eChina.news

Innovation

AI economics: rise of machines’ token use could work to advantage of Chinese models

Agentic AIs now use five times as many tokens as humans, giving lower-priced models from Chinese firms a competitive edge, analysts say The global artificial intelligence boom recently crossed a landmark threshold: autonomous agents now consume more than five times as much data as human users, triggering a surge in processing costs that is giving lower-priced Chinese models a competitive edge, according to analysts. Unlike traditional chatbots that simply respond to human prompts, these agents operate independently to execute multi-step workflows, such as writing software, conducting research or managing business operations. A single assignment can trigger a cascade of automated model calls as an agent plans its task, searches databases, invokes software tools and audits its own results. This shift has dramatically altered the economics of AI deployment. Agentic requests consumed about 15 times more tokens – the basic units of data processed by an AI model – than standard human queries, according to data from model aggregator OpenRouter compiled by venture capital firm Andreessen Horowitz.

September 4, 2026

Innovation

US-China AI agreement, G20 trade spat, inflation data

The United States and China joined other G20 nations in supporting light-touch regulation for artificial intelligence (AI) this week, creating rare agreement between the tech rivals and a potential foundation for further discussions ahead of an upcoming presidential summit. G20 nations agreed to adopt sector-specific rules for AI and avoid creating new AI regulators at a meeting in North Carolina. The Carolina Principles also call for close cooperation between governments and developers in...

September 4, 2026

Innovation

Apple poised to unveil most expensive iPhone yet, fuelled by memory price surge

TrendForce says 20 per cent increase is “moderate”, with memory costs for 256-gigabyte iPhone 18 Pro model up nearly 400 per cent in a year Apple’s next flagship smartphone, the iPhone 18 Pro, could cost as much as US$1,319 for the base configuration as skyrocketing memory costs push smartphone prices to record highs, according to a recent report. Prices for the new iPhone line-up were projected to rise by 10 to 20 per cent, driven by soaring memory costs, market research firm TrendForce said in a report released on Thursday. Compared with the baselines of US$1,099 for the iPhone 17 Pro and US$1,199 for its Pro Max variant, consumers would be facing a substantial price increase. However, TrendForce said the increase represented a “moderate pricing approach” by Apple, given that memory costs for the 256-gigabyte iPhone 18 Pro model were estimated to be nearly 400 per cent higher than at this time last year. Apple would absorb part of the soaring component prices to maintain market share, amid a weak global macroeconomic environment and more cautious consumer spending, TrendForce said.

September 4, 2026

Innovation

How China Works author on the data economy, robots and AI

Lan Xiaohuan is a professor of economics at China Europe International Business School. His book, How China Works: An Introduction to China’s State-led Economic Development, has sold millions of copies in China and has been translated into multiple languages. Here, he discusses the economic realities behind China’s record trade surplus, the case for a stronger social safety net, and how public data infrastructure shapes the artificial intelligence race with the United States. SCMP Plus readers...

September 4, 2026

Innovation

China’s robot revolution may not arrive in the way you expect

Electric vehicles generally travel abroad as finished, branded products. Chinese robotics could spread differently Unitree is a leading example of the first approach. Its strength in locomotion and motion control has helped push Chinese humanoids forward at remarkable speed, yet founder Wang Xingxing has also been candid about the software challenge ahead. He said last month that a breakthrough allowing robots to handle unfamiliar environments and tasks could come in two or three years in an optimistic case, or it could take as long as a decade.

September 3, 2026

Innovation

Are Z.ai and MiniMax heading down opposite financial paths months after Hong Kong IPOs?

Post-earnings results suggest an emerging divide as differing revenue growth and model capabilities may shape their market standing, analysts say When two of China’s leading AI pioneers went public in Hong Kong in January, they pitched investors on a shared promise: capturing the explosive demand for artificial intelligence at home and abroad. Their first-half earnings, however, suggest that narrative could be splintering into two different trajectories. Z.ai on Monday reported a nearly 400 per cent year-on-year surge in first-half revenue to 953.9 million yuan (US$142 million). By contrast, MiniMax’s revenue grew 283 per cent to US$116.6 million, according to its financial report released last week. The commercial gap appeared even wider when measured by annual recurring revenue (ARR), a metric used by software companies to project 12-month revenue based on current monthly subscriptions. Z.ai co-founder and chief scientist Tang Jie said during an earnings call that the company’s ARR had reached US$1.6 billion based on August’s results. MiniMax founder and CEO Yan Junjie, meanwhile, said its ARR had reached US$800 million in August – half that of Z.ai.

September 3, 2026

Innovation

ByteDance to expand massive AI data centre cluster in Inner Mongolia, source says

As Chinese tech giants ramp up artificial intelligence spending, TikTok’s parent company is set to expand data centres in Ulanqab TikTok parent ByteDance is seeking to expand its computing capacity by building out more data centres in north China’s Inner Mongolia autonomous region in the next two years, according to a person familiar with the matter, another example of a Chinese tech giant stepping up its investments in artificial intelligence infrastructure. The company had been in preliminary talks with local data centre vendors to add between five and six gigawatts’ worth of compute in the city of Ulanqab, the person said, who asked not to be identified due to the private nature of the information. Typically, the cost of upfront investment for a one-gigawatt AI data centre was around 160 billion yuan (US$23.8 billion), according to a research note published by Soochow Securities last month, meaning that ByteDance’s new facility – at between five and six gigawatts – could cost roughly 800 billion to 960 billion yuan.

September 3, 2026

Innovation

Chinese chipmaker Enflame 4,073 times oversubscribed in Shanghai IPO amid Nvidia race

The last of China’s ‘four little dragons’ to go public has drawn millions of retail investors, despite odds of securing an allocation at less than 0.03 per cent Enflame Technology, one of China’s leading AI chipmakers, has sparked a retail frenzy for its 6.12 billion yuan (US$910.9 million) initial public offering in Shanghai’s tech-heavy Star Market, underscoring growing investor appetite for home-grown alternatives to Nvidia. The Tencent Holdings-backed chipmaker saw the retail portion of its IPO oversubscribed by 4,073 times, with about 7 million online investors alone submitting orders for 42.1 billion shares, according to its filing to the exchange on Wednesday. The high demand amounted to an allocation rate for individual investors of just 0.025 per cent. That made it one of the lowest in mainland China this year alongside robotics maker Unitree Robotics, whose rate was just 0.018 per cent, or about one successful lot for every 5,500 applications. Priced at 142.18 yuan (US$21.16) per share, Enflame expects to raise 6.12 billion yuan to fund the research, development and production of its fifth- and sixth-generation AI chips.

September 3, 2026

Innovation

Frontier AI at a cost: what Anthropic’s Fable 5.1 means for the US-China model race

The US firm’s new models claim top position on leading benchmarks, though steep operating costs against Chinese rivals highlight a growing divide Anthropic’s powerful new Claude Fable 5.1 model has widened its lead in performance benchmarks over Chinese rivals, even as budget-friendly open-weight models from China continue to gain commercial traction globally. Fable 5.1 also claimed first place on San Francisco-based Vals AI’s index for handling complex, real-world tasks across sectors such as finance, coding and law. Anthropic’s earlier Opus 5 and Fable 5 models trailed closely in second and third place. The benchmark results prompted industry insiders to weigh in on whether leading labs in the United States were pulling further ahead of Chinese competitors. Yuchen Jin, a technical staff member at US artificial intelligence platform Databricks, called Fable 5.1’s capability leap “insane”, while some observers noted its performance shattered claims that Chinese open-weight developers had already closed the gap with the frontier. However, a stark price gap points to a sharp divide between the two countries’ approaches.

September 2, 2026

Innovation

China’s No 2 foundry Hua Hong invests US$2b in new fab to meet surging AI-driven demand

Project is via a joint venture with state-backed investment entities under China’s National Integrated Circuit Industry Investment Fund Major Chinese foundry Hua Hong Grace Semiconductor is pouring US$2 billion into a massive capacity expansion in the Chinese chip production hub of Wuxi, racing to meet skyrocketing domestic demand for AI infrastructure and bypass US tech curbs. The fresh capital would fund the construction of a new 12-inch speciality line, its third facility in the eastern Chinese city, the country’s second-largest contract chipmaker said in a filing to the Hong Kong stock exchange on Tuesday. Once fully operational, the expansion will add 55,000 wafers to its monthly production capacity, representing a roughly 30 per cent increase to its total capacity in Wuxi. Hua Hong and its Shanghai subsidiary are providing US$1 billion and US$1.1 billion respectively, holding a controlling 51 per cent stake, while the state entities will provide the remaining capital. The state capital injection will bring total funding for the project to US$4.2 billion, according to the filing.

September 2, 2026

Innovation

China’s AMEC unveils 6 chip-making machines in a day, boosting self-reliance ambitions

A rapid R&D drive is powering the firm’s transition beyond its core business The new line-up, announced at an industry conference in Wuxi on Tuesday, spans several steps in chip production, including etching tools that carve microscopic patterns into wafers, deposition systems that build ultra-thin layers of material and the complex equipment used to make silicon-carbide power chips. Unveiling six tools simultaneously marks a notable shift for AMEC as it accelerates its transformation from an etching specialist to an all-around semiconductor equipment firm. Four of the six new products are deposition tools. The global etching and deposition markets are dominated by US and Japanese giants like Lam Research, Applied Materials and Tokyo Electron. During an investor briefing last month, AMEC said it had cut its product development cycle from three to five years down to two years or less. The company poured 2.04 billion yuan (US$303.5 million) into research and development in the first half of the year – a 36.9 per cent year-on-year surge representing 30.5 per cent of its total revenue. It currently has more than 20 new machines under development across six equipment categories.

September 2, 2026

Innovation

Tencent’s Hy4 model gains in open-source AI rankings after ecosystem-driven training

On the DeepSWE benchmark, the Hy4 preview scored 64.3, surpassing Alibaba’s Qwen-3.8 Max at 56.6 and DeepSeek-V4 Pro at 62.7 Tencent Holdings’ use of its vast product ecosystem to train its new Hy4 preview model gives it an edge in developing AI agents and brings its flagship model suite back into the top tier of open-source offerings, according to analysts. The Chinese tech giant’s “differentiated product-plus-model strategy”, where preview models were first deployed across Tencent’s suite of products, enabled it to collect user data before feeding the information back into subsequent rounds of training, Goldman Sachs analysts said in a research note on Monday. “We view this closed-loop approach as particularly relevant for productivity and coding workloads, where real-world task trajectories, user interactions and evaluation signals drive model differentiation in the agentic AI era,” said the analysts led by Goldman’s head of Asia internet research Ronald Keung. They added that the Hy4 preview brought Tencent’s Hunyuan series back to the forefront of open-source models, with notable gains in coding capabilities over Hy3.

September 2, 2026