America continues to lead the AI race even amidst the recent product developments coming out of China. The Council on Foreign Relations puts China six to eight months behind the frontier models and argues that maintaining even that distance has made Chinese laboratories more dependent on American technology. On the underlying capacity, analysts estimate the United States holds an advantage of 21 to 49 times in AI compute produced during 2026, even in a scenario where chip exports to China are fully permitted.
The contest has now shifted from what the companies build to what governments will allow. On July 7th Bloomberg reported that American officials are looking at how Chinese firms rent computing power in other countries, which is legal today. That follows guidance issued on May 31st saying a Chinese company needs a licence to buy advanced chips even when the office doing the buying sits outside China.
Washington has changed its mind twice on selling chips to China. In January it allowed the sales, capped at a volume analysts put near a million units. The sales never happened. Security scrutiny from both countries stalled these sales. A bigger rewrite of the rules is now promised before September 30th, and it will take effect immediately, without a comment period.
The argument has since also moved to the usage of Chinese systems by American firms. The administration is considering rules that would stop American businesses and cloud providers from using Chinese AI, in the same way Chinese cars, robots and routers are already blocked. Treasury Secretary Scott Bessent has called the way Chinese firms learn from American systems intellectual property theft. American companies do not agree. Twenty-five of them, including Nvidia, Microsoft, Meta, OpenAI and Google, signed a letter warning that restrictions would push the work abroad. Anthropic did not sign it.
China is finding workarounds for this problem. Beijing has drafted a $295 billion plan for a national computing network running on 80% domestic chips. The difficulties are the factories, the memory supply and the software. One large Chinese firm lost three months of work when it moved from American hardware to a local alternative. The switch is being reflected by the current market share. Chinese chips took close to 41% of their home market last year, roughly half of that from Huawei. Before 2023 foreign suppliers held 90% or more.
America’s position in the race for dominance depends on a very large bet. AI companies produced 85% of the gains in the S&P 500 this year. Data centre building is half of all business investment. More than 80% of global venture money in the first half went to AI startups. That spending ends up at buildings that require copious amounts of power which is currently a major limitation. Bank of America expects data centres to add about 125 gigawatts of American demand by 2030, and the turbines needed to supply it are already spoken for.
What does this mean for Sri Lanka?
The exposure primarily comes through the IT sector. ICT and BPM exports earned $150.49m in June, up 11.31% on the year, while services exports as a whole grew 3.75%. Cheaper machine-written code lowers what clients will pay for the coding and back-office work that produced most of that growth.
The second channel is the data centre plan. Deputy Minister of Digital Economy says procurement is underway to set up AI data centres for local and regional customers, and to earn foreign exchange from them. Sri Lanka AI Week runs from September 28th to October 4th, against a target of $5bn in technology exports.
Looking at the labour market, birth rates are falling and a large share of the workforce is moving towards retirement. Over the medium to long term, it leaves the country trying to grow with fewer people in the working population. This concern raises the question of whether capital can do what labour used to? If the lack of labour results in a situation where growth is affected, could we plug this gap by using AI-augmented capital? Much of this is still unknown, but if this is a possibility it would be a considerable shift from the traditional workflows Sri Lanka is currently used to.