The competitive frontier in artificial intelligence is quietly shifting. From recruiting civil engineering talent to making major bets on intelligent computing centers, and now energy giants investing in embodied AI companies, AI is pushing tech companies’ asset boundaries into traditional resource domains like land, electricity, and data centers.
Recently, a job posting appeared on DeepSeek’s official website stating that “every generation encounters its own infrastructure revolution,” listing electrical engineering, HVAC, environmental engineering, and civil engineering among the required disciplines. This move sends a subtle signal: model companies that previously relied primarily on external computing resources are now extending deeper into computing infrastructure.
At the same time, energy state-owned enterprises are also moving deeper into the AI industry chain. On August 10, Unitree Robotics Co., Ltd. officially opened subscriptions, positioning itself as China’s A-share market’s “first embodied intelligence stock.” PetroChina’s Kunlun Capital Co., Ltd. and China Southern Power Grid Industrial-Finance Holding Group Co., Ltd. both appeared on the strategic placement list, drawing market attention. The offering combines strategic placement, offline issuance, and online issuance, with total strategic placement shares of 8.0893 million, accounting for 20% of the offering. Kunlun Capital and China Southern Power Grid Industrial-Finance participated as “large enterprises or their subsidiaries with strategic cooperation relationships or long-term cooperation visions related to the issuer’s business operations,” each receiving allotments of 903,290 shares worth approximately 136 million yuan (approximately $20.2 million), with a 12-month lock-up period.
On the other end of computing demand, AI concept stocks are seeing explosive earnings growth. According to statistics, over 100 A-share concept stocks are involved in the data center industry chain, with 45 having released semi-annual performance-related announcements. Among them, four turned losses into profits in the first half, two narrowed losses year-on-year, and 24 achieved year-on-year growth in net profit attributable to parent companies. Liton Technology and Fullhan Microelectronics both posted growth rates exceeding 1,000%.
Tech Giants Continue Organizational Restructuring
Organizational changes around AI are also accelerating among tech giants. ByteDance recently established a new Tier-1 division—AI Data and Security—operating at the same level as Seed, Flow, and Douyin, headed by Wang Yinglei. This follows the establishment of two AI Tier-1 divisions, Seed and Flow, at the end of 2023, marking another Tier-1 division ByteDance has created around its AI business.
Market sources also indicate that Fu Yue, head of AI Data and Security at ByteDance’s Seed division, will depart soon, with Wang Yinglei set to take over the role. ByteDance has not yet issued an official response to this news.
Overseas giants are also making strategic adjustments. Meta CEO Mark Zuckerberg published a lengthy post titled “The Future Belongs to Everyone,” stating that Meta will promote the popularization of “personal superintelligence” and continue supporting open-source AI models. He said open models help avoid excessive concentration of AI capabilities, allowing more individuals and enterprises to access advanced AI tools. Zuckerberg revealed that Meta’s Superintelligence Lab has been launched and will resume releasing some open-source models in the future.
OpenAI, meanwhile, faces personnel turmoil. As external scrutiny of its AI model development methods intensifies, multiple safety researchers have departed. Against this backdrop, OpenAI’s head of ethics, Chloe Bakalar, left less than a year into the role. This departure is one of several high-profile executive exits at OpenAI in recent weeks, alongside safety systems head Johannes Heidecke, chief futurist, and former goal alignment head Joshua Achiam.
Semiconductor and Computing Infrastructure Expansion Accelerates
Memory chip giant SK Hynix is restarting construction of its No. 2 NAND flash production base in Dalian, China, planning to expand capacity by approximately 50%. The Dalian No. 2 plant broke ground four years ago but remained stalled for an extended period due to the memory industry downturn. SK Hynix plans to bring in semiconductor production equipment by the end of this year and begin formal production in the first half of next year.
Microsoft is placing heavy bets on in-house chips. Sources familiar with the matter revealed that Microsoft plans to significantly increase production of its next-generation self-developed AI chips next year, hoping to attract major cloud service customers like Anthropic to adopt its own chips. The current-generation in-house chip, Maia 200, has seen relatively low market acceptance, but Microsoft is still pushing forward with expansion plans. Microsoft plans to officially unveil the new Maia 300 chip this fall, potentially as early as next month. The cloud computing giant has already held discussions with chip foundry TSMC, securing over 300,000 wafers of Maia 300 foundry capacity, with delivery scheduled for 2027—an order scale far exceeding the tens of thousands of Maia 200 chips Microsoft has produced to date.
Rapid-Fire Chinese AI Model Releases Draw Global Attention
Over the past two months, Chinese AI companies have been releasing next-generation large models at an almost relay-race pace. On July 16, Moonshot AI launched its next-generation large language model Kimi K3, ranking among the global top tier in multiple authoritative benchmarks. On July 31, DeepSeek officially announced the public beta of its latest large language model DeepSeek-V4-Flash API. According to testing data from U.S. research institutions, its operating cost is less than 1% of Anthropic’s flagship model Claude Fable 5. On the same day, ByteDance officially released its next-generation video generation model Seedance 2.5. On August 3, Alibaba released the Qwen 3.8-Max model, scoring on par with Anthropic’s Fable 5 and sometimes even surpassing it.
Bloomberg recently published a report noting that the rapid-fire release of new models from China’s AI sector is quickly narrowing the gap with Silicon Valley, creating a competitive environment dubbed the “death zone,” where players lacking cutting-edge technology or disruptive pricing may struggle to survive.
France’s Le Monde analyzed the reasons behind China’s rapid AI development from another angle. The report noted that Chinese engineers, constrained by chip limitations, have placed greater emphasis on algorithm optimization and resource utilization efficiency, extensively adopting advanced architectures such as “mixture of experts” models that activate only relevant modules when processing different tasks, significantly reducing computational costs. Most industry experts believe Chinese engineers have successfully turned “chip shortages” into an advantage.
Humanoid Robot Market Structure Takes Shape
In the humanoid robotics space, Chinese companies have established absolute dominance. According to data from California-based research firm Smart Analytics Global, global humanoid robot shipments totaled approximately 19,100 units in the first half of 2026, more than triple the 5,100 units in the same period last year. Among these, Chinese humanoid robot manufacturers accounted for over 97% of global market share.
Specifically, Shanghai-based AGIBOT shipped 8,400 units, representing 44% of global shipments; Hangzhou-based Unitree Robotics shipped 5,900 units, holding a 31% market share. Both companies’ shipment volumes far exceeded those of leading U.S. humanoid robotics firms such as Tesla, Figure AI, and Agility Robotics.
Multiple Industries Post Strong Earnings
Several companies released impressive semi-annual results. Guangju Energy issued a revised 2026 semi-annual earnings forecast, with the company now expecting net profit attributable to shareholders of listed companies of 65.56 million yuan to 68.56 million yuan (approximately $9.7 million to approximately $10.2 million), representing year-on-year growth of 603.02% to 635.24%. The revision was due to the company’s 14.42% equity stake in investee Mawan Electric Power, which confirmed 2025 dividend distributions at the end of the second quarter, adding 40.39 million yuan (approximately $6.0 million) in investment income for the reporting period.
Sea Limited achieved second-quarter GAAP revenue of $7.788 billion, up 48.1% year-on-year; gross profit reached $3.550 billion, up 47.3%; and net profit reached $458 million, up 10.6%. Yuewen Group posted first-half revenue of 3.53 billion yuan (approximately $524.3 million), up 10.7% year-on-year, with short drama and AI comic drama business revenue reaching 430 million yuan (approximately $63.9 million), more than triple the same period last year, and IP copyright revenue reaching 1.61 billion yuan (approximately $239.1 million), up 41.9%. Tingyi (Cayman Islands) Holding Corp. posted first-half revenue of 40.54 billion yuan (approximately $6.0 billion), up 1.1% year-on-year, with adjusted profit attributable to shareholders of 2.43 billion yuan (approximately $360.9 million), up 15.2% year-on-year.
South Korean President Lee Jae-myung on Tuesday called for large-scale strategic investments in future industries during a cabinet meeting, including small modular reactors, renewable energy, quantum computing, aerospace, and advanced biology. He stated that strategic investments in these areas should be pursued as thoroughly and intensively as investments in large-scale projects such as semiconductors and artificial intelligence.
A spokesperson for China’s Ministry of Commerce recently responded to U.S. plans to investigate Chinese AI companies, stating that China has consistently opposed the U.S. politicizing and weaponizing technology and trade issues, stigmatizing Chinese companies with unfounded allegations, and imposing sanctions. Innovation is not the patent of any single party, the spokesperson said, noting that Chinese AI companies have long been deeply engaged in fundamental research and have persisted in pursuing both technological self-reliance and open cooperation.
