What China's R&D Lead Means for Where Your Tech Comes From

Read the headline with the PPP caveat attached

The figures come from the OECD's 31 March edition of its Main Science and Technology Indicators. Science reported the totals as $1.03 trillion for China and $1.01 trillion for the United States in 2024, adjusted for purchasing power. China also grew faster: its R&D spending rose 9.7% that year, nearly triple the US increase of 3.4%. The OECD's own release expresses China's figure as USD 0.86 trillion in 2020 PPP dollars, which in 2024 prices means both countries crossed the USD 1 trillion mark.

The caveat is not a footnote. As Research Professional News points out, China comes out ahead only when purchasing power is factored in. At market exchange rates, China's official figure of more than 3.6 trillion yuan works out to about US$506 billion, roughly half the PPP number. PPP tries to measure what money actually buys at home: researcher salaries, lab space, pilot lines. For judging the depth of an innovation ecosystem, that is the more useful lens. For judging how much imported equipment a country can afford, it flatters.

Method also changes the totals. Evertiq cites another PPP estimate that puts China at about USD 785.9 billion and the US at about USD 781.8 billion. The absolute numbers differ, the conclusion does not: the two are roughly level and China is growing faster. The trend is what companies should plan around, not the exact crossing year. A year earlier the OECD found that China's 2023 spending had reached 96% of the US level at PPP, up from 72% ten years before.

Chinese R&D is mostly corporate and mostly applied

Where the money is spent tells you what kind of output to expect. According to a republished translation of the National Bureau of Statistics' explanation, enterprises performed 2,821.16 billion yuan of R&D in 2024, research institutions 423.16 billion and higher education 306.55 billion. The enterprise share has stayed above 75% for many years. This is a business-driven system, not a primarily state-lab one.

Basic research is a small slice. Preliminary bureau data show it rose 10.5% to 249.7 billion yuan, or 6.91% of the total, with progress cited in quantum technology, life science, materials and space science. My reading: the bulk of the money goes into engineering, scaling and product iteration. That is exactly the kind of R&D that buyers feel quickly, as products that improve every year and get cheaper while doing it.

The sector comparison with the US is revealing. In 2023, the OECD found China at 95% of the US level in business R&D but only 74% in higher education, while its government-sector R&D was 1.6 times that of the US. Relative to its economy, China's R&D intensity of 2.69% of GDP sits close to the OECD average of 2.70% and above the EU's 2.13%. In other words, China spends like a typical rich economy as a share of output; its sheer size and an average annual growth of 10.5% from 2021 to 2024 do the rest.

Open-weight AI is where buyers feel it first

The clearest commercial signal is in AI models. A Stanford HAI analysis reported by The Decoder found that Alibaba's Qwen family replaced Meta's Llama as the most downloaded language model family on Hugging Face in September 2025. Between August 2024 and August 2025, Chinese developers accounted for 17.1% of downloads against 15.8% for US developers.

Downloads are not usage, but usage tells the same story. A study by OpenRouter and Andreessen Horowitz covering 100 trillion tokens found that Chinese open-source models grew from 1.2% of global usage in late 2024 to nearly 30% within a few months in 2025, led by Qwen, DeepSeek V3 and Moonshot's Kimi K2. That happened despite US limits on access to advanced GPUs, which suggests the constraint pushed labs toward efficiency rather than stopping them.

MIT Technology Review notes that DeepSeek released R1 under a permissive MIT license and that Chinese firms have repeatedly matched leading Western models at a fraction of the cost. It also observes that businesses are moving from pilots to deployment, a phase where cheaper and more customizable tools tend to win. That matches what practitioners see: once a use case is proven, the conversation shifts from "which model is smartest" to "which model can we afford to run at volume and tune to our data."

Open weights change the supplier question itself. Running a Qwen model on your own servers is not the same as buying a service from a Chinese vendor; you are adopting a file, not a relationship. The work does not disappear, though. You still need to check the license of the specific release, test behavior on your own tasks, and decide who in the company owns evaluation. Firms without the capacity to host and test models themselves get much less of this benefit.

Patents and batteries show spending turning into output

Spending is an input; patents are one of the few comparable outputs. According to WIPO data summarized by the National Law Review, China was the top source of international PCT applications in 2024 with 70,160 filings, ahead of the US with 54,087, which marked the third straight US decline. Huawei was the top corporate filer and battery maker CATL ranked fifth with 1,993 applications.

The broader picture from WIPO's own figures is starker. China-based applicants filed about 1.8 million patent applications worldwide in 2024, compared with 503,271 from the US. In 2025 China's PCT filings rose 5.3% to 73,718 while US filings fell 3.0%, and CATL appears among the leading filers with 2,203 applications.

Patent counts deserve some skepticism. Domestic filings can be inflated by incentives and say little about quality. International PCT filings are a better signal because they cost money and indicate an intent to protect and sell in foreign markets. CATL's position matters for anyone buying batteries, energy storage or electric vehicle components: the IP behind that hardware increasingly sits with Chinese firms, and licensing or partnership talks in those sectors will involve them more often.

How supplier and partnership strategy should change

European and Turkish companies now face a two-bloc market. The best price-performance in some categories, open models and batteries among them, may come from China, while US export controls on advanced chips, customer expectations and contracts with US-linked partners pull in the other direction. Pretending the choice does not exist is the worst option, because it gets made by default in procurement.

A practical approach is to separate the layers instead of treating "Chinese technology" as one thing. Open model weights you host yourself carry different risks from an API served out of a foreign cloud, and both differ from hardware with closed firmware. Sending personal data to a service hosted abroad raises data transfer questions under European and Turkish data protection rules; self-hosting open weights avoids the transfer but not the testing and documentation work.

Be honest about who this does not apply to. Defense suppliers, firms with US government customers and some regulated sectors will often be told which suppliers they can use, and their strategy is compliance, not optimization. For everyone else, a short checklist helps.

• Map each technology dependency by layer: model weights, hosted service, hardware, firmware, support.
• For each layer, note where data goes and which jurisdiction the provider answers to.
• Keep at least one alternative from a different bloc for critical components, and test it, not just list it.
• Check export control exposure in both directions, including whether your own customers restrict certain origins.
• Review licenses release by release for open models; terms are not uniform even within one model family.
• Revisit the map yearly, since price-performance in AI and batteries is moving fast.

Türkiye between two innovation blocs

Türkiye starts from a modest base. Its R&D spending rose from 198.7 billion lira in 2022 to 377.5 billion lira in 2023, and R&D intensity went from 1.32% to 1.42% of GDP. Because the lira figures are nominal and inflation was high, intensity is the more honest measure of progress. Companies performed 65.1% of spending, universities 30% and government 4.9%.

Set against China's 2.69%, the EU's 2.13% and the OECD average of 2.70%, Türkiye spends about half the OECD level relative to its economy. More recent data exist: TÜBİTAK notes that TurkStat published 2024 R&D spending and staffing figures on 20 October 2025, with series by performing sector and source of funds. Anyone building a business case should pull those directly rather than rely on older summaries, and check current incentive schemes with TÜBİTAK and their advisors.

The useful lesson from China is not scale, which a mid-sized economy cannot copy. It is that R&D is led by companies working close to products, and Türkiye's business share is already a majority. A realistic position is to be very good at adaptation, integration and domain-specific applications, and to be a credible manufacturing and engineering partner for both blocs. Open-weight models lower the cost of building on top of frontier work, which helps here.

The risk is becoming a pure integrator with no IP of its own, dependent on whichever bloc sets the terms next year. Integration work only builds lasting value if companies capture something from it: proprietary datasets, fine-tuned models for local languages and industries, or patents on the applied layer. That requires the R&D intensity to keep climbing, not just the lira totals.

Indicators worth tracking from here

A single crossing point says little; the next few years of data will show whether China's lead holds and deepens. These are the signals I would watch, because each connects spending to something a company actually experiences.

One more trend is worth noting. The OECD's latest release reports that government R&D budgets are declining and reorienting toward defence. If that continues in OECD countries while Chinese business R&D keeps growing, the civilian technology gap could widen faster than the headline totals suggest. Spending is an input. What matters for buyers is the price, performance and availability of what that spending produces.

• International PCT filings by country and by company, especially in batteries, telecoms and semiconductors.
• China's basic research share, which signals whether it moves from fast follower to frontier originator.
• The higher-education R&D gap, still the weakest part of China's system relative to the US.
• Open-model downloads and real usage shares, not just benchmark scores.
• Price-performance of the components and models you actually buy, tracked year over year.
• Türkiye's R&D intensity in the latest TurkStat series, and the business share within it.