Permission Has a Price

The Force

On 16 October the Ministry of Commerce and the Ministry of Science and Technology published Announcement No. 48 of 2026, a draft revision of China's Catalogue of Technologies Prohibited or Restricted from Export. It adds forty technologies in three fields: twenty-three in artificial intelligence, covering large models, agents, computing clusters and information processing; five in biomedicine, covering cell and gene therapy, small-nucleic-acid drugs and pathogenic micro-organisms; and twelve in photovoltaics, covering polysilicon, wafer and cell manufacturing. Fourteen would be prohibited outright and twenty-six would need a licence. Comments are open until 15 November, but a registration requirement for covered exports took effect the same day and applies to transfers already under way.

The design is what matters. Sinocism's Bill Bishop, reading the drafts and the ministry's commentary, notes that technology already in the public domain is excluded, that open releases of restricted AI technology could qualify for automatic licensing, and that API and software-as-a-service delivery generally fall outside large-model export management. The controls bite on transfers of the underlying capability: a production line, a model's weights and training process, the engineers who commission a factory. A scientist at the Chinese Academy of Sciences told The Paper that the aim is to stop "root-pulling" transfers that take the technology itself out of China.

Solar shows how far the definition reaches. Specified thin-wafer, fluidised-bed polysilicon and perovskite and tandem production technologies would be banned; qualifying TOPCon, heterojunction and back-contact processes would need a licence; and the equipment that embodies them is covered too. The ministry's own explanation says, in translation, that exports of key photovoltaic technology "often occur alongside the export, installation, commissioning, improvement and upgrading of production lines", so a foreign factory built with Chinese machines stays inside the controls for as long as it needs Chinese engineers to keep it running. Finished solar modules, Bishop notes, are generally not covered.

The Consequence

For twenty years the question about Chinese technology was whether other countries would let the products in. The draft reverses the direction: the question becomes whether China will let the know-how out, and on what terms. A data-centre developer buying Chinese solar for its campus is unaffected; a government trying to build its own module industry with Chinese equipment, or a company planning to deploy a Chinese model on its own servers abroad, now needs Beijing's permission and may be asked for something in return. The ministry has already said that members of the World Artificial Intelligence Cooperation Organization will get "additional facilitation".

The Claim

The AI build-out is good for the electricity system: data-centre demand brings the capital that the grid needs, hyperscalers co-invest in generation and wires, and customers benefit from a larger, cleaner, better-funded network.

The Evidence

The claim comes from the top of the industry. "I think AI is the best thing that could've ever happened to the grid," Patti Poppe, chief executive of PG&E, told Fortune's Diane Brady in an interview published on 23 October. She welcomed technology companies such as Google co-investing in energy infrastructure, noted that California has added 33 gigawatts of renewables and storage in five years with 22 more due by 2030, and summarised: "The grid needs investment; AI and that whole revenue stream brings investment to the grid."

The same interview records the other side of the ledger. PG&E's shares, at $12.98, are down a third from their 52-week high after state lawmakers rejected wildfire legislation that would have stopped insurers suing utilities over wildfire claims in order to keep rates in check; San Francisco is trying to take over the company's local distribution grid to lower rates, at a cost the city puts at $3.4 billion and PG&E at $22.5 billion; and third-quarter earnings of $776 million on $6.18 billion of revenue missed forecasts. Poppe said she was open to breaking the company apart to isolate wildfire exposure, because "the status quo is not acceptable".

Nationally, the bill is arriving before the benefit. Fortune reported on 18 October that US electricity prices are up roughly 40% on average over five years and that utilities requested a further $23.1 billion of rate increases in the first three quarters of 2026, on course to match the record of nearly $31 billion set in 2025; much of the future spending is for new generation to serve data centres. Fortune's 22 October CEO Daily adds that most Americans do not want more data centres, and on 21 October Amazon abandoned a plan to turn the $427 million George Washington University campus in Loudoun County, Virginia, into data centres, pledging $65 million of land donations instead. "We just heard from the community," Joshua Weissman, Amazon Web Services' vice-president for data-centre delivery, told the magazine.

The Connection

On 18 October Anthropic said that, subject to stockholder approval, 5% of its shares would be committed to a new Anthropic Public Benefit Fund, with a further 2% added at the board's discretion for every 50% increase in the company's value above a set baseline. The fund could make large donations or direct distributions to specific populations, and the company named labour-market impacts as one example: "AI may generate enormous economic value, but it could also cause instability for many workers." The announcement came as Anthropic prepares an initial public offering that Bloomberg expects to be one of the largest ever.

Hours later, by Bloomberg's account, Musk wrote on X that "next year, I will give one of my SpaceX shares to every kid in America". Bloomberg put the pledge at about $10.5 billion for roughly 72 million children at the $146 share price; against about 13 billion shares outstanding, Matt Levine notes in Money Stuff, that is on the order of 0.5%. Levine also recalls OpenAI's July suggestion that it give 5% of its stock to the US government. Three labs, three mechanisms: a stake for the state, a share for every child, or a foundation that decides for itself what society needs.

Levine offers three readings of why a company would do this: altruism, marketing ("we look confident by giving some value away") and negotiation. The third is the one worth keeping: "Let's give society a share of our success, so that society will let us have that success." The sliding scale in Anthropic's design, which raises society's stake only as the company's value rises, reads most naturally as that bargain written into the capitalisation table.

What Transfers

The useful idea is to read these pledges as a line in the cost of capital rather than in the philanthropy budget. A company that expects to need public permission for something very large, whether a listing, a merger or a licence to displace labour, can now pre-pay for it in equity, and the going rate has been set in public at between half a percent and five. The same logic will be available to anyone else whose expansion depends on consent: data-centre developers facing county boards, utilities facing rate cases, and, after this week, technology exporters facing a ministry.