Fall 2026
Multimodal Algorithm Researcher Intern
Omni-Modal
Updated on 8/20/2026
Tencent
10,001+ employees
Global tech conglomerate: social, gaming, cloud
Compensation Overview
$38.54 - $60/hr
Palo Alto, CA, USA
In Person
Bachelor's
 AI & Machine Learning (1) |
|---|
Get referred to Tencent
See people who can refer or advise you
- A Bachelor's degree or higher in Computer Science, Artificial Intelligence, Mathematics, or a related field is required.
- A solid foundation in deep learning algorithms and practical experience in large model development is required.
- Proficiency in the underlying implementation details of deep learning networks and operators, model tuning for training and inference, CPU/GPU acceleration, and distributed training and inference optimization is required.
- Strong learning agility, communication skills, teamwork, and curiosity are required.
- Conduct research and development of Omni multimodal large models, including designing and constructing training data, foundational model algorithm design, pre-training, supervised fine-tuning, reinforcement learning optimization, model capability evaluation, and exploring downstream application scenarios.
- Scientifically analyze research and development challenges, identify bottlenecks in model performance, and devise first-principles solutions to accelerate model development and iteration.
- Explore diverse paradigms for achieving Omni-modal understanding and generation capabilities, research next-generation model architectures, and advance the capabilities of multimodal models.
- Hands-on experience in large-scale multimodal data processing and high-quality data generation.
- Familiarity with diffusion models and autoregressive models.
- Publication in top-tier conferences or experience in cross-modal research, such as audio-visual research.
- Practical experience with deep learning network and operator implementation, model tuning, CPU/GPU acceleration, and distributed training and inference optimization.
- Participation in Association for Computing Machinery or National Olympiad in Informatics competitions.
Tencent is a Chinese technology conglomerate that operates a wide range of consumer platforms and enterprise services. It connects over a billion users through WeChat and QQ, combining messaging, social features, and mobile payments, while Tencent Cloud offers AI, big data, and cloud infrastructure for businesses. It stands out by blending a huge user base with major investments in gaming studios and an integrated ecosystem that spans media, fintech, cloud, and enterprise tools. Its goal is to create a large, connected digital ecosystem for people and businesses in China and worldwide, using AI-powered products and services.
Company Size
10,001+
Company Stage
IPO
Headquarters
Shenzhen, China
Founded
1998
Get referred to Tencent
See people who can refer or advise you
Simplify's Take
What believers are saying
- 2Q2026 revenue rose 11% and marketing services climbed 22% on AI-driven ads.
- Domestic games grew 17% in 2Q2026, led by Delta Force and VALORANT.
- Tencent increased renewable electricity use to 48.5% in 2025, improving data-center economics and optics.
What critics are saying
- Tencent’s 2Q2026 free cash flow was negative RMB13.8 billion after RMB52.8 billion capex.
- The Pentagon blacklist and lost Washington lobbyists threaten U.S. partnerships through 2027.
- WeChat AI agents face execution risk; Xiaowei remains a small-scale test against ByteDance and Alibaba.
What makes Tencent unique
- WeChat and QQ still anchor 1.4 billion MAUs, giving Tencent unmatched distribution.
- Hunyuan 3 launched July 2026 with agentic features and open-source weights.
- Tencent blends gaming, payments, ads, cloud, and enterprise AI inside one ecosystem.
Help us improve and share your feedback! Did you find this helpful?
Benefits
Professional Development Budget
Growth & Insights and Company News
6 month growth
↑ 0%1 year growth
↑ 0%2 year growth
↑ 0%Tencent details progress toward Carbon Neutrality goals and outlines ai-era priorities. Aug 16, 2026, 21:00 ET New report outlines how infrastructure efficiency, renewable electricity and AI-enabled innovation can support lower-carbon growth SHENZHEN, China, Aug. 16, 2026 /PRNewswire/ - Tencent (0700.HK) published its Carbon Neutrality Mid-Term Report and launched an accompanying interactive microsite, sharing its progress to date and pathway toward achieving carbon neutrality across its own operations and supply chain by 2030. The report comes as the rapid growth of artificial intelligence is creating a defining energy challenge for the technology sector. AI workloads are raising rack power demand by up to tenfold, placing new pressure on electricity supply, cooling and infrastructure. At the same time, AI is emerging as a powerful tool for improving efficiency, integrating renewable energy and accelerating low-carbon innovation beyond the technology sector. The report showcases Tencent's response to the AI-energy challenge: improving the efficiency of computing infrastructure, aligning electricity demand more closely with renewable energy availability, and applying AI and digital technologies to support decarbonization beyond its own operations, while tracking progress toward its 2030 carbon neutrality commitments. As Tencent Chairman and CEO Ma Huateng writes in the report, "We are applying our digital and AI capabilities to cut emissions within our own infrastructure - and we believe AI holds far greater potential still: to accelerate low-carbon innovation, lift efficiency, and support broader environmental transformation in industries beyond our own." Redesigning infrastructure for the AI era One example is T-AIDC, Tencent's next-generation data center architecture for the AI era. AI workloads have increased rack power demand from 6-8 kW to 30-100 kW or more, placing new requirements on power delivery, cooling and system design. T-AIDC addresses these demands through an integrated architecture designed for high-density computing and delivers power-supply efficiency of up to 98%. Applying AI and innovation beyond our own operations Beyond its own infrastructure, Tencent is exploring how AI and digital technologies can support emissions reductions across a range of sectors. In steel production, AI-driven scheduling and digital twin technologies can help plants become more flexible in when and how they use energy, reducing emissions, lowering costs and supporting grid stability. Tencent is also working with partners to better align data center workloads with renewable energy supply through AI-powered scheduling, helping integrate more variable renewable electricity into energy-intensive digital infrastructure. Tencent is also helping climate technologies move from research to real-world deployment through CarbonX Program, supporting innovation in areas such as carbon removal, carbon utilization and long-duration energy storage. This work is complemented by TanLIVE, Tencent's climate intelligence platform, which helps connect governments, investors and climate-tech stakeholders while improving access to climate solutions, data and investment opportunities. Delivering progress toward carbon neutrality Alongside these broader efforts, the report demonstrates Tencent's continued focus on reducing emissions at source through energy efficiency, renewable electricity procurement and low-carbon infrastructure, guided by a carbon-mitigation hierarchy that prioritizes avoiding, reducing and replacing emissions. That approach is translating into measurable progress. Renewable electricity consumption increased from 22.0% in 2024 to 48.5% in 2025, while owned data centers achieved an 82.9% renewable electricity share. Since announcing its carbon neutrality commitment, Tencent has procured more than 6.5 billion kWh of green electricity. Sharing progress to advance the conversation By sharing both its progress and the work behind it, Tencent aims to contribute to a broader conversation about how the technology industry can manage the energy demands of AI while using digital innovation to support the transition to a lower-carbon economy. Explore the Carbon Neutrality Mid-Term Report and interactive microsite at https://www.tencent.com/climate/ For media queries, please contact: [email protected] About Tencent Tencent is a global technology and entertainment company focused on creating connections and experiences that matter. Founded in 1998, Tencent is driven by its mission to create "Value for Users" and apply "Tech for Good." Tencent's communication and social services connect more than one billion people around the world, helping them to keep in touch with friends and family, access transportation, pay for daily necessities, and even be entertained. Tencent also develops and publishes some of the world's most popular video games and other high-quality digital content, delivering rich and immersive interactive entertainment experiences. Tencent also offers a range of services such as cloud computing and other enterprise services to support our clients' digital transformation and business growth. Headquartered in Shenzhen, Tencent has been listed on the Main Board of the Stock Exchange of Hong Kong since June 2004. SOURCE Tencent
Daily AI Digest: Trimble's office agent meets WeChat's Xiaowei. This is the GolemWorkers daily AI digest on practical agents that can act, coordinate and finish work. Today: one agent spans an entire business, another enters WeChat, and enterprise AI is shifting from answers to completed work. Trimble launches one agent for the whole business. Trimble Arc Agent is live for organizations using supported Trimble products. From one conversational interface it can retrieve data, draft documents, create records, send messages and complete multi-step work, while following role permissions, approval paths and a recorded audit trail. Tencent tests a hands-free agent inside WeChat. Tencent's Xiaowei trial moves beyond answering questions: inside WeChat it is meant to act across daily services, such as finding a restaurant, making a reservation and placing an order. An early 24-hour test found a promising assistant that could complete some routines but still stumbled on others. Enterprise agents spread beyond software teams. OpenAI's latest enterprise data shows agents moving beyond software work. Since February, weekly Codex use grew 108-fold in legal, 41-fold in sales and recruiting, and 26-fold in marketing; Virgin Atlantic reports using agents to refactor legacy code and complete competitive research, with permissions, governance and human review around the work. OpenClaw beta strengthens secrets, backups and work journals. OpenClaw 2026.8.1-beta.2 can bind stored secrets to exact HTTPS hosts, warn before installing untrusted executable plugins and create verified SQLite backups. It also adds an optional Logbook that turns paired-device snapshots into a private work timeline, but the beta's full post-release validation was not green and it should not be treated as a stable upgrade. Claude Opus 5 targets longer work and stronger self-checks. Anthropic has made Opus 5 the default model on Claude Max and the strongest option on Claude Pro. The company reports gains in end-to-end business tasks and computer use, while early users describe agents that monitor systems, revise decks, review legal documents and stop for human judgment instead of rushing ahead. Agent idea of the day Build an agent that keeps wildfire smoke out of your home. What this agent does. Watch indoor and outdoor air quality, explain what changed, and recommend the safest next action before smoke or stale air becomes a surprise. Best for: Households in wildfire-prone areas, people with an air purifier or ERV/HRV, and anyone who wants useful alerts instead of another dashboard. Give it. * Indoor and outdoor PM2.5 readings from sensors or a trusted weather API * The state of your ventilation, purifier and any approved smart plugs * Your household threshold, quiet hours and recovery rules * One notification channel for alerts and approval requests Tell it to. * Read indoor and outdoor PM2.5 every 10 minutes and keep a timestamped trend. * Flag outdoor readings above the household threshold for 30 minutes and ask before closing windows or pausing fresh-air intake. * Flag rising indoor readings and ask before running an approved purifier or Corsi-Rosenthal box. * Recheck conditions until they remain safe, then recommend reversing any temporary action. * Send one concise event report with readings, actions proposed, approvals received and the time conditions recovered. Run it: Run every 10 minutes, with an immediate alert when a configured PM2.5 threshold persists for 30 minutes. You get. One quiet, time-stamped alert that says what changed, what action needs approval, what the agent will keep watching and when it will check again. Keep a human in control. * Do not give medical advice or modify HVAC wiring. * Never disable ventilation or switch powered equipment without explicit, time-limited human approval and a tested smart-plug boundary. * Treat stale or conflicting sensor data as an alert-only condition. * Keep a time-stamped record of readings, recommendations, approvals and reversals. Feasibility: A published Home Assistant setup uses indoor and outdoor PM2.5 readings, smart plugs, an ERV and a Corsi-Rosenthal box to pause polluted outdoor air, clean indoor air and signal household conditions. The agent recipe adds explanation, trend review and human approval around those proven actions. Source: Highly Probable
Nexon wants to replicate ARC Raiders success with future games. Nexon is looking to replicate the success of ARC Raiders across its future portfolio after the shooter became a major contributor to the company's revenue and provided evidence that its Western expansion strategy can work. ARC Raiders generated more than ¥18.3 billion in revenue during Q2 2026, representing approximately 15% of Nexon's total quarterly revenue. The game has now surpassed 16.3 million units sold worldwide and generated more than ¥88 billion in revenue since launching in October 2025. Nexon President and CEO Junghun Lee said the game's performance represents more than strong commercial results. "ARC Raiders stands as proof that Nexon's Embark Studios team has developed a roadmap for creating and publishing games that can break through in highly lucrative Western markets dominated by legacy franchises. It's a new and potent capability for Nexon, and one we intend to replicate at scale." Nexon plans to support the shooter with Frozen Trail, its largest content update since launch, in October. The update will add a new map, weapons, enemies, gadgets and outfits, as well as a paid Premium Reward Pass. The company expects the update to help bring inactive players back while attracting new players. ARC Raiders is also being tested in China through a partnership with Tencent. A closed alpha held in July exceeded registration targets, while Nexon said early feedback regarding the game's localisation has been encouraging. Embark Studios CEO and Nexon Executive Chairman Patrick Söderlund also highlighted the game's development as evidence that the company's strategy of focusing on fewer, larger projects can succeed. "More recently, ARC Raiders has shown how the fewer, bigger thesis can work, offering a new model for development. A small team, working with new technology, a fresh idea, over 16.3 million units in nine months." The game's performance comes as Nexon attempts to diversify beyond its established franchises. Its three major franchises generated ¥81.2 billion in Q2 revenue, down 5% year over year, while newer and other titles increased 20% to ¥39.9 billion. ARC Raiders was a significant contributor to that growth. Nexon now plans to apply lessons from ARC Raiders and its MapleStory expansions to other projects. Upcoming releases include Arad: Idle RPG, DAVE THE DIVER Mobile, Azur Promilia and TEMPPAL: OVERGEARED in 2026, followed by several Dungeon and Fighter projects and other titles from 2027 onwards.
AI utilization becomes mainstream in China's game industry, offering immersion through natural conversations with characters Mainichi Shimbun 2026/8/13 13:54 (last updated 8/13 14:40) 456 characters In China's game industry, the use of generative artificial intelligence (AI) is becoming mainstream. At the game exhibition "ChinaJoy" held in Shanghai from late July to early August, works that allow conversations with AI characters were on display. China is leading globally in the application of AI to games and is increasing its presence in development as well. At ChinaJoy, around 900 companies from home and abroad exhibited, showcasing over a thousand new game titles. The event drew a total of over 430,000 visitors over four days. A crowd formed at the booth of "Yanyun Shiliu Sheng" (Japanese title: "Fūen Den"), an RPG distributed globally by NetEase, a leading Chinese game company. Using AI, the characters engage in natural conversation in response to players' questions and actions. Previously, characters could only speak set lines, but with the use of AI, they can give different responses to each player, and this is expected as a technology that enhances immersion. The use is also advancing in game development. Tencent Holdings, a major Chinese IT company, exhibited "LAP", an AI-powered game production management tool. It can provide materials for backgrounds and characters and create animations. (Kyodo)
Tencent rejected the opportunity to make immediate returns on its $53 billion quarterly capital expenditure, instead choosing to develop its own AI models for long-term gains. Company executives revealed during Q2 earnings that demand for computing resources is so strong, Tencent could rent its infrastructure at 30 per cent profit margins compared to costs paid months ago. Chief strategy officer James Mitchell said the company could recover depreciation costs "almost immediately" through rentals. However, president Martin Lau said Tencent is pursuing a different strategy, allocating substantial computing resources to build state-of-the-art models and AI applications for the Chinese market. The company plans to generate returns by selling tokens for services like WorkBuddy and CodeBuddy. Tencent released its 295-billion parameter Hunyuan-3 model in July, with larger versions planned.