‑based asynchronous入账, budget control, stability measures, and fund‑safety guarantees, and includes practical solutions for hot‑key handling, budget throttling, and multi‑stage activity isolation.
This article shares a step‑by‑step account of a candidate's three‑round ByteDance interview, covering background, technical questions on Java, JVM, Redis, MySQL, Kafka, and algorithm challenges, along with personal reflections on pacing, preparation, and why concise answers matter.
This article explains why A/B testing is crucial for data‑driven product decisions, outlines ByteDance’s A/B testing system architecture across multiple layers, describes client‑ and server‑side experiment workflows, shares statistical best practices, and presents real‑world case studies illustrating hypothesis generation, evaluation, and future industry trends.
ByteDance’s data platform, built over seven years, now handles exabyte-scale data and over 100 million TPS, using a hybrid “middle‑platform + Business Partner” model, custom engines like ClickHouse/ByteHouse, agile governance, and a suite of products to support internal and external businesses, illustrating large-scale big-data engineering practices.
This article reviews recent advances in speech translation, discusses ByteDance's practical deployments, compares cascade and end‑to‑end modeling approaches, introduces improved encoder‑decoder architectures and training strategies, and reports state‑of‑the‑art results on the IWSLT 2021 benchmark.
In an InfoQ interview, ByteDance’s data platform leader explains how the company operates over 15,000 ClickHouse nodes storing more than 600 PB of data, the architectural choices behind its analytics workloads, cloud‑native transformations, and the practical lessons learned from large‑scale deployments.
The article examines ByteDance’s terminal technology team, outlining current front‑end trends, the challenges of complex processes, component chaos, and massive architecture, and presents their veMARS solutions for DevOps, component management, and application frameworks to improve platform development efficiency.
Shifting from growth-focused tactics to user‑centric product development at ByteDance, the author shares mindset changes, team rituals like daily TikTok reviews, and how these efforts drove CapCut’s rapid growth while significantly deepening personal product thinking.
ByteDance’s AI‑Lab leader Dr Yin Xiang discusses how the company’s intelligent speech technologies—spanning voice synthesis, recognition, and multimodal interaction—have been integrated across its global content platforms since 2017, boosting productivity in short videos, audiobooks, education, and more.
This week’s WecTeam Front‑end Weekly spotlights two technical deep‑dives: a Swift‑Objective‑C mixed‑compilation technique that slashes build times by 35%, and ByteDance’s large‑scale data‑tracking governance framework that underpins its trillion‑plus real‑time analytics pipeline.
ByteDance’s multi‑billion‑dollar acquisition of VR maker Pico mirrors Facebook’s Oculus purchase, intensifying a tech rivalry that spans metaverse ambitions, short‑video dominance, and aggressive investments in VR hardware, gaming, and social platforms.
This article details ByteDance's implementation of an exabyte‑scale data lake using Apache Hudi, covering scenario requirements, engine selection, functional support, schema management, extensive performance tuning, and future directions, while also noting recruitment opportunities within the team.
This article explains how ByteDance extended the Apache HDFS architecture with a multi‑datacenter design, introducing components such as DanceNN, NNProxy, and BookKeeper to achieve scalable storage, cross‑datacenter data placement, and rack‑level disaster recovery for petabyte‑scale workloads.
This article reviews ByteDance's transition from a fragmented, legacy front‑end toolchain to a unified modern web development platform, describing the shortcomings of traditional scaffolding, the rise of meta‑frameworks, and the design and capabilities of the internal MWA and Jupiter frameworks that power the open‑source Modern.js project.
The article recounts the winning project of ByteDance’s second Intelligent Creation Hackathon—a low‑cost virtual avatar production platform that uses 2D material‑driven technology and facial analysis to create 3D‑like avatars for meetings, live streams, and games, highlighting its technical details, team story, and future vision.
ByteDance’s data centers, including Douyin’s, operate with terabit‑level outbound bandwidth, housing up to hundreds of thousands of servers across China and abroad; the article breaks down server counts, total export bandwidth, CDN acceleration, and the network designs that enable billions of daily video streams.
The article examines how ByteDance, Douyin, TikTok and other Chinese tech giants operate massive data centers with terabit‑level outbound bandwidth, millions of servers, and extensive CDN and load‑balancing architectures to support hundreds of millions of concurrent users.
The ByteDance Volcengine meetup showcased four expert talks covering RTC scalability for interactive entertainment, zero‑latency VOD optimization, cutting‑edge H.266 video codec development, and reliable live‑stream event practices, offering deep technical insights into powering billion‑user short‑video services.
The article explores how TikTok (Douyin) and its parent company ByteDance operate massive data centers with terabit‑level outbound bandwidth, thousands of servers, CDN acceleration, and multi‑link designs to support hundreds of millions of concurrent users, detailing server counts, bandwidth estimates, and the underlying networking technologies.
The article examines ByteDance's massive data‑center infrastructure, detailing server counts, multi‑terabit outbound bandwidth, dual‑link designs, CDN acceleration, and comparisons with other Chinese tech giants, illustrating how such scale enables seamless video streaming for hundreds of millions of daily users.
This article examines ByteDance’s self‑built data centers, estimating server counts, total outbound bandwidth reaching several terabytes, and the role of CDN acceleration and multi‑link designs that enable billions of users to stream video smoothly.
The article chronicles Liang Rubo’s journey from a university roommate of Zhang Yiming to ByteDance’s trusted successor, highlighting his technical background, diverse responsibilities, and the organizational challenges he must navigate as the company expands globally.
This article examines the massive scale of ByteDance’s data centers, detailing server counts, outbound bandwidth estimates, dual‑link designs, and the role of CDN acceleration in delivering smooth video experiences to hundreds of millions of daily users.
An in‑depth look at ByteDance’s ultra‑flat hierarchy reveals how rapid project turnover, relentless performance pressure, and constant leadership reshuffles create a high‑stress environment that wears out middle managers and fuels a cycle of burnout and turnover.
ByteDance successfully delivered the Spring Festival Gala 2021 project within 27 days, leveraging advanced infrastructure and organizational strategies to handle unprecedented traffic and ensure a seamless user experience.
This article presents ByteDance's real‑world use of Apache Flink, covering the platform's overall architecture, SQL extensions, custom connectors, UI‑driven SQL platform, performance optimizations such as window mini‑batch and custom windows, dimension‑table enhancements, checkpoint recovery improvements, stream‑batch integration, and upcoming roadmap items.
This article analyzes ByteDance’s rapid growth by examining its product portfolio, commercial efficiency, algorithmic advantages, and recent expansion into education, gaming, automotive, literature, office tools, and healthcare, revealing why some products thrive while others lag behind.
The article recounts ByteDance’s 2016 decision to go all‑in on short‑video, analyzes the contrasting thinking models of its product manager, director, and founder Zhang Yiming, and explains how multi‑thinking, optimal‑solution, and super‑allocation mindsets drive strategic product success.
From Huawei’s 1995 global expansion to ByteDance’s TikTok surge, this article chronicles Chinese tech firms’ aggressive overseas strategies, the geopolitical push‑back from the United States, and the ongoing legal and political battles shaping their quest for worldwide market dominance.
TikTok’s explosive growth—surpassing 2 billion downloads and 800 million monthly active users—has been driven by aggressive acquisitions, localized content strategies, and massive marketing spend, but it now faces intense regulatory scrutiny and bans in key markets like India, the United States, and Hong Kong.
This article describes ByteDance’s four‑year series of customizations to Hadoop YARN—covering utilization improvements, multi‑load scenario optimizations, stability enhancements, and multi‑region active‑active deployment—along with practical production experiences, architectural details, and future work directions.
This article compiles practical experiences and technical lessons from developing Taopiaopiao mini‑programs on Alipay, Baidu, ByteDance and Quick‑App platforms, covering architecture choices, common pitfalls, component adaptation, template syntax, login flows, card integration, multi‑end differences and maintenance strategies.
The article analyzes the fierce competition in China's search market, comparing WeChat's cautious, low‑key approach to ByteDance's aggressive expansion, and explains how content ecosystems, strategic partnerships, and product‑level decisions shape the future of mobile search services.
This article presents an in‑depth overview of ByteDance’s large‑scale HDFS deployment, describing its unique access layer, metadata and data layers, the evolution through multiple growth stages, and the key architectural improvements such as NNProxy, DanceNN, lock redesign, startup acceleration, and slow‑node mitigation techniques.
The author recounts a nerve‑wracking video interview with ByteDance, detailing questions on Django, Nginx, uWSGI, Redis, data structures, message queues, a live coding task, and shares hard‑earned advice to help future candidates prepare effectively.
This article dissects Toutiao’s 2012‑2013 B‑round financing, its early product positioning, the tension between information efficiency and social ties, team dynamics, growth metrics, and the broader lessons for entrepreneurs and product managers navigating rapid scale.
This article examines ByteDance’s unique management model—transparent information flow, trust‑based culture, context‑over‑control principles, and rigorous OKR practices—that together enable a 5‑digit employee base to scale quickly while maintaining high performance and innovation.
ByteDance has entered the Chinese search market with its new mobile‑first “Toutiao Search”, assembling a team from top tech firms and leveraging AI, NLP and computer‑vision technologies, sparking a fresh rivalry with Baidu’s long‑standing dominance.
This article examines how ByteDance’s rapid growth and content ecosystem have challenged traditional Chinese internet giants like Tencent and Baidu, analyzing the strategic battles over user engagement, algorithmic recommendation versus social recommendation, and the implications for product management and market dynamics.
The article examines ByteDance’s rapid rise to a $75 billion valuation, detailing its product‑by‑product expansion from early apps to TikTok, its strategic investments across content, education and gaming, and the organizational tactics that turned the company into a dominant ‘app factory’.
An in‑depth personal account details how the author prepared for ByteDance’s (TouTiao) recruitment, outlining a month‑by‑month study schedule covering Java, big‑data technologies, algorithms, and system fundamentals, describing each interview round, sharing successful test strategies, and offering practical advice for landing offers at top tech firms.
This article presents Gong Yunfei's 2019 talk from the Beijing Apache Flink x Apache Kafka conference, detailing how ByteDance leverages Kafka for its large‑scale streaming and data processing needs.
The article analyzes ByteDance's strategic pivot from consumer‑focused apps to B2B solutions like Lark, explaining market pressures, investment history, and how this shift could reshape competition among Chinese internet giants.
