New AI from China, sports and diplomacy, the future of robotaxis: what was written about in blogs

A startup from China released a GPT-4 level model - what conclusions did the market draw? Gulf countries and China are turning sports into a global tool of influence. Will unmanned cabs win the car market? The most interesting in economists' blogs - in our review. 

Last week, Moonshot AI, a Beijing-based startup in which Alibaba has invested, released Kimi K2, a large open source language model. The release shows that it outperformed Anthropic's Claude Opus 4 on two metrics and showed better overall performance than OpenAI's programming-oriented GPT-4.1 model. 

The new release conjures up unpleasant associations for the U.S. market with the release of Deep Seek in January. "This is another 'DeepSeek moment' where even those who thought they were on top of AI developments are being forced to revise their predictions for the coming years," writes Nathan Lambert, a researcher of large language models at the Allen Institute for Artificial Intelligence, on his blog.

The cost of training the model is not yet known, but Lambert assumes that it will be comparable to the previous Chinese high-profile premiere - the DeepSeek model, that is, it will be much cheaper than its Western counterparts. The developers of the DeepSeek R1 model claimed to have spent up to $6 million to train it, while OpenAI's costs for a similar model were more than $100 million;

What conclusions can be drawn? Lambert writes that simply restricting Chinese companies' access to U.S. chips will not be enough to "stop" the race for leadership in AI.

In addition, the Kimi K2 release shows that High-Flyer, the foundation behind DeepSeek, is not unique to China; there are other cutting-edge developers. Importantly, China has already surpassed its Western counterparts as open source models, Lambert writes.

Why didn't the release of Kimi K2 cause a stir in the market, whereas after the DeepSeek model was released this January, the US market lost $1 trillion?  While it is true that it is cheaper to create Chinese AI models compared to their Western counterparts, they may still be higher than their developers have voiced. Investors have realized - only the final stage of training AI models is inexpensive, according to Lambert.

The release did not go unnoticed at OpenAI, he suspects. After the release of the new Chinese model, company head Sam Altman wrote that OpenAI was delaying the release of its open-source model, which was scheduled for release this week. However, Lambert writes, OpenAI assures that the postponement has nothing to do with Kimi K2. 

In recent years, Gulf countries have spent huge sums of money on sports - buying European soccer clubs, building stadiums, investing in their own national teams. Funds from the Middle East have turned sports into a showcase for their capital: a member of Abu Dhabi's ruling family, Sheikh Mansour bin Zayed Al Nahyanan bought Manchester City, state-backed Qatar Sports Investments bought 70% of Paris Saint-Germain for €70 million, and Saudi Arabia's Public Investment Fund paid £305 million for 80% of Newcastle United in 2021 - to name just the most high-profile deals. Entire clusters of arenas are being created for the sake of hosting global tournaments: Qatar spent $229 billion to organize the World Cup in 2022;

So why are Middle Eastern countries investing so heavily in sports? This question is answered on his blog by journalist, co-author of the book "Freakonomics" and author of the podcast of the same name Stephen Dubner. One obvious reason is reputation whitewashing, or so-called sportswashing. Sports projects are becoming a soft power tool. Such investments soften relations with Western partners and strengthen political ties.

But it wasn't always this way. Emlyon Business School professor Simon Chadwick identifies three stages in the development of sport in his conversation with Dubner. The first stage (late 19th and early 20th century) is characterized by the global dominance of European federations (FIFA, IOC and others). Sport was then perceived as a social good rather than a way to make money or a political tool. The second stage came in the post-war years: the American model of professional sports (NBA, NFL) became commercially successful, minimizing the influence of the state and subsidies. Now we are in the third stage, when sport is becoming a full-fledged instrument of foreign policy, says Chadwick. This involves not only Europe and the United States, but also China, Saudi Arabia, India and other countries.

But behind the investment in sports is not only a desire to improve the image, but also a well-thought-out strategy to diversify the economy. As global dependence on oil and gas gradually declines, Gulf countries are looking for new sources of revenue, and sports are ideally suited to boost tourism, hospitality and related industries, explains in the podcast another of Dabner's interviewees, Kash Shaikh, CEO of Baseball United, the first professional baseball league in the Middle East and South Asia, which Dabner called "America's export of sports."

Chadwick quotes Saudi crown prince Mohammed bin Salman, who has declared that sports should contribute up to 3% of the kingdom's GDP. 

China is becoming a major player in this new "geopolitical game". For the last 10 years, the country has been striving to become a global leader in building infrastructure abroad, especially in Africa. An example is the 2024 Africa Cup of Nations, held in Côte d'Ivoire. Three of the four stadiums for the tournament were built by China, receiving priority access to the region's raw materials or political influence through preferential lending.

Goldman Sachs Research analysts in their latest thematic report described the impressive prospects of the unmanned cab market in the U.S. (Alphabet's Waymo service is actively developing there, and Tesla began operating in a limited mode this summer). The bank estimates that the robotaxi market in the US will grow by about 90% per year from 2025 to 2030. By this time, the gross profit of companies in the sector can reach $3.5 billion, operators of autonomous cars will be able to reach a gross margin of 40-50% in the next 3-5 years. Now a little more than 1.5 thousand commercial robotaxis drive on the roads of five U.S. cities. By 2030, their number will grow to about 35 thousand, analysts predict;

The key question today is whether this forecast is overly ambitious or, on the contrary, too conservative? The answer depends on whether robotaxi operators will be able to scale their business, as well as on the development of machine learning technologies in this sector.

As the market grows, the cost of autopilot cars will continue to decline, Goldman Sachs believes. Nevertheless, it will remain higher than in China, where the technology appeared earlier. In parallel, the cost of traveling by robotaxi will also decrease. If now one mile costs about 35 cents, by 2040 it may fall to 15 cents. The cost of insurance will also drop from 50 cents to about 23 cents per mile over the same period. Additional savings would come from reducing the number of operators. While one operator now remotely monitors only three cars, by 2040 he or she could be handling up to 35 cars.

Another question Goldman Sachs analysts are asking is whether Americans will be able to give up personal cars as the robotaxi market develops. The bank believes they will not. A personal car in the U.S. costs the owner about $1 per mile, while a cab ride costs more than $2 per mile. Therefore, even as unmanned travel becomes cheaper, many people will choose to keep their personal cars and, in the long run, purchase their own unmanned vehicles.

This article was AI-translated and verified by a human editor