By Peter Clarke
What’s at stake: The global chip market – and particularly players like Nvidia, AMD and memory makers Samsung, SK Hynix and Micron – will enjoy a boost that is worth tens of billions of dollars in 2025 thanks to the relaxation of export restrictions on a few AI chips. There also seems to be a concerted effort in tension-reduction between the US and China that, if sustained, could benefit the global economy. But stability is vital.
The Trump administration has reversed a decision on advanced AI chip export controls to China, a move that is set to stimulate the global chip market for the rest of 2025 and hopefully into 2026.
The administration has allowed the resumption of exports to China of certain AI chips, such as the H20 from Nvidia and the MI308X from AMD. However, higher performance AI chips and subsystems from the US, such as the Nvidia Blackwell Ultra GB300, the H100 and AMD’s Instinct MI355X remain on the exclusion list.
The reasons for the change appear to be a mix of the transactional nature of the Trump administration and successful lobbying by Jensen Huang, CEO of Nvidia, backed up by many of the great and the good at the top of the chip sector.
The decision has also been linked to recent US-China trade talks and a de-escalation of the situation between the two superpowers. In recent weeks China has relaxed some restrictions on rare earth exports, and the US has done the same for EDA software exports to China.
Synopsys-Ansys
It is also notable that China’s State Administration for Market Regulation (SAMR) approved Synopsys’ $35 billion acquisition of Ansys on July 15. Synopsys announced the deal had completed on July 17, 18-months after the deal was first announced.
The US Department of Commerce under President Trump has taken a more liberal approach to the proliferation of AI than under President Biden, including the withdrawal of the ‘AI Diffusion Rule’ in May. Readers may remember that the AI Diffusion Rule was one of the last acts of the Biden Administration published on January 13, just days before the inauguration of President Trump. This author spoke against the move (see Has US Export Control Lurched Towards AI Tyranny?).
Others have raised flags about China’s past behavior with regard to acquiring technology and how Nvidia in its eagerness to sell to China is potentially setting itself to be the next victim (see Nvidia Set to Step Into China’s Tech-Transfer Trap). And that is a potential issue in the medium term with regard to semiconductor manufacturing. However, regarding AI it may be moot given the pace of development and China’s success in the softer side of the technology, such as DeepSeek large language models (LLMs).
And it could be that the current President is more concerned about the short term.
More recently the softening of the US stance about China may have been triggered by troubling economic forecasts for the US and decreasing product availability driving prices up in bellwether US stores such as Walmart.
Economists had been predicting that US tariffs will result in the domestic market seeing lower GDP growth, higher inflation, weaker manufacturing and agriculture sectors, and persistent negative effects on employment and investment.
The President seems to have little regard for experts – although his team may be savvier – but he does appear to have a soft spot for business executives and entrepreneurs.
Leading US AI chip companies had been losing swathes of sales under the ramped export controls. So, a reduction of the export embargo was bound to appeal. In addition to benefiting the likes of AMD and Nvidia the change will also boost sales of high bandwidth memory (HBM) chips.
In July Nvidia CEO Jensen Huang met with President Trump and US policymakers. He pledged support for the administration’s goals of onshoring of manufacturing and ensuring US supremacy in AI before heading off to Beijing where he met with Wang Wentao, China’s Commerce Minister. During this trip – his third to China in 2025 – Huang also praised Chinese AI companies such as Deepseek, Alibaba, and Tencent.
What is Nvidia H20?
The H20 is a hobbled version of Nvidia’s H100 series designed to meet US export restrictions. Originally scheduled to launch in November 2023, the H20 eventually saw mass-production and market release in October 2024, specifically targeting the Chinese market. It focused on increased power efficiency and cost-effectiveness, targeting enterprise AI deployments, especially for large language model (LLM) inference and cloud-based AI workloads.
Although the H20 sits below the H100/H200 in terms of raw compute power it reportedly outperforms H100 for many real-world inference operations for some medium-sized LLMs owing to memory and efficiency optimizations. Subsequently, in April 2025, US government restrictions were increased effectively banning the sale of the H20 in China.
The export ban had a sharp impact on Nvidia’s finances. The company recorded a $4.5 billion inventory write-down in the first quarter and projected another $8 billion loss in the second quarter. That meant setting aside nearly $15 billion in lost sales from key Chinese customers. AMD was also affected. China represented 25 percent of AMD’s sales, and the elevated restrictions were expected to cut $1.5 billion from AMD’s annual sales in 2025.
Meanwhile, Chinese chip companies had launched alternative products. These included the Kunlunxin chip P800 from Baidu, Moore Threads’ MTT S80, Huawei’s Ascend 910C and Cambricon’s Synaptics 590.
A policy change?
The administration of President Biden ramped restrictions because it held the view that providing China with access to anything close to leading-edge AI capability would stimulate progress in AI within China and increase the risks to US national security due to use of the technology by the People’s Liberation Army.
The Trump administration has raised and then lowered the bar. Having seen continued AI progress reported in China, President Trump now appears to have accepted that it is better to have China dependent on US-designed silicon.
Indeed, it could also soon be US-manufactured silicon headed towards China.
The H100 was widely reported to be made using TSMC’s 4N manufacturing process node which is somewhere around 5nm or 4nm. As the H20 is a redesign of the H100 it is likely that H20 is also on 4N. And TSMC said that its Fab 21 near Phoenix, Arizona, began high-volume production of 4nm chips at the end of 2024. Meanwhile AMD’s MI300 series chips are reported to be manufactured using TSMC’s 5nm technology.
To have TSMC in Arizona shipping to China would be a reversal of the Made in China paradigm and provide President Trump with an opportunity to boast that his rapidly changing trade tariffs and disruptive policies are working. It can also be argued that with the elapse of time H20 and MI308X and equivalents on 5nn or 4nm manufacturing processes look less like leading-edge silicon and AI.
There is also an argument that having TSMC manufacturing chips for China – whether in the US or Taiwan – provides a degree of protection for the island that China considers a rogue state. China has repeatedly said it will take control of Taiwan at a time of its choosing. Meanwhile US security guarantees for Taiwan appear less strong under the somewhat isolationist President Trump.
However, I think the biggest inhibition on an invasion of Taiwan is the global economic “winter” that would engulf the planet consequently. This would leave China and most other developed countries considerably worse off for a long time.
For now, finding an acceptable level of market intervention – tariffs and controls – particularly for key technologies such as AI and semiconductors and, then applying them steadily and consistently to produce a stable environment, is the best way forward.
Bottom line: The latest modification to the export controls is raising optimism around the chip market and is to be welcomed. Unfortunately, if rules and regulations are based on expediency rather than principles there is little to stop things changing again. Meanwhile, Nvidia CEO Jensen Huang appears to be relishing his role as a de facto US trade ambassador to China. He sits atop the most significant semiconductor and AI company in the world but should beware of stepping into shoes so recently worn by Elon Musk.
Related articles:
Has US Export Control Lurched Towards AI Tyranny?
Nvidia Set to Step Into China’s Tech-Transfer Trap
Peter Clarke is a veteran reporter and analyst covering the global electronics industry. He is a regular contributor to TechSplicit from his base in the United Kingdom.
Copyright permission/reprint service of a full TechSplicit story is available for promotional use on your website, marketing materials and social media promotions. Please send us an email at talktous@techsplicit.com for details.