A framework for public funding of innovation and startups

I blogged here exploring models of innovation funding generates the greatest bang for the buck in terms of achieving the primary objective of catalysing innovation. This post will provide some analytical frameworks to formulate a policy on innovation funding. This (on the importance of portfolio management activities), this (on an industrial policy for funding startup innovation largely through grants), and this (on the success of Maharashtra’s Defence and Aerospace Fund) are other recent blogs on the topic. This post will summarise all the takeaways and outline some guidance on startup innovation funding. 

The policy objective of startup financing is fundamentally to expand the envelope of investible startups and innovations and thereby crowd-in private risk capital. What is the best approach to achieve this objective?

Answering the question requires addressing the challenge of whether public innovation funding primarily expands the investable universe (genuine additionality) or primarily subsidises returns on investments that would have happened anyway (return-amplification / crowding-in of already-attracted capital). 

In the context of infrastructure, I have written here arguing that India’s efforts to crowd-in private capital into infrastructure sectors through the likes of IIFCL and NIIF, and IDFC earlier, have struggled to deliver the additionality (in terms of derisking sectors outside the traditional strongholds of public private partnerships) as the new institutions have ended up competing with the private sector for investments. 

What does the empirical evidence on these efforts globally report? 

A useful framework for thinking about this would be to categorise funding into three buckets: seed/angel stage (one that leads to proof of concept and lab validation, TRL 2-4), technology/product development stage (includes prototypes and pilots, TRL 4-7), and commercial scaling stage (TRL 7-9). The first category is pure incubation of ideas through grants; the second is about expanding the pool of scalable innovations; and the third is about derisking and crowding in private capital to scale innovations. 

The first stage, being the riskiest, will have the greatest additionality from public funding. It is invariably grant-funded, and gets the biggest share of public funding focus across countries, also because it is essential to create the pipeline of startups that can be feedstock for VCs and other investors. It is no good to have a VC ecosystem without a strong investible startup pipeline in the prioritised technologies. 

In the second stage, being the “innovation valley of death”, grants may be the best option. While instruments such as a Simple Agreement for Future Equity (SAFE), popularised by Y Combinator, and other forms of convertible funding are commonly discussed in the context of technology/product development, the evidence from successful global cases points to grants, with at best hybrid forms like clawbacks or profit sharing. Interestingly, apart from India (BIRAC and MEITY MSH), no major country uses SAFE in public funding.

This is because while investors obviously favour equity instruments like SAFE in pure private market contexts, they create problems with the determination of future cap tables, significantly diluting entrepreneurs and diminishing their incentives at so early a stage of the startup’s journey, and also making them significantly unattractive for commercial investors (who generally prefer startups without the encumbrances from public shareholding). 

It can be observed that those with risk capital instruments tend to kick in only at the TRL 6-7 stages. 

In this context, it is worth briefly discussing the critiques of grant funding to startups. Those who critique giving grants to startups do not realise the central role of public funding in deepening the innovation ecosystem for commercial capital to then come in. In countries like India, where early-stage risk capital is tiny, public funding is critical to create a deep and broad pipeline of innovations. The concern of possible incentive distortions from giving away free money is largely minimised through milestone-based tranches or conditional grants. Critics also tend to conflate these two categories of funding with the third stage of commercial scaling capital, which we now turn to.

The dilemma between expanding the pool of capital and returns-amplification is most relevant to this stage of commercial scaling capital. The global evidence on this is mixed. The Israeli Yozma program, which deployed funds through Fund of Funds (FoFs), is thought to have catalysed the country’s vibrant VC industry. 

However, other examples point to returns-amplification. A study of IFC’s blended finance deals in the 2000-20 period finds comparable financial returns to non-blended projects but “no statistically significant excess private mobilisation beyond what IFC’s standard lending would attract.” In other words, blending did not increase the quantum of private investment — it redistributed risk between IFC and private co-investors. 

This finding is echoed in a 2023 study of SIDBI’s Fund of Funds for Startups (FFS) 1.0 by the Impact and Policy Research Institute (IMPRI), India’s Startup Engine: A Policy Review of the Fund of Funds Initiative. It finds that most FFS 1.0 capital went to established VC funds that would have raised capital independently. It also found “crowding-in effect was primarily reputation/signal, not financial additionality,” the government’s involvement functioned more as a validation of fund managers to private investors than as a necessary injection of capital. While FFS 1.0 delivered a 2x mobilisation ratio, it was mostly in already-functioning VC markets.

This brings us to the question of the most cost-effective approach to achieve the public finance objective while supporting commercial scaling. The options span the spectrum from directly investing in the startup to indirectly investing through FoFs. 

While the former allows for targeting the riskiest innovations/startups, it creates the challenge of due diligence, which can be addressed through co-investment with professional investors that piggyback on their diligence. While the latter limits the control over who/what is funded, it allows full play for professional investment practices. 

In either case, the nature of the entity that deploys the public funds is important. A fully public or majority public shareholding corporation, whether non-profit or for-profit, will struggle to deploy risk capital and will be hobbled by the restraints of the General Financial Rules (GFR) and the vigilance from oversight agencies. It is for this reason that there is no instance from India of a government-owned entity directly making risk capital investments (apart from the Maharashtra Defence and Aerospace Fund). Its alternative, a majority privately owned entity or a Category I Alternative Investment Fund (AIF) with private Limited Partners (LPs), cannot avoid the returns-amplification problem. 

In the circumstances, the most prudent and effective strategy would be the FoFs route with some sharply defined conditionalities. The funds could be committed at concessional terms - subordinate equity, first loss buffer to a certain threshold, capped returns, warrants with lower liquidation preference, etc. It should be complemented by broad mandates on the nature of investments made, specifically on the TRL stages of the innovations, and pre-defined technology areas. 

When public capital is subordinated to private capital in the waterfall through any of the aforesaid approaches, the public subsidy is targeted precisely at the risk premium that blocks private investment. Return-amplification is minimised because private investors bear disproportionate upside — they are not getting a free subsidy on already-viable deals.

In this context, the RDIF is instructive. For a start, all its funding is debt or equity and only for TRL 4 and above stages. It has three modes of investing based on where it stands in the returns waterfall. In the first mode, RDIF effectively absorbs the first losses and receives distributions after private investors have received their hurdle rate. In the second mode, it receives distributions pari passu with other contributors at the same hurdle rate and IRR. In the third mode, it receives distributions at a higher priority or higher IRR than private contributors. 

While the first mode is a good example of concessional lending as discussed above, the second and third modes may need to be justified on other considerations. Scarce public capital should flow to those areas where it has the highest additionality. While it prescribes the broad areas of investing, it may not suffice in pre-empting returns-amplification investing. 

In the circumstances, the RDIF runs the risk of ending up with the same problems as those with the likes of IDFC and NIIF in infrastructure (whose portfolios clearly indicate that they tend to compete and crowd-out rather than crowd-in private capital). It may struggle to realise the promised additionality. For instance, it is most likely that most of the funding will flow into the TRL 8-9 innovators in the less risky among the defined areas. Finally, I’m not sure how Focused Research Organisations (FRO) can deploy returnable capital in startups, unless they merely act as pass-throughs to FoFs. 

If the second-level fund managers (SLFMs) of RDIF are required to meet additionality criteria (invest in TRL 6-9 companies they would not otherwise fund; report on portfolio-level additionality; face consequences for drift towards safe/commercial deals), the public mandate will be preserved. But without this discipline, every SLFM will cherry-pick the best deals, and the public capital risks becoming a subsidy for private returns. At best, public capital ends up competing with private capital and marginally expanding the large enough and growing pool of risk capital. 

Finally, the biggest constraint to scaling is finding the deployment platform in a country where the indigenous product ecosystem, especially domestic OEMs, is very limited. In the circumstances, public policy must play an important role in value addition by facilitating the creation of scaling pathways. This could be through direct procurements (solar cells, smart meters, street lighting LEDs, etc.) or domestic content mandates (cameras, telecom equipment, etc.). This has been a very important pathway for commercial scaling in both the advanced countries and in China, but it will be a challenge for India’s public policy. It is also for this reason that investors should pursue proactive portfolio management in terms of actively facilitating the linking of startups with the public procurement pathways. 

In conclusion, a few points to be borne in mind. One, grants at Stage 1 (TRL 1-4) are the highest-additionality instrument globally. No other instrument produces a comparable expansion of the investable universe. The evidence is unambiguous. Two, since private capital will remain scarce, public capital is critical to develop the pipeline of risky innovations and startups. Three, this nature of funding and additionality will also largely apply to the stage of technology/product development.

Four, a blended fund with a derisking public tranche and a set of sharply defined target investment-related conditions, is the highest-additionality structured instrument for commercial scaling. The public tranche absorbs the risk premium, and private capital fills behind. Five, government procurement is the highest-additionality instrument for commercial scaling for hardware companies. Procurement creates more private capital crowding-in than any equity instrument, because it proves market demand.

Finally, as public policy interventions to realise the aforesaid objectives, there are perhaps two low-hanging fruits. One, there should be a portal that consolidates all the startups financed by state and central government departments, and it should become the primary universe of the pipeline for risk capital funding. This portal should be tightly integrated with the ecosystems of VCs and other investors. Second, there should be active portfolio management at the level of all departmental funds to facilitate access to larger public risk capital funds like RDIF and SIDBI FFS 2.0. The objective should be to ensure that promising publicly financed innovations do not remain stranded.

Weekend reading links

1. Power constraints could emerge as the biggest bottleneck to America's AI growth vis-a-vis China.

Beijing has already prepared by installing an eye-popping 1,500 gigawatts of new energy capacity since 2021, taking its total to 3,891GW. However, the US has not: its installed capacity has barely risen in recent years, and now sits around 1,373GW — or less than what China added in just four years. This is shocking. Worse, China will add over 3.4 terawatts of electricity-generation capacity in the next five years, according to Bloomberg — six times as much as the US.

2. Industrial policy and infrastructure development are back again as priorities for international development actors in Africa, after couple of decades of dalliance with RCTs and small interventions. 

3. The declining labour productivity growth rates.

Between 1950 and 1973, the metric, based on output per hour worked, rose at an annual average of 4 per cent across developed economies. But the rate halved to 1.9 per cent from 1973 to 2009. And since the financial crisis, it has slowed further, averaging just 1.2 per cent between 2009 and 2025.

Pension systems are grappling with increasing dependency ratios.

When the German retirement age was set at 65 in the 1910s, life expectancy was below 50. It has now increased to over 81, while the retirement age is only set to increase to 67... In the 1960s, Japan had eight or nine people aged 20 to 64 for every person aged over 65. Now it is just over one.

4. Richard Hass makes the point that since America chose the war, it must also make the choice on ending it. This is an important point.

America did have other viable options, above all diplomacy, especially as no convincing case has been made that an imminent threat had to be dealt with militarily. The contrast between Washington’s near-unlimited willingness to compromise and demonstrate patience when it comes to persuading Russia to end its aggression against Ukraine and its unrealistic demands and lack of patience with Iran in the run-up to this war is as stark as it is telling. Ukraine’s offer to help defend against Iranian drones while Russia reportedly provides intelligence to Iran only makes the double standard worse.

5. Around 14.5 million barrels of oil transit the Strait of Hormuz daily. It has shrunk rapidly.

More on the Strait

Iran’s ace in the hole has been its de facto blockade of the strait through which one-fifth of the world’s oil and liquefied gas normally flows. At its narrowest point, the strait is less than 21 nautical miles wide, putting tankers perilously close to drones and missiles from Iran’s southern coastline. Tehran now has near-total sway over the Gulf oil market, forcing neighbours such as Iraq to almost entirely stop production and trapping roughly 300mn barrels of oil and gas in the region, a number that rises by about 20mn every day...

With its new supreme leader, Ayatollah Mojtaba Khamenei, announcing his goal to keep the strait closed indefinitely, Iran has wrongfooted oil traders who had always presumed that US military might would keep the waterway open. Iran has never blocked the strait before, despite its previous threats.

6. Some facts about Indian equity markets.

If we look at data from January 1, 2025, to the end of February 2026... emerging market (EM) equities were up 51.4 per cent, international equities were up by 47 per cent, the United States rose by 18.4 per cent, and total world equity returns were 28.3 per cent. In contrast, over this same 14 months period, India was down 0.7 per cent, the second-worst performing market in the world, with only Saudi Arabia performing worse. In fact, India and Saudi Arabia are the only two markets that are actually down (all returns in US dollar). This is when Korea has tripled, Brazil is up 80 per cent, and Taiwan has risen by more than 50 per cent. We have underperformed the EM benchmark by 5,000 basis points in just 14 months... Absolute foreign ownership of the Indian market is at a 15-year low, and we see foreign portfolio investors (FPIs) selling on a daily basis. India has received zero net foreign flows over the last five years, a very long time indeed.

It has definitively debunked the There is no alternative (TINA) hypothesis.

Our markets had done very well, and many other large EM countries were seen as uninvestable. We are the fastest-growing economy in the world — where else will FPIs go? This was the narrative. This myth has been debunked. If they wish, FPIs can totally ignore us. Five years of net zero flows. There are always choices for capital, and capital only chases potential returns. If we do not offer a good risk/return proposition, nobody will come.

7. Middle East has hundreds of desalination plants.

8. On Monday, the benchmark Brent crude price surged to $119 a barrel before diving to $84, the biggest intraday swing in dollar terms ever. 

9. The Government of India employee count (incl Railways) has remained stationary for the last decade. 

10. M Govinda Rao on the 16th Finance Commission. 

11. After starting out importing everything from China, Ukraine can now make drones with no components imported from China. 

Ukraine will not be mass-producing drones with no Chinese components anytime soon, because it’s still much cheaper to use them. Given China’s dominance of global manufacturing, it is hard to define any drone as truly “China-free.” Many components made outside China still contain Chinese parts or raw materials... Ukraine is one of many nations that have been working to reduce their reliance on Chinese supply chains... Two companies in Ukraine that have built “China-free” drones were picked to compete for contracts in a Pentagon “drone dominance program” under which the United States plans to buy thousands of low-cost attack drones. One of the companies, Ukrainian Defense Drones Tech Corporation, where the men were soldering circuit boards in the basement workshop, was among 11 in all selected last week for possible American drone orders... Ukrainian Defense Drones began making drones in 2023. Initially, all of its components were Chinese. Within a year, however, it had localized production of carbon fiber frames and antennas. By 2025, Ukrainian Defense Drones had expanded to produce flight controllers, speed regulators, radio modems and video transmission systems. Essentially, all its components were made in Ukraine except for the cameras. The company has since gained technology for cameras, too, which it hopes to produce in Europe... 

In the first year after the Russian invasion in February 2022, nearly all of Ukraine’s drones came from China. As demand surged, Beijing imposed export restrictions in 2023 and expanded them in 2024... As the rules tightened, Ukraine resorted to middlemen to buy some parts, and Ukrainian companies began to view the Chinese market as increasingly unreliable. Kyiv turned its focus to building its own drones, and eventually to doing so with fewer Chinese components. By 2024, the vast majority of drones that Ukraine sent to the front were assembled domestically — but still almost entirely with Chinese components. A year later, however, the share of parts from China in Ukraine’s drones had fallen to about 38 percent... Ukraine still buys cheaper Chinese components because the Ukrainian military needs huge numbers of drones and has a limited budget to buy them. Drone missions fail at very high rates, another reason that Ukraine tries to keep costs down.

12.  Finally, on how AI has impacted the Iran-US/Israel war. 

AI is reshaping how the US military makes decisions in war — a shift clear in Iran, where the Pentagon says it struck more than 2,000 targets in just four days... “If we look at the campaign against Isis, the coalition struck around 2,000 targets in the first six months of the campaign in Iraq and Syria,” said Jessica Dorsey, who researches the use of AI and international humanitarian law at Utrecht University... The unprecedented tempo of targeted attacks has been driven in part by AI systems that sift the torrents of intelligence data from drones, satellites and other sensors, generating strike options far faster than traditional human-led planning. The conflict also marks the first battlefield use of “frontier” generative AI models... helping commanders interpret data, plan operations and provide real-time feedback during combat. Over the past two years, the US Department of Defense has extensively integrated AI-enabled technology within its operations. The primary operating system for the Pentagon’s data is Palantir’s Maven Smart System, which alongside Anthropic’s Claude model forms a real-time data analysis dashboard for operations in Iran... 

During a live military operation such as Operation Epic Fury in Iran, Palantir’s Maven platform acts as the software “brain”. It supports the entire so-called kill chain — finding and hitting a target during active conflict. That ranges from identifying and prioritising the target to selecting the appropriate weapon and finally assessing the battle damage. Traditionally, kill chains involved printing off documents and waiting for a senior commander to study and approve it. “Those [older] kill chains are measured in hours and sometimes days,” said a defence tech expert who asked to remain anonymous. “The point of [AI] is to shrink that into seconds and minutes, almost instantaneous.”... As of May 2025, the Maven system was used by more than 20,000 users across 35 military entities in the field, according to public comments by Vice Admiral Frank Whitworth, director of the National Geospatial-Intelligence Agency. That number may be closer to 50,000 users in the US today, according to defence researchers, with Nato also signing up to use Maven in 2025...

The bombing of a girls’ primary school in Minab, in southern Iran, further illustrates the lethal risks of quickly generated or improperly vetted targets... In Iran, AI has potentially already been involved in identifying exponentially more targets than in previous wars, said Utrecht University’s Dorsey. Those targets could have existed beforehand — or they could have been generated quickly by AI systems, creating a serious concern about how carefully these have been vetted as required by law, she said.

This about the Minab school bombing.  

China update - March 2026

The Trump chaos has been a godsend to China by deflecting attention from the world economy’s China problem. It has allowed the country the space to continue flooding the world with its heavily subsidised exports. 

1. Ruchir Sharma has an excellent description of the fundamental problem with China’s growth model. First, some facts about its rising trade surplus. 

This decade, Beijing has dropped export prices by nearly 20 per cent, producing a 40 per cent surge in volume. Booming exports along with weak imports increased China’s trade surplus last year by 20 per cent to a record $1.2tn. Net exports accounted for almost a third of its 2025 GDP growth, a bloated share even by China’s standards. As a share of global GDP, no nation has ever had a larger trade surplus, and that includes Japan during its 1980s heyday. China’s dumping offensive is deindustrialising rival exporters the world over, idling car factories in Thailand and textile plants in Indonesia. Across Asia, nations where Chinese imports are rising fastest also tend to have the weakest job growth. More than 50 of the world’s 70 largest economies have taken steps to defend themselves against Chinese dumping.

This has come at a prohibitive cost and engendered economy-wide distortions.

The root of the problem is the country’s growth target. Of the nearly 40 nations that rose into the developed ranks after the second world war, none faced the twin hurdles confronting China today: depopulation and massive debt. No other major nation in history has managed to sustain growth above 2 per cent with a shrinking labour force. And at 340 per cent of GDP, China’s total debts are higher than any other emerging economy by far. Beijing is trying to engineer a historically implausible miracle. Given its demographic decline, China can hit its target only by raising output per worker, but maintaining overall productivity growth near 5 per cent would be an unprecedented feat at this stage of development. Lately China has been moving in the opposite direction. Productivity growth includes contributions from labour, capital, and a critical “total factor” that aims to capture how much growth labour is squeezing out of the investment. The Conference Board estimates that this key third factor has fallen to near zero this decade, implying that China is generating growth only by investing more heavily.

China keeps pumping out credit to fund more investment, but mostly it’s getting a bigger debt pile. To generate $1 of GDP growth in China, it now takes $6 of new debt, up from $1 two decades ago. Beijing is counting on investment in new technologies including AI to boost productivity, but it’s highly unlikely that boost will be big enough to sustain productivity growth near 5 per cent. China’s real potential growth rate is probably between 2 and 3 per cent… The real problem is that investment keeps growing faster than consumption, compelling China to flood the world with its excess production. Chinese exports, now $3.8tn a year, recently surpassed US imports for the first time.

2. The record $1.2 trillion trade surplus in 2025 and the claim of having achieved 5% GDP growth rate for the year should not detract from the increasing pile of problems facing the economy. This is a good summary of the signatures of China’s faltering economy.

The growing dichotomy between China’s thriving trade-focused sector and its anaemic domestic economy. While China’s world-conquering exporters and powerhouse innovation clusters can make the country seem like an unblemished economic success, painful technology transitions and faltering domestic demand mean that for many businesses and citizens, these are times of increasing hardship. Cities such as Shenyang, which turned itself from a centre of heavy industry to an automotive hub during market reforms in the 1990s and 2000s, are struggling to evolve. Shenyang now wants to pivot into electronics and other industries but, like many Chinese provinces, is unwilling to let favoured businesses die…

Yet investment was the weakest since the 1990s as property prices fell and new construction starts declined… A four-year real estate slump has undermined domestic demand and added to deflationary pressures. In November, retail sales growth hit a three-year low. Meanwhile, Beijing’s interventionist policies from currency depreciation to subsidies to support for favoured industries are driving overcapacity in sectors ranging from automotives and batteries to solar panels. Plunging profitability — industrial profits fell 13 per cent year-on-year in November — has made companies reluctant to hire or pay high salaries. 

Zombie companies now account for more than 12 per cent of total registered companies, more than double the level in 2018 and nearly double the global proportion, according to a study led by Alicia García-Herrero, chief Asia-Pacific economist at French investment bank Natixis… Unemployment among those aged 16-24 was 17 per cent in November compared with about 11 per cent pre-pandemic. While China’s overall official unemployment rate remains stable at about 5 per cent, many ordinary Chinese people say the figures do not mirror reality.

Ironically, the rising trade surplus may carry the seeds for the weakening of the domestic economy,

As long as Beijing can rely on exports for growth, analysts expect it to let the housing market continue to deflate and to concentrate on boosting the high-tech sector to compete with the US — choices that will only deepen the economic divide.

3. Another outlet to channel domestic capacity on the face of weakening domestic demand is the projects under the Belt and Road Initiative (BRI). China’s investments under the BRI Project surged in 2025.

The surge in new investment and construction deals was dominated by gas megaprojects and green power, according to research by Australia’s Griffith University and the Green Finance & Development Center in Shanghai. Beijing signed 350 deals last year, up from 293 worth $122.6bn in 2024... Last year’s figures brought the total cumulative value of BRI contracts and investments since its launch to $1.4tn, the study found. The growth in 2025 was driven by multibillion-dollar megaprojects including a gas development in the Republic of the Congo led by Southernpec, Nigeria’s Ogidigben Gas Revolution Industrial Park led by China National Chemical Engineering and a petrochemical plant in North Kalimantan, Indonesia, led by a Chinese joint venture of Tongkun Group and Xinfengming Group.

Fossil fuel exploitation became the main focus of China’s energy engagement with its BRI partners.

The value of energy-related projects last year was $93.9bn, the highest since the BRI’s inception and more than double the 2024 level. It included $18bn in wind, solar and waste-to-energy projects, underscoring China’s lead in clean technology. Metals and mining also hit a record at $32.6bn, including a majority of spending on minerals processing abroad, highlighting how Beijing has used the BRI to secure long-term access to resources. That included a surge of investment in copper in the second half of the year.

The scale and direction of the shift in 2025 point to China tightening global mineral supply chains toward itself and excluding the US. 

4. The NYT writes that China’s campaign to dominate rare earths has its origins in the 1960s. In April 1964, China discovered that an iron ore mine near Baotou, 50 miles from the Mongolian border, held the world’s largest deposit of the 17 metals that belong to the rare earth family. Today, China produces 90% of the world’s rare earths and rare earth magnets. 

Deng Xiaoping, then a high-ranking Chinese Communist Party official, visited the remote desert mine, owned by a military steel maker, to inspect the massive cache.“We need to develop steel, and we also need to develop rare earths,” declared Mr. Deng, who over a decade later would emerge as China’s top leader… China’s centrality in rare earths didn’t happen by accident. It is the result of decades of planning and domestic and overseas investment, often directed from the highest levels of the party and the Chinese government. In the early 1970s, the People’s Liberation Army launched a little-known research program to develop potential military uses of rare earths. 

Mr. Deng kept pushing China’s rare-earth advancement forward in the 1980s and 1990s together with Wen Jiabao, a geologist by training who went on to serve as China’s premier from 2003 to 2013. Under Mr. Wen, China consolidated what was a highly splintered web of mostly private companies into a tightly run arm of the Chinese government. Mr. Wen closed mines run by smugglers and cleaned up the industry’s most severe pollution. The sector grew in size and expertise. In 2019, seven years into his reign as China’s top leader, Xi Jinping described rare earths as “an important strategic resource.”.. In April and again in October, 2025 China enacted new export controls that allowed it to withhold supplies of rare earths and rare-earth magnets and force Mr. Trump to compromise on tariffs…

This is a fascinating account of how Deng and Co stitched together the rare earths industry in China. 

The father of the industry was Xu Guangxian, a tall, thin man from Shaoxing, a town near Shanghai… Shortly after World War II, Mr. Xu completed a doctorate in chemistry at Columbia University. He returned home to teach and do research at Peking University in Beijing… Purifying rare earths is extraordinarily difficult. Early chemists named them rare not because they were hard to find — they are not — but because of the challenge of separating them from one another… At Peking University, Mr. Xu and his wife, Gao Xiaoxia, also an accomplished chemical engineer… had a revolutionary breakthrough: Rare earths could be purified using inexpensive hydrochloric acid and cheap plastic holding tanks rigged together. 

Mixed rare earths were poured in one end, and specific kinds of rare earths, after binding to various solvents, emerged from different outlets at the other end. It was the first rare-earth assembly line, a crude version of a process still used today. Production costs plummeted with Mr. Xu’s techniques. Mr. Xu installed his first production lines, in Baotou and at a chemical plant in Shanghai, and started training engineers from all over China… A Five-Year Plan drafted by Mr. Deng and Vice Premier Mr. Fang Yi, covering 1981 to 1985, ordered that China “increase the production of rare-earth metals.” More than 100 towns and villages across China built rare-earth refineries in the 1980s, many of them state-owned and few with meaningful pollution controls. By 1986, China was the world’s largest producer of rare earths.

Rare earths were then used in low-tech manufacturing. Thanks to research in the US and Japan, rare earths became central to advanced manufacturing. But Chinese expertise in rare earths also owes to the strategic short-sightedness of the US. 

In 1983, engineers at General Motors and the Japanese magnet maker Sumitomo Special Metals each announced they had developed powerful rare-earth magnets. The magnets were immediately put to use in electric motors in the auto industry and beyond. China lacked expertise to turn rare earths into magnets. It would purchase that know-how from the United States. General Motors had turned its discovery into a thriving magnet manufacturing subsidiary in Indiana, called Magnequench. But a decade later, G.M. decided to stop making many of its own auto parts.

Magnequench was sold in 1995 to a consortium of investors that included two Chinese companies led by sons-in-law of Mr. Deng: Wu Jianchang and Zhang Hong. Under President Bill Clinton, the U.S. government allowed the transaction to proceed because a majority of the owners were American. The American owners were mainly institutional investors. Mr. Deng’s sons-in-law had led companies with deep ties to low-cost magnet manufacturing in China. Magnequench started moving its equipment in 2001 to Tianjin and Ningbo, China, and shut down in Valparaiso, Ind., by 2004… the Chinese magnet factory at Tianjin had previously been using processes “that were at least 10 years behind” what Magnequench had developed in Valparaiso. The move by Magnequench, which was then bought in 2005 by a Canadian rare-earth processor with operations in China, taught China how to make rare-earth magnets.

Under Wen Jiabao, China focused on the pollution stemming from rare earth refineries, which were contaminating the Yellow River. In 2006, China imposed annual quotas on exports to limit processing and stem pollution, and has since consolidated companies under state ownership. Its dominance also stems from capabilities 

Today, China is working to cement its lead in rare earths by churning out more trained technicians and researchers than any other country. Programs in rare earths are offered by 39 universities. The United States and Europe have no such programs — not even at Iowa State University, an institution that once trained generations of American engineers in rare earths. Iowa State has not offered a course in rare earths for the last several years and has one graduate student doing independent study in the field. It plans to offer a course next year. China has hundreds of scientists exploring rare-earth technologies. 

Technicians at a refinery in Wuxi, a city near Shanghai, spent seven years doing experiments to refine dysprosium, a rare earth, to extraordinary purity. The refinery is now the world’s sole source of that rare earth, which is used in capacitors — tiny devices to control electricity — found in Nvidia’s Blackwell artificial intelligence chips. Most of the refinery’s shares were until this year owned by Neo Performance Materials, the Canadian company that acquired Magnequench in 2005. A Chinese state-controlled company bought most of the shares on April 1. Then on April 4, Beijing halted exports of dysprosium and six other kinds of rare earths to the United States and its allies… Beijing has halted most exports of rare-earth processing equipment. It has also taken away the passports of rare-earth technicians to prevent them from leaving the country with valuable information.

5. In a good illustration of regulation with Chinese characteristics, Beijing manages to strike a balance between strictly controlling sales of unregulated drugs inside the country, while allowing their massive and growing illegal exports. 

FT has a story about the burgeoning trade in injectable peptide drugs that have become popular in the West due to the success of GLP-1 weight-loss peptide drugs like semaglutide and tirzepatide. 

Industry insiders estimate there are about 1,000 Chinese sellers targeting overseas customers, a figure that has climbed sharply in recent months. Rising competition has pushed prices down. Sellers quoted prices such as $65 for 10 vials of BPC-157 at 10mg, or $70 for the same quantity of semaglutide — which compares to $110 and $1,000 for the same quantities on US websites… Most peptides are produced by about a dozen factories clustered in Shenzhen and Changsha, the capital of Hunan province. These facilities originally manufactured active pharmaceutical ingredients for the pharmaceutical industry before pivoting towards the grey market. 

Multiple layers of intermediaries now sit between factories and consumers. “We never touch the product. We don’t know who makes it,” said one seller… one seller said he had no intention of trying the products himself. “I’m overweight, but I wouldn’t dare take these drugs,” he said, laughing. “It’s westerners who are obsessed with them. I just sell them.”… Chinese authorities have been cracking down on domestic sales of unregulated GLP-1 weight-loss drugs, arresting scores of sellers. A review of court records shows at least 40 cases of people being charged with black-market peptide sales… 

Finding sellers online is easy. They advertise openly on cross-border ecommerce platforms such as Global Sources, as well as on Facebook, Telegram and WhatsApp… The FT visited the registered addresses of eight suppliers and found that most used false locations, with no functioning phone numbers or email contacts… Inside one office in China visited by the FT, a group of young women chatted with customers in Brazil, the US and Canada. They used ChatGPT to draft sales copy for WhatsApp messages and worked with western influencers who promote the products on TikTok and Facebook in exchange for commissions… Sellers also shoulder the risk of shipments being seized by logistics companies or customs authorities… 

6. Scott Kennedy has a very good paper on China’s high-tech sector—some headline facts.

Chinese industrial policy spending was estimated to be 4.9 percent of GDP. More recently, an International Monetary Fund (IMF) study, applying a similar methodology to data from 2011 to 2023, found that Chinese industrial policy spending ranged between 4.0 and 5.5 percent of GDP. The authors did not compute a parallel estimate for the United States, but they found that China’s spending was over three times the amount outlaid by the European Union… 

According to the World Bank, China contributed $625.2 billion in domestic value added to manufactured goods in 2004, just 8.5 percent of the world total. By 2024, China’s output had jumped to $4.7 trillion, accounting for 28.0 percent of all manufacturing value added globally… China’s share of domestic value added in manufacturing has risen across multiple industries since 2000… Critically, the share of China’s exports by foreign-invested domestic firms has dropped over the same period, from nearly 60 percent to 27 percent, meaning that this shift toward higher domestic value added is not a reflection of the influence of foreign-invested firms in China. 

The report points to the wide variations in the outcomes of China’s industrial policy interventions. It uses a four-fold framework to categorise them.

Chinese EV firms are estimated to have received nearly $231 bn in fiscal support in the 2008-23 period.

An area where China’s industrial policy has struggled is in the semiconductor industry. Despite outspending competitors by a large margin…

… Chinese firms have struggled to make headway in the value chain.

7. Finally, did China really eliminate poverty? 

But while the country has eliminated poverty according to the World Bank’s $3 per day income standard, its definition of poverty is much lower than what the lender considers poverty in an upper-middle-income country such as China. By 2022, more than one in five people in China remained in poverty according to the World Bank’s definition for an upper middle-income country, set at $8.30 of income per day at 2021 prices.

Labour market in times of technological changes

The impact of AI on the economy, especially on the labour market, is most likely to be the defining political economy issue of this generation. 

In this context, it is useful to understand the trajectory of the evolution of jobs with technological changes over the last century or so. Daron Acemoglu and Pascual Restrepo have shown that roughly half of America’s employment growth between 1980 and 2010 came from the creation of entirely new occupations. They argue that the automation effect of the displacement of workers is offset by the reinstatement effect arising from the creation of new occupations. 

David Autor, Caroline Chin, and Anna Salomons have a paper which examined the substantive content of emerging job categories (or new work) over the 1940-2018 period in the US, where it comes from, and its effect on labour demand. Augmentation innovations are those that increase capabilities, quality, variety, or utility of the outputs of occupations, thereby generating new demands for worker expertise and specialisation. They constructed a database of new job titles linked both to US Census microdata and to patent-based measures of occupations’ exposure to labour-augmenting and labour-automating innovations. 

We find, first, that the majority of current employment is in new job specialties introduced after 1940, but the locus of new work creation has shifted—from middle-paid production and clerical occupations over 1940–1980, to high-paid professional and, secondarily, low-paid services since 1980. Second, new work emerges in response to technological innovations that complement the outputs of occupations and demand shocks that raise occupational demand; conversely, innovations that automate tasks or reduce occupational demand slow new work emergence. 

Third, although flows of augmentation and automation innovations are positively correlated across occupations, the former boosts occupational labour demand while the latter depresses it… Employment and wagebills grow in occupations exposed to augmentation innovations and contract in occupations exposed to automation innovations… augmentation innovations increase occupational wagebills by boosting both employment and wages suggests that ‘new work’ may be more valuable than ‘more work’—plausibly because new work demands novel expertise and specialization that (initially) commands a scarcity premium… we establish that the effects of augmentation and automation innovations on new work emergence and occupational labour demand are causal. Finally, our results suggest that the demand-eroding effects of automation innovations have intensified in the last four decades while the demand-increasing effects of augmentation innovations have not.

Routine task-intensive occupations gained substantial new titles between 1940-80, and few between 1980-2018.

What most stands out from this figure is the shifting fortunes of routine task-intensive occupations—both blue-collar occupations such as operative and kindred workers, metal workers, and mechanics; as well as white-collar occupations such as shipping and receiving clerks; stenographers, typists, and secretaries; bank tellers and bill and account collectors; and library attendants and assistants. 

So what does this all mean for the labour market in the times of AI? In a much discussed essay, Dario Amodei, CEO of Anthropic, sounds alarm that AI could displace half all white collar jobs in 1-5 years.

The pace of progress in AI is much faster than for previous technological revolutions. For example, in the last 2 years, AI models went from barely being able to complete a single line of code, to writing all or almost all of the code for some people—including engineers at Anthropic. Soon, they may do the entire task of a software engineer end to end… it just implies the short-term transition will be unusually painful compared to past technologies, since humans and labor markets are slow to react and to equilibrate... AI will be capable of a very wide range of human cognitive abilities—perhaps all of them. This is very different from previous technologies like mechanized farming, transportation, or even computers. This will make it harder for people to switch easily from jobs that are displaced to similar jobs that they would be a good fit for… 

AI is increasingly matching the general cognitive profile of humans, which means it will also be good at the new jobs that would ordinarily be created in response to the old ones being automated… Across a wide range of tasks, AI appears to be advancing from the bottom of the ability ladder to the top. For example, in coding our models have proceeded from the level of “a mediocre coder” to “a strong coder” to “a very strong coder.” We are now starting to see the same progression in white-collar work in general… AI, in addition to being a rapidly advancing technology, is also a rapidly adapting technology… Early in generative AI, users noticed that AI systems had certain weaknesses… But pretty much every such weakness gets addressed quickly— often, within just a few months.

However, in a recent issue, The Economist disputed such alarming prognostications.

We analysed employment and wage trends across more than 100 large white-collar occupations in America since the second half of 2022. Employment across the sample has risen by 4% and real wages by 3%. To get a sense of AI’s impact on different roles, we used occupational descriptions to classify white-collar roles into four groups depending on the bundles of tasks involved: technical specialists, managers and co-ordinators, care workers, and back-office employees. We then tracked employment in each group starting in late 2022, using six-month moving averages. 

Roles that combine technical expertise with oversight and co-ordination have enjoyed the biggest gains. Employment among project managers and information-security experts has risen by 30% or so. Other occupations which combine deep expertise in maths-related fields with problem-solving are also thriving. So are jobs which involve interpersonal care work and those which demand judgment and co-ordination. Only routine back-office work has shrunk. Over the past three years or so the ranks of American insurance-claims clerks have shrunk by 13% and those of secretaries and admin assistants by 20%.

It also finds AI generating all-new jobs - data annotators, forward-deployed engineers, chief AI officers, and mainly those without settled names (“other occupations”). 

On the impact of AI on jobs, the Yale Budget Lab has a meta-study that compares evidence from across studies. It uses seven different measures of occupation-level AI exposure calculated by researchers and urges caution in reading too much into the findings. These measures are based on human and AI assessments of whether a job’s constituent tasks can theoretically be performed by an LLM, linking tasks with AI-related patents, and using real-world data on how LLMs are being used to carry out particular work-related tasks. 

Its headline findings:

AI exposure metrics broadly agree with each other, but they disagree with each other more on highly exposed occupations. The key point of disagreement between different AI exposure metrics is in the magnitude of exposure, not whether an occupation is exposed. Occupational exposure to AI is not indicative of a jobs AI will automate out of existence. Rather, it indicates places in the labor market where AI could have an impact.

The study uses each occupation’s exposure and variance across the various scores, with low variance pointing to a consensus on exposure. They regressed the two and found greater disagreement for highly exposed occupations, “driven more by how much an occupation is exposed more than whether it is exposed.”

Clearly, occupations focused on computational, text-based, or administrative work tend to have both higher variance and higher average exposure, whereas manual fields like construction and maintenance have lower disagreement and variance. 

The high degree of variance and disagreements point to the perils of forming opinions on the trajectory of AI’s evolution and its impact on the labour market. The meta-study urges caution in drawing conclusions “about where AI disruption to the labor market could be going”. 

John Burn-Murdoch and Sarah O’Connor in the FT point to a few perspectives that are important while considering the impact of AI. Specifically, they point to the nature of the AI exposure and the regulation of AI adoption. 

It is more than 20 years since David Autor and his co-authors argued convincingly that, like the waves of technological change that came before, computerisation threatens jobs where workers are mainly performing tasks to meet a specification, but is a complement to those who determine the specification. Viewed through this lens, the AI revolution may pose less risk to (or even benefit) a software developer who exercises considerable autonomy over what they work on and how they do it, than to a warehouse worker who loses out to a new generation of AI-enhanced robots, or a retail sales assistant whose store is closed as technology drives ever more commerce away from brick and mortar stores and onto the web…

Or consider the role of regulation. As we have written previously, AI models can now evaluate medical scans more accurately than experienced radiologists, but regulatory barriers and insurance policies have made it virtually impossible for fully autonomous systems to be used. Meanwhile, laws have sprung up across the US prohibiting AI tools that “provide services that constitute the practice of professional mental or behavioral healthcare (such as therapy)”. Whatever one’s views on the rights or wrongs of these particular cases, they are clear demonstrations that vulnerability to occupational displacement in the age of AI comes down to far more than “Is AI capable of performing the tasks that make up your job?”

In conclusion, on the overall likely impact, the central questions revolve around three trends - labour automation (associated displacement), labour augmentation (and associated redeployment), and the emergence of new work categories. What will be the relative impacts of the three? Will the first far exceed the second and third? Or will they offset the first? What will be the pace of the first? Will the second and third lag the first significantly?

It is impossible to answer any of these with a degree of confidence. We may only be able to wait and watch how the trends play out and respond accordingly.