When XI met Joe
This is post originally appeared in Pascal's newsletter.
Click here to read it and more.
The main China-related event in November occurred when Xi Jinping met Joe Biden in San Francisco at the APEC (Asia-Pacific Economic Cooperation) summit. Until a week before the meeting, it was unclear whether Xi would attend, but he did. Expectations were set very low on both sides, and so any news would be good news.
Building on APEC’s November 2022 meeting in Bali they discussed the principles moving forward the U.S.-China relations. After a year of US-China turbulence with weather balloons, chip wars, a lack of bilateral military communication and a Chinese economy in trouble, the restoration of US-China presidential talks has to be seen as a mutual effort to avert a crisis and the start of a positive momentum into the relationship. For Chinese, it is fine to like the US again. For American (business people) it is fine to continue making money in China.
The outcome was positive, because five critical communication disconnects were patched together again. All five could potentially lead to disasters for the world.
- A military-to-military dialogue. Having a hotline is extremely important with regards to the South-China Sea tensions, especially when it comes to Philippines-China dispute. It’s quickly escalating into a potential hot conflict. The next war?
- Taiwan. Xi Jinping claimed that China doesn’t intend to use military force to resolve its demands over the island. With the Taiwan election coming up next January, the ‘Taiwan independence’ trump card could be picked up in the US election game.
- Hegemony. Xi re-iterated once again that China does not seek to act as a hegemon. In my view, the American misreading of China’s intensions to become the number one, creates the dangerous binary view about China that could lead into a Thucydides trap.
- The environment. Both countries, the world’s biggest carbon emitters, pledged to work together again; declaring the climate crisis “one of the greatest challenges of our time’. In advance of COP28, it was a key signal despite the absence of both leaders at the summit.
- Fentanyl and AI. China agreed to crack down on chemical firms that contributed to 75,000 overdose deaths in the US last year. We won’t have a ‘revenge opium war’. The two leaders also discussed AI and agreed that the technology is a real risk to be averted when used in military or nuclear operations. National regulation, guardrails and governance on AI are of utmost importance. US-China talks on AI can help to understand each other's constraints and to exchange best and worst practices for safe deployment.
From a business side, China recorded its first ever foreign direct investment (FDI) deficit last quarter, reflecting the concerns of foreign enterprises, yet it may also be due to repatriation of earnings, according to Goldman Sachs. Positive signs of a recovery in foreign business trust in China were visible from the sixth edition of China’s International Import Expo (CIIE) held in November, where 78 billion USD in deals were signed. In October Chinese imports rose unexpectedly with 3% to declined again in November by 0,5%. Chinese exports grew for the first time in six months with just 0,5%.
Could this be the turning point for the world to get its confidence back in China’s economy? I do believe so. I expect 2024 to beat expectations – also due to the low expectation that the media and economists have set. FDI should bounce back as China is still the second largest and growing economy. The Chinese government has adjusted course since the summer to now support business, lower geopolitical tensions with U.S. and deal with real-estate and debt crisis head-on by adding more liquidity into the market and giving more room for lending.
After almost four years of crisis management in China – covid, real-estate, local debts, trade, geopolitical, national security, energy, climate, tech war, FDI, ... – we are now likely entering an upward cycle of recovery, evaluation and strategy building. This could be the best time for investors, businesses and entrepreneurs to invest in China. 35 years of China experience has taught me one thing: In China, when the going gets tough, the tough get going!
CHINESE INNOVATION – NOVEMBER 2023
(Click the links in titles below to read a larger news article on each topic)
The biggest barrier for consumers to buy electrical vehicles (EVs) is price and driving range. China is solving both at lightning speed. Price is lowering fast due to China’s (over)capacity of new EVs, a price war, leap in high-quality, and an incredibly efficient supply chain. As for range, that is where battery manufacturers like BYD and CATL have been innovating faster than the competition.
Chinese battery manufacturer CATL announced that its CIIC skateboard platform completed testing in China and achieved a 1,000 km range at a consumption of 10.5 kWh/100km, and managed to charge 300 km in just 5 minutes. The battery pack efficiency was 75%, and in low temperatures of -7 degrees, the range was merely reduced by 30%.
The CIIC (CATL Integrated Intelligent Chassis) skateboard integrates batteries, electric motors, and other critical units into a single platform, lowering production cost, vehicle weight, and energy consumption while maximizing passenger space. CIIC brings a high level of integration between the car and chassis, reducing cost and the overall development cycle. Car and battery are one. The first automaker that will use the CIIC in a production model will be Hozon Auto (picture on top). This new concept could become a new competitive disadvantage for foreign brands who often hold firmly to their existing chassis designs and platform R&D.
In last newsletter, we mentioned how China is building the world’s largest deep-sea telescope to hunt for cosmic neutrinos. Now, China began building an underwater data center with performance equal to 6 million PCs.
The installation of the world's first commercial undersea data center will not only store data, but also serve as an undersea "supercomputer". It benefits of higher computing power density, better security, and as it also provides a dust-free, oxygen-free environment, it further protects server electronics and reduces faults.
Weighing roughly the equivalent of 1,000 cars, the equipment has been placed at a depth of 35 meters on the coastal area of Hainan Island. The equipment is being monitored remotely on-shore. On the seabed, seawater can provide natural cooling for the data modules. This data center will save 122 million kilo-watt hours, 68,000 square meter of land and 105,000 metric tons of fresh water per year compared to a traditional land-based data center of same size.
Early November, Alibaba launched a RISC-V chip to boost performance of its cloud data centers. The Zhenyue 510 chip is a controller IC for enterprise SSDs. An SSD controller is responsible for managing the underlying NAND flash memory in a server, and handling data input and output requests from the host – supporting applications such as AI training, online transactions and big data analysis.
Late November, Alibaba claimed they had built the first commercial server powered by a processor designed on the RISC-V CPU architecture. The system, using indigenous Sophon SG2042 chips, is a gigantic 3,072-core server with 48 nodes. It's the first cloud-facing commercial server built with RISC-V processors. Each processor has 64 cores, with a 2GHz frequency, 64MB system cache and connectivity through PCIe 4.0.
RISC-V is an open-source instruction set architecture (ISA) for designing chips. It's based on Reduced Instruction Set Computing (RISC) principles, aiming to be simple, efficient, modular, and adaptable for a wide range of applications. The key difference with the popular ARM or x86 architectures is that it’s free to license – so perfect for Chinese designers to circumvent the US chip bans. It means China has beaten the US in a key milestone in the race to expand the RISC-V ecosystem and the adoption of the embryonic chip architecture.
Three Chinese cancer drugs have recently been approved by the American Food and Drug Administration (FDA). Toripalimab, developed by Shanghai Junshi Biosciences is a monoclonal antibody used for the treatment of melanoma and nasopharyngeal carcinoma.
Fruquintinib, developed by the Chinese HutchMed, is an oral medication to treat adults with previously treated metastatic colorectal cancer. Efbemalenograstim alfa, is an injectable drug developed by Evive Biotech, to treat chemotherapy-induced neutropenia, or low levels of a type of white blood cells.
For example, Fruquintinib, to treat colorectal cancer – which begins in the colon or rectum – is the third most commonly diagnosed cancer, accounting for 7,8% of all cancer cases in the U.S., and third leading cause of cancer related deaths. For more than a decade there has been limited innovation for patients with metastatic colorectal cancer. Today, patients receiving Fruquintinib have their risk of death cut by 34 per cent.
In China, the cost of a single-dose vial of Toripalimab is around US$280. The wholesale cost in the US will be 31 times more or US$8,892.03. The price of Fruquintinib will have similar price spike in US. This is however still a competitive price in the US market. In China, the price of the drug is based on enrolment into the national healthcare insurance system which enrolls 95% of Chinese citizens already. The Chinese regulators put a price cap on medicine.
Drug companies in the US have claimed the high cost of R&D is behind the price of drugs in the country. However, a study published last year in the Journal of the American Medical Association challenged this idea. They found that there was no association between estimated research costs and the drugs. Drug companies charge what the market will bear, and that is made obvious from the price difference of Chinese drugs in China and in the US, but also from the price discounts American Pharma companies apply when selling into China. The US has the highest prices for new drugs in the world.
It is clear that China has just started to flood foreign markets with drugs developed in China. Chinese medicinal drugs could create an export tsunami like Chinese Electrical Vehicles did. China has caught up with Western BioPharma companies in a wide range of drug innovations and will produce even more medicines as long as they get FDA approvals for their products.
Chinese cities soar in science ranking while US and European rivals drop. The “Leading 200 Science Cities” list – maintained by the academic journal Nature – tracks research output. The Nature Index measures two different scores – “count” and “share”. A city receives a count value of one for each article with at least one author affiliated with an institution in that city. For the share score, each article is assigned a value of one, which is split equally among all authors. The sum of these values is the city’s count score.
When Nature first launched the global science city ranking in 2018, just 10 Chinese cities ranked among the top 50. Six years later, that number has already doubled. Five out of the top 10 cities on the 2023 list are in mainland China. The top five cities – Beijing, New York, Shanghai, Boston, and the San Francisco Bay Area – remained unchanged.
The other 18 mainland Chinese cities that made the top 50 jumped to a higher ranking compared to last year, while most major cities outside China, including Tokyo, Paris and London, fell. The only exception was Seoul, which rose two spots.
A total of 32 cities in mainland China made the 2023 list, and almost all of them climbed up the rankings compared to their positions on last year’s list, with only one exception: Taiyuan. The city of Hefei in southeast China – which was very poor two decades ago – even surpassed both Los Angeles and London in its share score, rising by three places to 13th place on this year’s list.
Chinese have produced a light-based chip that is significantly faster and more energy efficient than current high-performance AI chips when it comes to performing some tasks such as image recognition and autonomous driving. The new chip – known as the All-Analogue Chip Combining Electronics and Light (ACCEL) – is light-based and uses photons, instead of electric current, for computing and transmitting information to achieve a faster computing speed. The breakthrough innovation comes from a computing framework that fuses photonic and analogue electronic computing.
In a laboratory test, the new chip reached a computing speed of 4.6 peta-floating point operations per second (PFLOPS), 3,000 times faster than one of the most widely used commercial AI chips - Nvidia’s A100 (subject to US sanctions on China). The Chinese chip also consumes 4 million times less energy.
Its low power consumption may help overcome the problem of heat dissipation, which currently places a significant barrier in further miniaturizing ICs. Although the new chip cannot immediately replace those used in devices such as computers or smartphones, it may soon be used in wearable devices, electric cars or smart factories. The new chip was built by China’s SMIC using a 20-year-old transistor fabrication process. Improvements are therefore still very likely to happen.
Chinese are finding the light at the end of the AI tunnel which the US has been digging for China.
7. China Generative AI updates
Here are some of the China newsflashes of November on Generative AI.
For 300,000 RMB (41,190 USD), you can get your own personal digital person powered by Baidu’s ERNIE Bot language model. It’s readily available on digital human Du Xiaoxiao’s Taobao shop. It also stocked with single-purpose ERNIE-driven chatbots, such as one that flatters you, one that comforts you, or even one that purports to “reset your life”. Not sure what that means in China?
Alibaba eCommerce platforms, Taobao and Tmall, have debuted 10 free AI-powered features for merchants as well as a generative AI chatbot named Wenwen to answer user’s queries on Singles' Day. Personalization is going to be the way forward. It will make it a lot easier for consumers to get custom recommendations based on what they’re looking and will also play a crucial role in advertising strategies for Singles' Day to increase conversion rates.
A Chinese court just awarded copyright protection to AI-generated images in a landmark ruling. They ruled that it possessed "originality” and reflected human creative intellectual input – and should be recognized as works protected by copyright law. The logic was that it was a human being who set the relevant parameters for the AI model and ultimately selected the image in question. The ruling is in stark contrast to the stance of Western courts.
The 2023 World Internet Conference (WIC) issued the "Developing Responsible Generative Artificial Intelligence (AI) Research Paper and Consensus. It provides a reference for all stakeholders on AI development, as well as to promote and coordinate AI governance and build an open, fair and effective technological environment for the benefit of the world.
Didi (the Chinese Uber) is leveraging advanced AI models to offer more intelligent and efficient travel solutions. These include selecting optimal flights and hotels and seamlessly integrating with third-party ticket booking platforms and corporate reimbursement systems.
Ctrip (China’s booking.com) also launched its first vertical generative AI model in July for the tourism industry called "Ctrip Wenda”.
The world’s fourth largest smartphone manufacturer OPPO officially unveiled its upgraded Pantanal cross-platform smart system and self-trained large language model-AndesGPT which empower the brand new ColorOS 14, and the exploration in an open ecosystem for global developers. OPPO has established AI-related partnerships with more than 45 key institutions in China and abroad. OPPO is working with 320,000 developers and 750,000 creators to serve 600 million OPPO users.
The professional plan gives users access to make 100 inquiries every three hours to the recently released Ernie Bot 4.0 at US$8.18 per month. This is cheaper that ChatGPT 4, which according to Baidu’s founder Robin Li is by no means inferior compared to GPT-4. Baidu’s move comes as Chinese tech firms race to monetize and commercialize their large language models.
Bytedance’s (TikTok’s parent company) released its generative AI bot Feishu Intelligent Buddy. It is being added to its office app to help users generate emails, spreadsheets, mind maps, survey forms or analyse content from PDFs, video and audio files. The Feishu bot is able to work off a number of different large language models as rivals bake their own home-grown models into their office suites.
Pinduoduo (TEMU’s parent company) offers up to US$8,451 per month for roles including large language model developers. Candidates for the LLM referencing engineer role will be responsible for the research, development and optimisation of the company’s LLM reference engine. It is not clear yet whether Pinduoduo will launch its LLM publicly and how it intends to use it within the business.
Tencent is integrating several of its social media tools including WeCom, the enterprise version of ubiquitous messaging app WeChat, with its in-house large language model (LLM) Hunyuan, to develop industry-specific AI applications for China’s online pharmaceutical and healthcare market, which is expected to be worth US$311.5 billion by 2026.
Chinese start-up 01.AI founded by computer scientist Lee Kai-fu (author of AI Superpowers) has become a unicorn in less than eight months on the strength of anopen-source GenAI model called Yi-34 (trained on 34 billion parameters) that outstrips Silicon Valley’s best, available in Chinese and English. The company admitted recently that its AI model is built on Meta’s Llama.
If you know of any other breaking China innovation news from September that I missed, do let me know so I can add it in next month’s newsletter! Special thanks to Jurgen Coen for sharing some innovation links.
This is post originally appeared in Pascal's newsletter.
Click here to read it and more.