China's battery plan and DeepSeek's Huawei tools show a policy turning inward

On 28 September, China's government publicly released a five-year plan for the new-type battery industry. The document is dated 14 September and was issued jointly by seven bodies, including the Ministry of Industry and Information Technology, the National Development and Reform Commission and the National Energy Administration, according to pv magazine, which reported the release on 1 October.

It is not a capacity target. The plan avoids a specific output figure, which is unusual for Chinese industrial policy, and instead sets technical goals for 2030: preliminary large-scale application of all-solid-state batteries, a cycle life of 15,000 cycles for long-life lithium batteries, and defect rates at leading manufacturers measured in parts per billion. It also calls for work on electrode materials, new electrolytes and high-end auxiliary materials.

The chemistry section is where the policy turns interesting. Lithium-ion stays the core technology, but the plan explicitly calls for a supply system in which sodium-ion and flow batteries develop alongside it. Research lines include new alkali-metal-ion batteries, multivalent-ion systems, metal-air and metal-sulfur batteries, and nuclear batteries. Standards work for sodium-ion is listed next to solid-state and other emerging technologies.

For stationary storage, that matters. Sodium-ion competes on cost, cycle life, safety and supply-chain diversity rather than energy density, and the plan puts it inside a multi-chemistry portfolio rather than as a replacement for lithium iron phosphate. pv magazine notes the document sets no sodium-ion deployment target and does not position it as a substitute for LFP.

MIIT said authorities will strengthen capacity early-warning mechanisms, product-quality supervision and regulation of price competition to avoid sharp swings in supply and demand.

That governance language is the other shift. The plan supports mergers and consolidation, recycling, safety management and more orderly overseas investment, and it wants deployment expanded from generation and grid storage into industrial parks and data centers. The Ministry's accompanying interpretation calls for new storage scenarios and tighter coordination between supply expansion and actual demand.

The battery document landed two days after a very different kind of announcement, and the two together say more about where Chinese technology policy is heading than either does alone. On 30 September, DeepSeek said it was releasing open-source programming tools for Huawei's Ascend chips, according to a post on its official WeChat channel. The centerpiece is TileLang, a programming language developed by researchers at Peking University that DeepSeek has used for about a year and now treats as its main tool for artificial general intelligence work, The New York Times reported, as relayed by The Decoder.

The release includes libraries for computation and for moving data between chips, and DeepSeek says Huawei fully supported the work. The two companies also optimized a supernode, a cluster of 128 Ascend 950 chips.

Why does a programming language matter more than a chip? Because Nvidia's dominance does not rest on silicon alone. It rests on an estimated four million developers worldwide who build with CUDA, an ecosystem rivals like AMD have not crossed even when their hardware looked competitive on paper. DeepSeek's argument is that an independent software ecosystem needs a universal language that is easy to program but still extracts full performance. In its view, TileLang offers a simpler model than CUDA.

The timing is not accidental. Two weeks before DeepSeek's announcement, Huawei unveiled new AI processors and supernode systems and said they would be widely used for model training next year. Huawei also admits it cannot keep up with demand at home and plans to sell fewer chips abroad. Referring to US export controls, rotating chairman Eric Xu said the company cannot accept a future that hinges on whether others are willing to sell chips to China.

Independent analysts are more cautious about how much of Nvidia's moat has actually eroded. SemiAnalysis tested Jalapeño, OpenAI's inference chip, and called the CUDA moat "potentially dead" because OpenAI gets new models running on its own hardware quickly, with Jalapeño beating Nvidia's Blackwell on performance per watt in most scenarios tested. The same analysts added their own caveats: they only tested relatively easy-to-optimize workloads of about 8,000 input tokens and 1,000 output tokens, and had not yet run AgentX, a benchmark for multistep agent tasks.

That gap is exactly where SemiAnalysis found Nvidia well ahead in August. With AMD's current software stack, Nvidia would still come out cheaper per token even if AMD gave its hardware away, and the authors locate Nvidia's lasting advantage not in silicon but in the software that links many chips into one system. Huawei's chips were not part of the AgentX comparison, though an earlier SemiAnalysis analysis of DeepSeek V4 noted that Huawei's CANN stack was the only one besides CUDA to support the model on day one.

Both stories are about the same thing: building domestic stacks for hardware, software and standards that do not depend on foreign suppliers. The same instinct is visible in US-facing electric-vehicle competition, where Tesla treats China as the market that gets features first. On 1 October, Tesla China announced that every Model 3 it sells there gets a 16-inch center touchscreen, up from 15.4 inches, and that every Model 3 and Model Y on sale in the country can export up to 2,200 W of AC power, according to Electrek.

The US Model 3 gets neither. It still has a 15.4-inch display, and when Tesla brought vehicle-to-load to the US in August it limited the feature to the Model Y Premium, Model Y Performance and Cybertruck, at 2.4 kW through an $80 adapter, leaving the entire US Model 3 lineup out. Chinese buyers also get a light gray premium interior and a new round of incentives: Model 3 orders placed by 31 October get 5,000 yuan off the final payment, stacked with five-year 0% financing, an 8,000 yuan insurance subsidy and other credits. Tesla advertises the Model 3 from 222,500 yuan, though the base RWD list price remains 235,500 yuan, so the headline figure only works once discounts and subsidies are counted. BYD's new Formula S starts at 189,900 yuan for the RWD sedan, according to Electrek.

There is a harder edge to the inward turn. On 24 September, DW reported that new entry-exit rules took effect giving Chinese authorities the power to stop engineers, founders and other specialists from leaving the country if their expertise in batteries, rare earths or artificial intelligence is judged a threat to industrial and technological security. The same report notes tightened outbound-investment rules and restrictions on posting technical staff overseas.

Henry Gao, a law professor at Singapore Management University, told DW the measures "offer a rare glimpse into the true state of China's economy" and suggest Beijing is deeply concerned about economic weakness and substantial capital outflows. Alicia Garcia-Herrero, chief economist for Asia-Pacific at Natixis, told DW she thinks the talent block will matter more than the capital squeeze: money can still move slowly with approvals, she said, but a process engineer who cannot board a plane is harder to replace. Bloomberg Intelligence estimated about $1 trillion in Chinese wealth left the country last year, the largest hot-money outflow since records began in 2006.

The travel restrictions follow earlier reports about AI talent. In May, Bloomberg reported that Chinese authorities required top AI researchers, founders and executives at firms including Alibaba and DeepSeek to obtain approval before traveling abroad, and The Information reported last year that some DeepSeek staff were asked to hand in their passports. Beijing has neither confirmed nor denied the practice, DW noted.

What connects a battery plan, a chip programming language and exit controls is a preference for building capability behind controlled borders. The battery plan's own framing makes the shift explicit: away from capacity-led expansion and toward technology, quality and supply-demand discipline. The DeepSeek release makes the software equivalent explicit: a domestic layer that sits between Chinese models and Chinese chips. The exit rules make the third leg explicit, treating scarce expertise as a resource to be kept in place.

None of this settles whether the strategy works. SemiAnalysis's caveats about which workloads were tested are a reminder that ecosystem claims are easy to make and hard to verify. China's biotech record, described by physician and consultant Ruby Wang in Rest of World on 24 September, shows what a decade of patient capacity building can produce, with China now accounting for half of global drug licensing deals and potentially up to 60% this year, alongside acknowledged weaknesses in immunology and commercialization. The battery and chip plans are earlier in that arc.

There is also a counterargument that the controls themselves are the risk. Gao told DW the curbs may prove counterproductive, because the harder it becomes to move capital or people abroad, the greater the incentive for those with means to find other routes. "Over time, that could further erode confidence, accelerate capital flight and deepen the very economic problems the measures are intended to contain," he said.

For now, the dated record is clear enough. A plan released on 28 September puts numbers on 2030 battery ambitions. A tool released on 30 September puts open-source software under Huawei's chips. And a rule that took effect on 24 September puts a border around the people who build both.