Your confiscated power bank is a lesson in geoeconomics

China treats standards as strategy. The West treats them as a PDF to sell.

Cover illustration generated with Gemini: engineering documents flowing into a machine that produces finished devices.

TL;DR

In 1994 the Perry Memo cut roughly 45,500 US military standards to about 28,300 as a cost exercise. It looked backward and assumed the private sector would write what came next. What arrived was paywalled documents and a large audit industry.
China went the other way: free, descriptive, fast-moving standards tied to its factories, with 4,929 issued in 2025 alone. It is the same bet open source won against proprietary software.
Read STEM as a value chain, Science, Technology, Engineering, Manufacturing, rather than an education silo. China, Japan and South Korea keep that chain connected. The US, Europe and Australia lose the E layer.
Australia ships the rocks and rents back the physics. New minerals processing is a step in the right direction, but at the scale of tens of billions in forecast iron ore and coal declines, it is a ribbon-cutting, not a decades-long strategy.
The metric nobody publishes is time to first export certification. On the bridge to export, Australia ranks last of seven.

Disclaimer. This article is the author’s personal analysis and opinion, published for general information and discussion. It is not professional, legal, engineering, regulatory, investment or certification advice, and should not be relied on to certify a product or to make a business or policy decision. Figures come from public sources and consultancy estimates. Several are reconstructed or are rough orders of magnitude, and the STEM-chain scores and the certification comparison are the author’s qualitative reading, not an index. See the appendix. Company, standard and product names are used for identification only.

The CCC mark on a power bank, the free download of a Chinese national standard, and Qwen overtaking Llama on Hugging Face are the same story told three times. China runs standards as infrastructure. The West, which invented that model with MIL-STDs and RFCs, retired it in 1994 and now sells the rules of its own economy at a few hundred dollars a copy.

This piece moves through three time horizons. Then, 1994: how the Perry Memo got the West here, a cost-cutting exercise that looked backward and assumed someone else would write the next generation of standards. Now, to 2035: China’s long-running, forward-looking decision to own the standards layer. Next, to 2050: STEM as a value chain rather than an education silo, and why the nations with an engineering industrial strategy sit where they do. Australia is the worked example throughout, because Australia is the clearest case of a rich country that sells the raw material and rents back the physics.

THEN, 1994: HOW THE PERRY MEMO GOT US HERE

The day the West retired its engineering commons

In June 1994 the US Secretary of Defense, William Perry, issued a memo that banned most military specifications and standards unless a waiver was granted. At the time the US held roughly 45,500 MIL-STDs and MIL-SPECs. By 2003 about 28,300 remained. Thousands were cancelled and nothing was reinstated. A partial reversal in 2005 restored no content. What followed was fragmentation, not succession.

Read the motive carefully, because it matters for everything after. The memo was a cost exercise. Maintaining a corpus of tens of thousands of documents was an expense line, and expense lines get cut. It looked backward: it counted what existed, priced the upkeep, and cancelled what looked redundant. It had no theory of what a standards system is for. And it carried one load-bearing assumption, that “performance specifications” and commercial standards, ISO 9000 above all, would fill the gap and the private sector would supply whatever engineering knowledge the military no longer wrote down.

That assumption was never tested before it was acted on. The National Academies noted the counterfactual had never been measured, and it still hasn’t been, twenty-five years on.

Graphic 1. National-scope standards in force, thousands. Filled markers are verified anchors (MIL-STD about 45,500 pre-Perry and about 28,300 by 2003; GB 47,378 in April 2025). Every other point is a reconstruction from those anchors, plus or minus 20 to 30 per cent. Sources: Standards Portal (SAC database, April 2025); Wikipedia and UW-Madison ASSIST notes for MIL-STD counts; ISO, IEC, BSI, Standards Australia catalogues.

Two kinds of knowledge, and only one of them works

The distinction the memo missed is between descriptive and prescriptive standards.

A descriptive standard records what physics and failure have taught. MIL-STD-810 on environmental testing, MIL-STD-461 on electromagnetic interference, MIL-STD-704 on aircraft power, and the IETF’s RFCs are distilled failure analysis. They say what happens to a device in vibration, heat and noise, and how to pass. You read one, you build to it, and you find out quickly whether you were right.

A prescriptive management-system standard is a different animal. ISO 9001 and ISO 27001 are deliberately content-free about the product. They specify a process: document what you do, do what you documented, audit the gap. They are written to apply to any organisation in any industry, and that universality is the design flaw, because a standard that applies to everything can say nothing specific about anything. A standard whose compliance is measured by documentation can only produce documentation.

The failures are on the record. The 737 MAX was certified in a world of AS9100. Dieselgate happened inside companies certified to ISO 14001. The paperwork was in order.

What the private sector actually supplied

So what filled the vacuum? Documents for sale, and an audit industry to go with them.

ISO earned about CHF 19.5 million (roughly US$22 million) from document sales and royalties in 2019. That is pocket change, and ISO still litigates to protect it. The money is downstream. BSI’s assurance division, which does certification, training and audits, earned about £333 million (roughly US$424 million) in 2023. The global testing, inspection and certification industry is about US$256 billion in 2026: SGS, TÜV, Intertek, UL, DEKRA, BSI. The PDF is a loss-leader for the audit relationship.

Set that against what failure still costs. Medical device recalls run US$7 to 8.5 billion a year, about 2 per cent of sector sales, inside firms certified to ISO 13485. The compliance machine is large, growing, and staffed by capable people. It is not obviously reducing the failures it exists to prevent.

Graphic 2. Approximate US dollars on a log axis, so each step right is ten times the money. ISO figure from The Register (31 July 2021); BSI from its 2023 annual report via IIOA; TIC market from Coherent Market Insights and Aventis Advisors; recall costs from Lumafield citing McKinsey. Converted at CHF 1.13 and GBP 1.27 to US$.

A word on the other side of this ledger. Summed nation by nation, the lost earnings from a weak standards system are an informed estimate, not a measured series. My reading is that China is strongly positive, the US negative and getting worse, and Australia worst per capita. Treat that as a hypothesis the data would need to test, and note that the test has not been run.

The paywall is an SME-exclusion device

The cost lands unevenly. Participation costs money, access costs money, and fixed certification overhead is regressive by firm size. A two-person firm in Vietnam can build to China’s GB EV-charging specifications for free. A US start-up has to buy the standards behind UL 2056, IEC 61851, ISO 15118, SAE J1772 and FCC Part 15 before it has revenue.

That is Schumpeterian entry throttled at the regulatory layer. The firms most likely to produce a radical alternative are the ones least able to afford the entry ticket, so the system selects for incrementalism. This is not a conspiracy. It is what a revenue model does when it is applied to infrastructure.

NOW, TO 2035: CHINA OWNS THE STANDARDS LAYER ON PURPOSE

The vacuum got filled deliberately

China’s national standards come in three tiers: GB (mandatory), GB/T (recommended) and GB/Z (guidance). In April 2025 there were 47,378 in force. In 2025 alone China issued 4,929, up 56 per cent on the year. Under the China Standards 2035 programme this is not drift. A group-standard layer, the T/ standards written by industry bodies, works as an RFC-to-standard pipeline: a consortium proves something works, and it graduates upward. China has signed 108 cooperation agreements with 65 partner countries, and standards travel with Belt and Road financing, bundled with the hardware.

Compare the tempo. In its golden age the US produced about 1,200 MIL-STDs a year. ISO runs at a committee tempo of 500 to 700 a year. China’s curve spiked to nearly five thousand in 2025, and it has not inflected yet.

Graphic 3. New standards issued per year, five-year period averages. Only the 2025 GB point (4,929, +56% year on year) is verified; every earlier point is reconstructed. Sources: gov.cn (5 February 2026); Standards Portal.

Who charges to read the law

The second half of the contrast is access. China publishes GB full text free, in Chinese. The US government publishes its own standards free, and the ASSIST database still holds about 22,000 active MIL-STD/SPEC documents. IETF RFCs, fewer than 10,000 documents, are free, and they run more of the world economy than all of ISO combined. On the other side of the line sit the paywalled corpora: BSI with about 27,000 documents at £100 to 300 each, ISO with about 26,500 at CHF 100 to 300, ASTM with about 12,000, IEC with about 10,500, Japan’s JIS English editions at about 10,000 sold by the JSA, and Australia’s AS/NZS at about 10,000 documents at A$100 to 500 each.

Many of those paywalled standards are cited into law, so you pay to read the rules you are legally obliged to follow. Europe is only cracking after the European Court of Justice’s PublicResource ruling. The UK’s wiring standard, BS 7671, costs about £100. Australia privatised standards publishing in 2003. The toll booth sits on top of Western soft power.

Graphic 4. Documents in force by corpus, coloured by access model, with the price of reading one. GB total includes ISO/IEC adoptions; ISO total includes management-system standards. Sources: Standards Portal; UW-Madison ASSIST; ISO, IEC, ASTM, BSI, JISC and Standards Australia catalogues; The Register (31 July 2021) on ISO paywalls.

Open source already ran this experiment

None of this is new. Software ran the same experiment and recorded the result. Linux against Solaris. Apache against IIS. Postgres against Oracle. TCP/IP against OSI, the committee-designed rival that lost to the thing that was free and worked.

Three mechanisms did the work. Radical innovation happens at the edges, not in the committee. Commoditise a layer and compete above it. And forking is parallel search: when anyone can copy and modify, a thousand variants test what a single owner would never try. China’s GB corpus is Linux in 1999. Nobody building new infrastructure in 2035 will want to build it on a US$300 PDF.

Same playbook, new substrate: open weights

The most recent run of the same playbook is AI. Qwen has passed one billion downloads, including 153.6 million in February 2026 alone. Qwen derivatives are more than 40 per cent of new Hugging Face variants against Llama’s 15 per cent. Chinese models went from roughly 1 per cent to roughly 15 per cent of global usage in a year. Six of Japan’s top ten corporate models run on DeepSeek or Qwen. DeepSeek-V4 is priced at about a twelfth of GPT-5.5.

There are three real friction points: trust, chips and content alignment. A free standard does not remove them. It does explain why a market that cannot get the open version from the West will take it from somewhere else.

One power bank, three marks

Back to the power bank. On 28 June 2025 China’s civil aviation regulator, CAAC, began requiring a 3C mark on power banks carried aboard, which is why so many travellers had theirs confiscated.

Mine was one of them. My near-new power bank was confiscated at the airport in 2025. I was not surprised. I had seen it coming from a digital nomads group on Facebook, where travellers swap exactly this kind of ground truth, and I had prepared. I had also read up on CCC and understood it as a more rigorous, mandatory and physics-based regime than the marks I was used to. I knew my other devices, the headphones, portable fans, laptop and phones, would not be touched, because their batteries are built into the products and the mandate targets the loose, high-capacity lithium cell in a bag. Losing the power bank cost me one gadget. Watching a rule be understood in advance, by travellers, from a chat group, and then enforced to the letter at the gate, taught me more about how this system works than any standards document had.

Three regimes treat the same product differently. CE is a manufacturer’s self-declaration. UL is rigorous but voluntary. CCC is tested, factory-audited, QR-traceable and enforced at the gate, and in 2025 the fires traced to separator film fed straight back into the standard, a feedback loop running through physics. The honest caveat: protectionism is part of the picture, and counterfeit CCC marks exist. But the structure shows a regime that treats the standard as a living instrument tied to the factory floor rather than a certificate collected once.

NEXT, TO 2050: STEM AS A VALUE CHAIN, AND THE M IS THE POINT

Not an education silo

STEM is usually discussed as an education pipeline: Science, Technology, Engineering and Mathematics, measured in enrolments, degrees and graduates. Read it instead as a value chain, and swap one letter. Mathematics drops out as a stage and Manufacturing takes its place. Maths does not stop mattering. It is the language underneath every stage, not a stage of its own. What the chain adds is the last step, the one where value is actually captured.

In the chain, Science produces papers. Technology proves them in a lab and finds a venture. Engineering turns the prototype into something certifiable, standardised and repeatable. Manufacturing builds it, deploys it and exports it. Value flows left to right, and each handover can break.

Most countries pursue one variant or the other. They fund the education pipeline and hope the chain follows, or they chase the chain and neglect the pipeline. China’s strategy is to run both: the education pipeline feeds a chain it has deliberately connected from S to M. That, more than any single statistic, is the difference between a country that counts graduates and a country that ships product.

The E stage is the one nobody talks about. It is where standards, certification and design-for-manufacture live, and it is where the nations with an engineering industrial strategy have concentrated their effort. China, Japan and South Korea all keep the chain connected, in different ways.

China is fully coupled, deliberately: standards written for its factories, certification exported with the hardware. South Korea is ruthlessly aligned from T to M, a middle power that chose the manufacturing end of the chain. Japan is coupled but closed. Its JIS engineering is superb and locked in Japanese, exported as product and not as influence. Japan’s strengths show where a connected chain pays off: warships, submarines, industrial automation and the older mature manufacturing it dominates. Its weakness is the modern M. It lags in EVs at scale, in leading-edge chips and in critical-minerals processing, which are the three places the new industrial base is being built. A connected chain buys you what you built it to make. It does not buy you the next thing automatically.

The US is the broken case: designs in California, builds in Shenzhen, retired its own E layer in 1994. Europe is captured by rent, paying a certification complex to administer its decline. Australia is decoupled so far: it funds the S that produces papers and starves the T, E and M that produce things.

Graphic 5. Left: approximate raw-to-finished value multiples per unit of material, a research summary range and not a measured series (lithium’s 10 to 50 times is the loosest). Sources: ANZ; ASPI/IEEFA; industry.gov.au Resources and Energy Quarterly (December 2025); Epic Sourcing (2026); RFF; The Policymaker (APPI); RBA Bulletin (October 2025). Right: the author’s qualitative scoring, 1 (weak) to 5 (dominant), of each economy’s strength at each stage, not an index. The per-country verdict column has been removed from the poster version for space and is in the text.

Australia ships the rocks and rents back the physics

Iron ore, LNG and coal make up about 80 per cent of Australia’s exports to China. China supplies A$90 billion or more of manufactured goods a year in return. Commodity by commodity, the value left offshore is large.

Iron ore: about 85 per cent of exports go to China, which returns it as steel products worth roughly 3 to 6 times as much per unit of material. Copper is refined, then turned into magnet wire, motors and EV drivetrains, roughly 3 to 5 times. Coal becomes coke for steel and power for factories, roughly 2 to 4 times. Lithium is the starkest. Nearly all Australian spodumene is refined in China and returns as EVs and batteries, 10 to 50 times or more. Australia holds 0.53 per cent of lithium-battery value chain value, and over US$200 billion of processing value is realised offshore. All of these multiples are approximate ranges, not a measured series.

Read it through the chain. Australia is S plus mining, which is extraction engineering only. It skips refining, materials chemistry, cell fabrication, product manufacturing and certification. China runs T to E to M and sells the M back, with the E layer (GB 38031, GB/T 20234, GB/T 27930 for batteries and EV charging) free, fast-revised and coupled to its factories. Australia exports entropy and imports negentropy.

None of this is a failure of effort or of endowment. Australia is a reliable, rule-of-law supplier with some of the best resources and mining engineering on earth, and it has built the S and the extraction end of the chain very well. The question is what it does next.

The rent has an expiry date. Iron ore earnings are forecast to fall from about A$116 billion toward A$77 to 79 billion by 2030-31, and the RBA expects China’s demand for Australian ore to fall 80 per cent by 2050.

The right direction, at the wrong scale

Some of what is happening now points the right way, and it should be said plainly. The Critical Minerals Production Tax Incentive offsets 10 per cent of eligible processing and refining costs across 31 critical minerals from July 2027 to June 2040, valued at about A$7 billion over the decade. A Critical Minerals Strategic Reserve of A$1.2 billion has been announced, starting with antimony, gallium and rare earths. Iluka’s Eneabba rare earths refinery, backed by a A$1.65 billion federal loan, is due to start producing separated oxides, including the neodymium and praseodymium that go into EV magnets, with a take-or-pay offtake from 2028. IGO’s Kwinana plant has produced its first lithium hydroxide. This is real refining, on Australian soil.

It also arrives with warnings. Albemarle has idled both trains of its Kemerton lithium hydroxide plant, and cancelled the expansion, because it could not compete with Chinese processing on cost. The point is that processing without the surrounding system is fragile: Chinese plants sit inside a coupled chain with free, fast-revised E-layer standards, a deep supplier base and decades of accumulated process knowledge.

And the order of magnitude is unforgiving. Critical minerals earnings are growing by single-digit billions of dollars a year. Iron ore is forecast to lose tens of billions, and coal faces its own decline. Even if every announced project lands on time, processing adds a fraction of what the bulk commodities are about to take away.

My reading is that the gap is one of kind, not of effort. A plant opening is a ribbon-cutting. A GDP-moving industrial base is a decades-long strategy: free standards, test infrastructure, a supplier ecosystem, patient capital and a government that holds the course across election cycles and price cycles. Korea, Japan and China each ran that kind of strategy for thirty years or more. Australia has funded the announcements. It has not yet funded the E layer that makes the announcements compound.

The valley of death sits at E to M, and certification is the bridge

The usual story about commercialisation is a funding gap. UK Parliament, CEDA and AMGC all describe a “serious gap between the research and commercialisation worlds”. The chain locates it precisely: between the working prototype (T) and the manufacturable, certifiable, sellable product (M).

What a firm physically does in that valley is a standards-compliance process. It works out which requirements apply, designs for test, and certifies. Certification is the exit. That makes the valley a knowledge-and-access gap at least as much as a funding gap. US battery ventures with working prototypes at technology readiness levels 5 to 6 died needing US$50 to 100 million to build manufacturing and finish automotive certifications.

The same structure repeats at the border. Conformity assessment is the largest non-tariff barrier to trade. Only 7 per cent of Indian SMEs hold an international certification. Countries that support SME compliance see export growth of 8 to 10 per cent a year. Australian wine heading into Canada faced zero tariffs and still hit conformance barriers. A nation’s export capability is downstream of its standards infrastructure.

Time to first certification, and Australia’s missing bridge

No nation publishes official figures for how long it takes to certify a product, which is itself finding number one. The best available evidence is consultancy-published lead times for a mid-complexity consumer electrical product, in weeks, shown below for the fast, typical and slow case in each regime. These are commercial estimates, and the slow paths are documented worst cases, not distributions.

Graphic 6. Consultancy-published lead times in weeks for a mid-complexity consumer electrical product; no nation publishes official figures. Colour bands are the author’s: green 6 weeks or less, amber 7 to 14, red 15 or more. Rows are ordered by typical time. Sources: consultancy guides for CCC, KC, PSE, UL, CE and RCM.

Read the shape rather than the headline. Physics gates compress and process gates queue. CE self-declaration is fast because nothing is verified, and CCC stays true after issue. Korea’s two to three week IEC CB conversion path shows the West’s own tools can be fast.

Australia looks mid-table on paper, and the paper is the problem. The RCM is an import gate, not an export bridge. Its ecosystem exists to help foreign manufacturers in, and Chinese labs sell “RCM conversion from CE reports”. The standards behind it (AS/NZS 4417, 62368.1, CISPR 32) are paywalled even for domestic compliance. Fixed costs run A$5,000 to 25,000 per product and another A$15,000 to 50,000 per foreign market, set against a 26 million person home market. NATA, Australia’s accreditation body, is genuinely good: the world’s first comprehensive lab accreditation body and a founder of ILAC. But it is thin, with expedite surcharges of 30 to 50 per cent. And Australia has mutual recognition agreements with the EU, UK and EFTA, and none with China.

Judge it by the right metric, time to first export certification, and Australia ranks last of the seven for lack of a bridge, not for the speed of its gate. Every other regime at least plausibly serves its own exporters. Germany built the bridge with DIN, Fraunhofer and TÜV. Japan has JIS and AIST. Korea has KSA and KTL. China has GB, CCC, CNAS and MRA diplomacy with the developing world. Australia’s regime serves the world’s manufacturers’ imports.

Three scales, one fork

The same choice shows up at three scales. Geopolitically, strategy beats revenue, as open source beat proprietary. In the market, the paywall’s measurable output is suppressed entry and therefore incrementalism. At the civilisational scale it is rent extraction against value creation: toll roads on old infrastructure make you rich now and hollow later.

History delivered the verdict once in software. It is delivering it again in engineering.

What I would measure

If I could ask for one number, it would be time to first export certification, published by every nation, alongside whether the standards behind it are free to read. Nobody publishes it. Until someone does, the argument above rests on mechanisms and consultancy estimates, and Australia will go on counting patents.

Appendix: notes on the data

Verified anchors: GB 47,378 (April 2025) and 4,929 issued in 2025; ISO 26,531; MIL-STD about 45,500 pre-Perry and about 28,300 by 2003; ISO 2019 sales and royalties; BSI 2023 revenue; TIC market size; Hugging Face and TrendForce figures as published.
Reconstructed: every pre-2025 corpus size and every per-year issuance figure except GB 2025.
Qualitative: the STEM chain scores and the CE/UL/CCC comparison are my reading of published mechanics, not an index.
Rough orders of magnitude: the Australian export and processing figures and the per-commodity value multiples are approximate ranges chosen to show the shape of the chain, not to settle a forecast.
Consultancy estimates: all certification lead times, for a mid-complexity consumer electrical product; the per-market fixed cost of A$15,000 to 50,000 is a research-summary estimate.
Informed estimate: the lost-earnings side of the national ledger.

Appendix: references

Wikipedia (United States Military Standard, MIL-STD-498, MIL-STD-1553, National Standards of China, Standards Australia)
Perspectiva Solutions
China Briefing
Clingendael (2025)
Ifri
The Diplomat (September 2021)
Standards Portal
gov.cn (5 February 2026)
The Register (31 July 2021)
Mesopartner
EU-Japan Centre
JISC
IIOA (BSI 2023 report)
Coherent Market Insights
Aventis Advisors
Lumafield/McKinsey
WTO TBT
National Academies (NAP 4921)
Treasury (2022) productivity paper
Parliament of Australia economic dynamism inquiry
Productivity Commission 2026 interim report
ANZ (China-Australia trade)
ASPI/IEEFA (green iron)
industry.gov.au Resources and Energy Quarterly (December 2025)
Epic Sourcing (2026)
RFF
The Policymaker (APPI)
RBA Bulletin (October 2025)
industry.gov.au Resources and Energy Quarterly (critical minerals and iron ore outlooks)
Department of Industry on the Critical Minerals Production Tax Incentive and Strategic Reserve
Mining Magazine and Engineering News on Iluka’s Eneabba loan and refinery
Benchmark Minerals on Albemarle’s Kemerton idling
UK Parliament (valley of death)
CEDA
industry.gov.au (technical barriers, standards and conformance infrastructure, NATA)
consultancy lead-time guides for CCC, KC, PSE, UL, CE and RCM
Forbes (28 April 2026)
MIT Technology Review (12 February 2026)
TrendForce (26 January 2026)
UW-Madison library (ASSIST)

Also published on LinkedIn.