Dragoman Digest

AI’s power demand creates unlikely commercial opportunities

Grid connection delays redirect data centres toward premium-priced, higher-emission alternatives
Data centre developers are increasingly turning to on-site fossil fuel generation as US grid connection wait times stretch to seven years. Data centres now consume five percent of US electricity supply, with demand projected to nearly double by 2030. The scramble for stable and dispatchable power solutions has seen Baker Hughes provide nearly 1.2GW worth of gas turbines to data centres in the first ten months of 2025. GE Vernova has reported that orders for aeroderivative turbines – units based on or adapted from jet engines – rose by a third year-on-year. The surge is creating unexpected commercial opportunities for adjacent industries. Oilfield service companies facing declining drilling activity – US rig counts have fallen seven percent year-on-year amid lower oil prices – are pivoting toward data centre infrastructure. Even aviation start-ups are capitalising: Boom Supersonic is selling turbines originally designed for its jets to help fund its core aircraft business.

The pivot comes at a significant cost premium. One ‘behind-the-metre’ gas plant being built for Meta in Ohio has been modelled at US$175 per megawatt hour – roughly double the average industrial electricity price. On-site power generation is also markedly less efficient and produces higher emissions than grid power derived from utility-scale gas plants and renewables. Regulators are nonetheless accommodating demand. In Virginia, home to the US’ “data centre alley”, authorities are considering allowing data centres to run diesel generators more frequently, whilst the EPA has said data centres can use generators to maintain stable power.

As the data centre buildout continues, growing global pushback over rising electricity costs and water use may drive demand for cheaper alternatives and/or further unconventional solutions. Countries like Saudi Arabia, where solar generation costs just over one US cent per kilowatt hour, are positioning themselves to attract data centre investment with abundant cheap power – potentially redirecting demand away from markets reliant on expensive workarounds. Where AI infrastructure ultimately concentrates may depend less on tech ecosystems than on which regions can deliver affordable, reliable power.


Chinese open-source AI models gain ground as US firms maintain proprietary approach
Open-source strategy and aggressive pricing win over cost-conscious developers

China’s AI industry is capturing global market share through open-source (software released freely rather than kept proprietary) models that undercut US rivals on price and accessibility. Airbnb recently revealed it favours Alibaba’s Qwen over ChatGPT for building customer service applications, including chatbots, because it is faster and cheaper. Four of the top five trending models on developer forum Hugging Face, the leading repository for open-source AI models, are now produced by Chinese firms or based on Chinese models. While US companies closely guard their training data and architecture to protect future revenue streams, Chinese alternatives come with permissive licences allowing users to modify parameters and fine-tune for specific tasks – building developer loyalty in the process.

Pricing reinforces this advantage. Chinese models typically cost around one-fifth of US equivalents, with some matching benchmark performance at under ten percent of the price. This market penetration (or ‘diffusion’) has raised alarms in Washington. A recent US Commerce Department report warned that Chinese models are eroding US developers’ historic global lead. Moreover, a leaked White House memo accused Alibaba of providing the Chinese military with AI services, signalling potential export restrictions. Meanwhile, Alibaba and ByteDance are training their latest models in Southeast Asian data centres equipped with high-end Nvidia chips, sidestepping export controls after the Trump administration scrapped rules designed to close this loophole.

Security concerns still constrain the use of Chinese models in (mostly) Western government and sensitive enterprise applications, where data provenance and political considerations are front of mind. Yet industry surveys suggest over 80 percent of users are open to Chinese models if they are hosted outside China. Frontier capability still resides with US proprietary systems. But if ‘good enough’ at a fraction of the cost proves sufficient for most commercial applications, US companies’ technical edge may be insufficient to prevent open-source models from steadily accumulating developer share.


Mexico erects tariff wall on Chinese exports

Mexico aligns trade policy with nearshoring ambitions and US pressure

Tariffs of up to 50 percent imposed by Mexico on more than 1,400 goods have now gone into effect. The tariffs, which cover goods including electronics, auto parts, textiles, plastics, and steel, were passed with support from both chambers of Congress which framed the tariffs as necessary for protecting domestic industry and boosting fiscal revenue. China is undoubtedly the target, though the tariffs notionally target countries with which Mexico lacks a bilateral trade agreement. In 2024, Mexico posted a record bilateral trade deficit with China, with imports reaching about US$129.5 billion and exports remaining below US$10 billion. The tariffs underscore Mexico’s willingness to subordinate its relationship with Beijing in responding to US concerns that China is using Mexico as a “backdoor” to bypass US tariffs. The tariffs also sit alongside President Claudia Sheinbaum’s industrial policy ambitions. Released at the start of last year, Plan México aims to increase local industries’ participation in global value chains by pairing trade measures with tax deductions and incentives for firms that expand local production.

China will be most concerned about auto exports, with Mexico having fast become one of its most important overseas markets. In the first half of 2025, Mexico was the world’s largest buyer of Chinese-made cars, and Chinese brands accounted for around 20 percent of light vehicle sales in 2024. De facto tariffs on Chinese vehicles imposed by Russia – previously the largest market for Chinese auto exports – have led to a noticeable contraction in sales. While Chinese authorities have formally protested and hinted at possible retaliation, Chinese companies with established North American operations are likely to adapt to tariffs by increasing local production. Mexico is keenly aware that the United States-Mexico-Canada Agreement (USMCA) will be reviewed later this year. Sheinbaum’s government will be hoping that tariffs on Chinese goods, combined with cooperation on border security and counter-narcotics, will be enough to placate a Trump Administration with renewed ambitions in the Western hemisphere.


EU to launch compact EV class to help stave off Chinese competition

Brussels targets more permissive regulation to bolster domestic electric carmakers

The EU is moving to support its domestic auto industry by reshaping regulation rather than relying solely on tariffs to counter the rise of Chinese EVs. Policymakers are developing plans for a new compact EV category that would relax technical requirements for smaller cars and lower production costs for European manufacturers. More broadly, the initiative reflects Brussels’ assessment that defending Europe’s auto sector will hinge less on tariffs and more on redesigning the single market’s rules to favour domestic scale and cost reductions.

Under the proposal, vehicles no longer than around 4.2 metres would qualify for a new “M1E” small EV class and benefit from regulatory incentives, most notably “super credits” that allow each qualifying EU-built vehicle to count as 1.3 units toward manufacturers’ 2035 CO₂ compliance targets. The framework is also expected to open the door to targeted subsidies, tax breaks, and preferential access to parking and charging infrastructure.

The initiative forms part of a broader EU strategy to protect market share as tariffs have proved ineffective in blunting Chinese market presence. China’s share of the EU’s EV market has now recovered to levels exceeding the ten percent figure seen in late 2024 when tariffs were first imposed. By lowering regulatory thresholds for compact vehicles, Brussels estimates production costs and final prices could fall by around 10 to 20 percent, bringing entry-level EVs into the €15,000 to €20,000 range. This would make European-made vehicles much more competitive with Chinese imports, especially when combined with tariffs of up to 45 percent. Smaller cars are also an area where European manufacturers retain a relative advantage against Chinese peers, particularly in urban markets, giving them a potential head start as the segment expands. Sustained consumer demand and scaled production will ultimately be necessary to ensure the success of Brussels’ efforts to prevent the electrification of European transport from being “made in China”.