Weekend reading links

1. A reality check on who owns agricultural land in South Africa.

White farmers still own roughly half of the country’s land although only 7 per cent of citizens are white.

2. Tej Parikh has a very good graphical summary of America's healthcare market.

The US spends more than $4.5tn annually on healthcare — and is projected to soon account for one-fifth of its economy. Even on a per capita basis, other large, rich nations spend about half as much as America. Healthcare is the largest component of US consumer spending on services (well above expenditure on recreation, eating out and hotels)… The economy has created 3.9mn private sector jobs since the start of 2023. More than half have come from healthcare and social assistance… studies have estimated that approximately 25 to 30 per cent of health spending could be considered waste.

Healthcare is such a major contributor to growth that any reduction will automatically impact job creation and economic growth. 

3. ExxonMobil, Occidental Petroleum, Equinor, and others are piloting a new drilling technique for lithium, direct lithium extraction (DLE), in the Smackover Formation area of the Southern US that has a massive brine aquifer. 

Underground brine reservoirs flowing across Arkansas and neighbouring states contain high concentrations of the silvery-white metal; a US Geological Survey study published in October estimated the total resource in south-west Arkansas alone at up to 19mn tonnes... “DLE could do for the US lithium industry and economy what fracking did for the US oil industry almost 20 years ago,” says Andy Robinson, a geoscientist and co-founder of Standard Lithium, which is seeking to develop a $1.5bn project near El Dorado in partnership with Norwegian energy group, Equinor.

Proponents say DLE offers a faster and less environmentally damaging alternative to existing extraction methods. For oil companies, which have extensive skills in drilling, pumping and processing fluids, it represents a useful way to diversify their businesses... But experts warn that US lithium pioneers must prove the new technology can be commercially successful at scale and compete with both existing extraction technologies and rival DLE projects in lower-cost countries...

Between 2020 and 2024 global demand for lithium tripled to around 1.2mn tonnes, according to energy research group Wood Mackenzie, which is forecasting lithium consumption will reach 5.8mn tonnes by 2050. To meet demand, producers have over the past decade expanded hard rock mining in Australia and China and lithium brine extraction in Latin America, giving these three regions control of more than 80 per cent of the extraction industry. Hard rock mining of lithium is much like any other metal production process; ores such as spodumene are excavated from open pit mines, crushed and chemically processed to separate the lithium. Brine extraction involves pumping lithium-rich brines into large ponds, typically in regions with a hot, dry climate. The water gradually evaporates, leaving behind concentrated lithium salts that can be processed...

Until recently, US-based lithium miners have struggled. They face higher costs, tougher mining regulations and less favourable geological and climatic conditions than in the “lithium triangle” in Chile, Argentina and Bolivia. The development of direct lithium extraction, which usually involves using solvents or ceramic materials to separate lithium from the brines, has changed all that. DLE takes a matter of hours to separate lithium from brines, while evaporation ponds can take as long as 18 months. Recovery rates are around 70 to 90 per cent, according to Wood Mackenzie, compared to 40 to 60 per cent for evaporation ponds, and DLE also uses less land and less water. Combined with the discovery of high concentrations of lithium in oilfield brines within the so-called Smackover Formation, which extends across Arkansas, Louisiana, Texas, Alabama, Mississippi and Florida, DLE has opened up an opportunity. Existing oil and chemical infrastructure in the formation also makes these resources more accessible than greenfield sites.

4. Great primer in NYT that has the list of all items Americans import from China. Goods that Americans import mostly from China.

The highest value of goods imported from China.

And America's biggest exports to China.

5. Tata Electronics bets big on iPhone manufacturing. But it comes with exacting standards on quality and productivity.

The company’s ambition to become an iPhone-manufacturing hub collides with the reality of high attrition, relentless production targets, and the ever-present pressure of Apple’s quality control. Inside the factory, each worker undergoes two to three weeks of intensive training before stepping onto the assembly line. Once there, their tasks are highly compartmentalised—a deliberate strategy to protect Apple’s intellectual property. Most workers only know how to assemble a specific section of the phone, with little visibility into the broader production process... An iPhone must pass through at least 600 quality checkpoints before it leaves the factory. A single defect can jeopardise an entire batch, sending costs skyrocketing and potentially damaging the supplier relationship. This is why Tata has invested heavily in automated equipment from suppliers like Delta Electronics, aiming to reduce defect rates and increase efficiency... the workers... shifts are standard eight-hour stints—6 am to 2 pm, 2 pm to 10 pm, or 10 pm to 6 am... That compartmentalised operation is about efficiency but also about Apple’s intellectual property. Keeping workers focused on their slice of the process helps prevent any accidental leaks of trade secrets.

6. Starlink compared to other telecom companies.

Also this article on Business Standard.

7. China tries to increase its soft power. The one area it appears to be having some success is gaming. 

Four of the ten highest-grossing mobile games of 2024 were made in China. One such is Genshin Impact, a role-playing adventure which rakes in over $1bn a year. Last year a Chinese firm released Black Myth Wukong, the country’s first blockbuster video game. Featuring the mischievous Monkey King, it is steeped in Chinese folklore. Some 30% of its 25m players are said to be outside the country.

8. One of the genuine successes of the Indian state, the taming of the Naxalite movement, which is perhaps in its end stages.  

9. Global Capability Centres (GCCs) are driving a boom in Grade A real estate in India that are ESG-compliant, and equipped with smart technology systems. 

Between 2022 and the first half of 2024, GCCs have leased 53 million square feet (msf) of office space. In 2024, they accounted for 36 per cent of total leasing activity, occupying 27.7 msf of the 77.2 msf transacted... The momentum has continued into Q1 CY25. Colliers reported that GCCs absorbed 6.5 msf of Grade-A office space in the quarter — constituting 41 per cent of overall office space demand across the top seven cities in India... In Q1CY25, GCCs leased 88 per cent of the total office space in green buildings as part of their broader commitment to achieving carbon neutrality...

According to Vestian’s sustainability report, green-certified office buildings commanded an average rental premium of 12–14 per cent over non-certified buildings. GCC-occupied office space in Bengaluru has been leased at a 50 per cent premium compared to non-GCC-occupied office space in FY25. The premium is 13 per cent in NCR and 9 per cent in Hyderabad... office rentals across the top Indian cities have grown between 9 to 28 per cent from 2022 to 2025, mainly driven by GCCs... real estate costs for the GCCs are not more than 6-7 per cent of their total cost of a GCC setup, which does not deter them from going for high-quality locations... According to Nasscom-KPMG report, the GCC market size in India tripled from $19.6 billion in FY15 to $64.6 billion in FY24. It is further anticipated to touch $110 billion by 2030, despite ongoing trade tensions and geopolitical frictions.

10. US-India pre-Trump merchandise trade tariffs.

11. Bola Tinubu's shock therapy appears to be working for the Nigerian economy.

On day one Tinubu removed a ruinously expensive fuel subsidy. More important still, the central bank has restored monetary policy orthodoxy after a shambolic era in which only cronies with access to cheap dollars benefited. After a dangerous overshoot, the naira has stabilised, with the gap between the official and black market rate shrinking to almost nothing. The central bank has stopped printing money to pay for government profligacy. Politicians still spend too much, often on fripperies like an extravagant presidential jet, but at least the government has begun to increase tax receipts. Investors do not live in constant fear of a devaluation and can readily access dollars. That may eventually help Nigeria to diversify, but shorter term it is positive that oil production has recovered from a nadir of 1mn barrels a day to nearly 1.5mn last month. Oil theft has been reduced and local companies are squeezing more out of marginal fields.

12. The decline of net FDI into India.

In 2020-21 and 2021-22, gross FDI inflows were adversely impacted by repatriation and outward investments to the tune of 46 per cent and 54 per cent, respectively. The extent of this impact rose sharply in the following three years — to 61 per cent in 2022–23, 86 per cent in 2023–24, and 99 per cent in 2024–25... the amount of repatriation and disinvestment in 2019-20 was about $18 billion, or about 25 per cent of gross FDI inflows. But the following two Covid years saw repatriation and disinvestment rising to account for a 33-34 per cent share of gross FDI inflows. In 2023-24, this trend became alarming, with the share of repatriation and disinvestment in gross FDI inflows jumping to 62 per cent. In 2024–25, the share inched up further to 63 per cent.

What this implied was pretty serious. Foreign investors in Indian companies were showing a marked preference for ploughing back their gains from here to reinvest in other markets elsewhere. Note that this trend has continued for the last two years... Indeed, reinvested earnings by existing foreign investors have stayed at well below a third of gross FDI inflows in these years. Nor has there been a marked desire on their part to increase reinvested earnings... Contributing to such gloomy prospects on the net FDI inflows front is last year’s data that shows how Indian companies are raising their outward FDI in a big way. Indian companies have stepped up their outward FDI during the post-Covid years — from $14 billion in 2022-23 to $16.6 billion in 2023-24, and to $29 billion in 2024-25.

In 2024-25, while India attracted $81 bn in FDI, foreign firms repatriated over $51 bn, and Indian firms' outward investment was $29.2 bn, leaving the net FDI inflows at only $0.35 bn. 

13. In a bid to overturn a system that the government believes is biased against it, Mexico goes to polls on June 1 to elect judges!

In elections on June 1, Mexico will replace almost 900 judges at the federal level and hundreds more across 19 state-level jurisdictions in a voting process never tried elsewhere that was implemented in just eight months... A random lottery decided which half of federal judges would be replaced on Sunday, and which in 2027. Most candidates for the vote were chosen by the ruling party and were not allowed any public or private funding. Some are openly associated with the ruling Morena party... The electoral institute expects turnout of about 8 to 15 per cent, compared with more than 60 per cent in last year’s presidential election... “Less than 1 per cent understand what they are voting for,” Jorge Sepúlveda, vice-president of the Mexican Bar Association. “Those that’ll vote will mostly be people propelled by the government.”... In Mexico City, voters must fill out nine ballots, choosing about 50 names from a choice of almost 300. Specialist judges were assigned to certain districts, meaning voters in parts of the capital will choose all the country’s competition and telecoms judges... An all-powerful disciplinary tribunal will be able to remove judges. Of 38 candidates for that, at least 10 have ties to the ruling party, including two who worked directly for López Obrador... One anti-corruption group identified 17 “high risk” candidates in judicial elections, including one who had worked for the Sinaloa Cartel’s leader and another who had worked for the leader of the Los Zetas criminal group. Saúl López, professor at Tecnológico de Monterrey’s school of government, said that the new system would offer “the maximum degree of capture, not just by organised crime but other economic powers”.

14. The disturbing monopoly in cloud computing.

Unlike traditional utilities, the dominant cloud providers Amazon, Google and Microsoft — which together control two-thirds of the global market — operate with minimal transparency or public oversight. This leaves governments, businesses and citizens vulnerable to systemic risks, while giving these corporations immense power to shape the digital economy to their advantage. It is no accident that the same behemoths that dominate ecommerce, digital advertising and operating systems also control the cloud computing infrastructure that underpins these services. Cloud is an extraordinarily capital-intensive business, with high barriers to entry and significant network effects. The data, technological capabilities and financial reserves controlled by these behemoths secured them advantages that smaller, independent rivals simply couldn’t match when cloud computing began to take off. But the companies haven’t just benefited from structural advantages; they’ve also engaged in anti-competitive practices, as documented by competition authorities across Europe, the US, Australia and Japan. These include opaque and discriminatory pricing, technical barriers to switching provider, excessive fees for data transfers and bundling cloud services with other products...

The dependence of many nations on a small number of US cloud giants is a geopolitical threat. Several existing US laws — including the Cloud Act — require providers to hand data to the American government when asked, even if stored on foreign soil... Big Tech’s cloud oligopoly undermines innovation. In artificial intelligence, for example, tech giants have been accused of trading cut-price access to cloud resources for intellectual property rights, equity stakes and strategic influence over leading start-ups, reinforcing their dominance across the sector.

Possible responses to this monopoly

Fortunately, most of the tools we need to address these problems already exist. Established frameworks — including utility regulation, competition policy and public procurement — can be drawn on to restructure and govern cloud infrastructure in the public interest. For instance, regulators should mandate fair and non-discriminatory access to cloud services, mirroring rules already applied to telecoms. This should include transparent, consistent pricing and a ban on unfair contract terms. Providers should be required to implement robust processes to ensure the stability and security of their infrastructure, with regular audits and stress tests. Governments should also rethink their procurement practices. Public institutions should not reinforce monopoly power by defaulting to the dominant providers. Finally — and most ambitiously — governments should consider structural separation. Requiring Amazon, Google and Microsoft to spin off their cloud divisions would eliminate their ability to use this critical infrastructure to extend their dominance into new markets.

15. With high tariffs comes trade crime.

In April, for example, Chinese exports to the United States fell 21 percent from a year earlier, but Chinese exports to Southeast Asian countries rose by the same percentage... An analysis by Exiger, a data analytics firm, found that more than 3,000 companies in Mexico depended on Chinese shipments for 75 percent or more of their supply chain. Many of these companies are subsidiaries of Chinese state-owned enterprises, and most sell products to the United States, the report said.

16. As PSG faces Inter Milan in this weekend's Champion League final, Simon Kuper writes that Paris has become global football's biggest talent pool.

Paris finally acquired a serious football club in 1970, when little Paris FC and Stade saint-germanois merged into PSG. (Paris FC soon walked out again.) At the time, the city’s growing suburbs, the banlieues, were filling with kids who had few entertainments besides football. In new towns short on markers of belonging, millions grew up supporting PSG as a way to feel Parisian. The popular claim that it’s a fake club with money but no fans is nonsense. The French state funded accredited coaches and artificial pitches in the banlieues. Soon, Greater Paris was producing more top footballers than certain continents. French teams packed with Parisians have reached four of the seven World Cup finals since 1998, winning two, and losing two only on penalty shoot-outs. The previous time PSG reached the Champions League final, against Bayern Munich in 2020, they lost to a goal by Bayern’s Parisian exile Kingsley Coman, but Parisian talent goes a long way down... PSG’s rise began in 2011, when the French president Nicolas Sarkozy, a fan, encouraged a wing of Qatar’s state to buy the club for a piffling €70mn or so. Sarkozy rooted out the hooligans, and Qatar bought superstar players. Two years ago, PSG’s front three were Kylian Mbappé, Neymar and Leo Messi. Yet PSG fans (I have two in my apartment) prefer today’s younger, harder-working, less-spoilt side.

17. Manish Sabharwal has a good compilation of "regulatory cholesterol"

Can women in India work the same jobs and the same way as men? No, they are banned from 32 operations and 200 sub-processes, including pottery manufacturing, cashew-nut processing, and glass manufacturing. Can employers think about hiring men and women for night shifts similarly? No: Women attract 59 special conditions for employers across states. Can factories use all their land? No: Fifty per cent of an industrial plot is lost to just three standards; micro and small factories lose the most land to standards more stringent than those of countries 10 times richer. Can workers work the hours they want? No: A factory worker loses 270 plus hours of annual earnings to working hour restrictions, and these limits force workers to give 156 to 416 fewer hours in a quarter than in Japan. Is building one 300-worker factory cheaper than two 150-worker factories in India? No: One 300-worker factory needs 40-80 per cent more land than two 150-worker factories. Do India and Singapore require the same number of floors to build a hotel with the same number of rooms? No: The same number of rooms requires three floors in Singapore and seven in Noida. Can all of rural India industrialise? No: Fifty per cent of rural areas cannot be industrialised due to minimum road width norms.

Corporate power and elite capture - Jeff Bezos and Amazon edition

I have blogged on multiple occasions (this and this) that the biggest challenge to the social contract from widening inequality comes from the consequent inevitable capture of the political processes and the rules-making institutions.  

Jeff Bezos and Amazon are only the latest examples of how this capture plays out. Here are two illustrations from recent newspaper reports.

It has just been reported that as the owner of the Washington Post, Jeff Bezos, has intervened to prevent the paper from endorsing any candidate in the US elections for the first time in 36 years. 

The newspaper’s editorial page staff had written an endorsement of Kamala Harris for US president, but it was not published following a decision by Bezos, the Post’s owner, to change its policy on endorsements, according to an article in the paper... Sir Will Lewis, The Washington Post chief executive, outlined the reasoning behind the policy change in an opinion article in which he acknowledged that it could be read as “an abdication of responsibility” but added: “We don’t see it that way.” However, the newspaper’s guild said the decision raised concerns that “management interfered with the world of our members in editorial”... This will be the first time that the Post has not endorsed a president since 1988... Lewis, a former executive at News Corp and The Telegraph, was appointed by Bezos last year to try to arrest mounting losses and a decline in readership. People close to Lewis have said in the past that he is in regular contact with Bezos, and would not make big decisions without his input... This summer, Lewis angered Washington Post journalists after replacing the executive editor and other staff with his former colleagues from The Wall Street Journal and The Telegraph. He faced investigations from rival newspapers — as well as his own publication — into his role in a phone hacking scandal in the UK while he was a senior executive at Rupert Murdoch’s media empire. The turmoil at the Post came as Murdoch’s New York Post endorsed Trump for president, with a front-page headline declaring that the “choice was clear”... The Post’s reversal on endorsements follows a decision by Patrick Soon-Shiong, owner of the Los Angeles Times, to block an endorsement of Harris. Mariel Garza, the editorials editor, resigned in protest.

At a time of deep social polarisation, digital media-induced perversions of public debates, and the rise of populist politics, corporate takeover of media is a matter of big concern. Apart from the disturbing political consequences of such interferences, there are the motivations that drive such decisions. Sample this.

The Associated Press reported that hours after the Post announced its endorsement decision, Trump greeted executives from Blue Origin, the space company owned by Bezos that has a $3.4bn contract with Nasa to build a spacecraft to carry astronauts to the moon and back.

There would be a temptation to weigh in on the issue by arguing that the media should stay neutral and not endorse any candidate, and to that extent, Bezos is right. While that would be a logical (and correct) argument in many political contexts, it would be a disingenuous abstraction from the context in this case. 

In the US, it's common to see even people like academicians flaunting their political affiliations. In this milieu, newspapers have been ideologically aligned to political parties and, therefore, have historically endorsed Presidential candidates. The increasing flock of corporate media outlet owners like Jeff Bezos are using their ownership position to interfere in editorial decisions to favour their corporate and financial interests. 

And since the interests of corporate owners on a few issues (taxation, anti-trust, widening inequality etc.) are divergent from the wider public interest, such capture of the Fourth Estate has the potential to erode an already fraying social contract further.

The second exhibit concerns carbon emission reduction. Being very large consumers of energy through their data centres and AI algorithms, Amazon and the Big Tech companies are keen to burnish their green credentials by hitting their net zero targets quickly and cheaply. An alternative to actually hitting the target is to manipulate the target itself and the means of hitting the target. This would involve lowering the standards and providing flexibility in measurements. The way emission reduction is calculated offers an opportunity to indulge in greenwashing. 

Consider this. 

Social media group Meta, for instance, says it has already hit “net zero” emissions in its energy usage. But FT analysis of its 2023 sustainability report shows that its real-world CO₂ emissions from power consumption the prior year were 3.9mn tonnes, compared to the 273 net tonnes cited in the report… Companies including Amazon, Meta and Google have funded and lobbied the Greenhouse Gas Protocol, the carbon accounting oversight body, and financed research that helps back up their positions…A coalition that includes Amazon and Meta is pushing a plan that critics fear will allow companies to report emissions numbers that bear little relation to their real-world pollution and not fully compensate for those emissions. One person familiar with the reform discussions describes the proposal as “a way to rig the rules so the whole ecosystem can obfuscate what they are up to”… A rival proposal by Google, which would require companies to offset their emissions using power generated by more closely comparable means, has been criticised by the Amazon coalition and others for being expensive and too difficult…

Tech companies invest in renewable energy but they cannot fully control how polluting the power their data centres draw from their local grid is. So under current accounting rules, the power used by a data centre during the night in a coal and gas heavy region such as Virginia can be cancelled out by buying a certificate tied to solar energy produced during the day in a region with a cleaner grid, such as Nevada… Each time a wind, solar or hydroelectric facility generates a unit of clean power, its owner can issue an energy attribute certificate, typically known in the US as a renewable energy certificate, or REC. These can either come “bundled” into a contract for clean power, or can be bought individually from a generator or market intermediaries. Companies can purchase RECs “to buy-down their environmental impact”… 

But Matthew Brander, a professor at the University of Edinburgh, says the system is akin to buying the right from a fitter colleague to say you have cycled to work, even though you arrived by a car that runs on petrol. Other experts have raised concerns about how RECs are being used to offset real-world emissions. At present, the certificates must come from the same defined geographic region as the pollution they are offsetting, such as Europe and North America, but not the same grid and not at the same time. That means the clean energy that offsets the emissions could be generated in a different country, at a different time of day — or even in the past…

But both timing and location matter in terms of real-world emissions. For example, one potential buyer hooked up to a coal-dependent grid and another on a much cleaner grid could buy the same certificate to offset one megawatt hour of power use — even though the emissions stemming from that usage will differ in each grid. The certificates are also very cheap. The average forward price of a single US renewable energy certificate to be bought in the next calendar year has been under $5 since at least 2022, commodity trader STX Group estimates… Academics and experts at Princeton, Harvard and the Greenhouse Gas Management Institute have shown that buying certificates typically did not drive either a new supply of renewables or a fall in emissions… Google’s proposed solution is to only match energy consumption with clean energy and certificates from the grids where power is consumed, and to take the time of day of its electricity use into account. Using certificates from one area while operating in another could allow buyers to understate their reliance on fossil-based electricity without addressing the emissions for which they’re physically responsible.

In this context, it’s disturbing that Jeff Bezos’ $10 bn charitable group, Bezos Earth Fund is trying to influence the operations of the carbon credit market to allow Amazon to dilute its emission reduction obligations. Amazon is lobbying to not only continue recognising carbon credits purchased from a different geography and time, but also get a higher credit for those purchased from a dirtier developing country grid than from a cleaner developed country grid. 

The Bezos Earth Fund is among the largest funders of the Science Based Targets initiative, a globally-renowned body relied upon by groups such as Apple and H&M to set voluntary standards and strict limits on the use of carbon credits to offset emissions… The SBTi is also in the middle of a process of rethinking its approach to offsets, a decision that could prove crucial to Big Tech groups at a time when artificial intelligence is resulting in a leap in emissions caused by the greater use of data centres. Experts and campaigners have grown concerned about the potential of Amazon and the Bezos fund… to influence SBTi, which holds sway over whether many corporate groups can achieve a credible “net zero” label… a former SBTi staff member raised fears about perceived influence of the Bezos fund on climate standards in a July complaint to the UK charity commission. The fund has also financed the organisations that employ three SBTi board members…

Grant-making organisations with current or historic ties to big business, such as Bloomberg Philanthropies, the Ikea Foundation or the Rockefeller Foundation are the financial bedrock of the climate standard setting and campaigning space. Google and its philanthropic arm have also funded bodies in this space. But the battle over the future of the SBTi could prove crucial to corporate efforts to achieve climate goals. Some companies have become frustrated at SBTi’s restrictions on the use of credits to just 10 per cent of emissions… The Bezos fund is also a backer of the top standard setter in carbon accounting: the Greenhouse Gas Protocol, which is also in the process of reconsidering its approach to offsets… Amazon is also seen as promoting alternatives to the SBTi’s standards… Amazon last year also contributed to the creation of a market label, Abacus, to test the quality of carbon credits… Buying credits is typically much cheaper than cutting supply chain emissions, making them a tool of choice for some chief executives in the face of pressure to keep climate promises made to shareholders.

There’s a real danger of green-washing here, to fake net zero. 

Here’s the problem. The two examples are only the latest to show that wealth brings outsized influence in the political and rules-setting process, thereby eroding democracy and the social contract itself. It’s Jeff Bezos’ outsized wealth that gives him the power to exercise such influence on a terrain that goes far beyond his business or even industry. It’s not possible through any institutional restraint or safeguards short of outright prohibitions (against, say, corporate ownership of media outlets) to insulate from elite capture. 

But, even such measures are likely to be blunt given the staggering magnitude of wealth concentration and the availability of instruments to exercise influence (electoral funding, ownership of media outlets, philanthropic foundations etc.) that allow the likes of Bezos to manipulate the political process from behind the scenes (though the likes of Elon Musk no longer make have even these pretensions) to suit their interests. 

The final word on the issue should go to this quote from Justice Louis Brandeis (HT: Matt Stoller), who famously said, "We may have democracy, or we may have wealth concentrated in the hands of a few, but we cannot have both." Just as business concentration and competition cannot co-exist, democracy and wealth concentration too cannot go together!

R&D expenses, productivity, and growth in the age of Big Tech

Econ 101 informs us that R&D investments spur innovation and productivity growth, which in turn lower production costs and create newer products, boost consumption, drive further investments, job creation and economic growth. In short, R&D investments and innovation trigger a virtuous loop.

Gillian Tett points to an interesting paradox that questions this conventional wisdom that R&D investments invariably result in innovation, productivity growth, and expansion of economic output. Sample some numbers.

On the one hand, American R&D has risen in recent decades, from 2.2 per cent of GDP in the 1980s to 3.4 per cent in 2021. That reflects a doubling of private sector R&D to 2.5 per cent of GDP. Meanwhile, the proportion of the population involved in patent production nearly doubled in this period. But there is a big catch. Although “conventional economic models” imply that increases in R&D spending on this scale “should have led to accelerated economic growth”, this has not occurred. Michael Peters, a Yale economist, lays out the grim news: while labour productivity rose on average by 2.3 per cent between 1947 and 2005, between 2005 and 2018 it fell to 1.3 per cent. This cost America a putative $11tn of output, he calculates.

Micheal Peters in the IMF’s latest F&D magazine points to the US productivity growth being on a secular decline, that has become pronounced during the digital age since the turn of the millennium.

There is growing evidence that the US economy is not as dynamic as it used to be. A key aspect of business dynamism is new business formation. It is often measured by the entry rate, or the share of enterprises that started operating in a given year. The entry rate fell from 13 percent in 1980 to 8 percent in 2018, according to the US Census Bureau. In addition, US enterprises became substantially larger, with the average number of employees rising from 20 in 1980 to 24 by 2018. Older and bigger companies thus account for a much larger share of economic activity than they used to. These trends indicate significantly declining dynamism in the US economy over almost four decades…

First, the rise in corporate concentration has been shown to go hand in hand with expanding market power. The average markup by publicly traded US companies surged from about 20 percent in 1980 to 60 percent today. Large incumbent businesses thus seem to be shielded more and more from competition, allowing them to jack up prices and widen profit margins. A second line of research shows the flip side of rising corporate market power: the weakening of workers’ bargaining position. Since 1980, labor’s share of the US economy has fallen by about 5 percentage points. The plunge was faster in industries that experienced more concentration… Third, there has been a secular decline in business-to-business reallocation since the late 1980s, as shown in a series of papers by John Haltiwanger and other researchers. This suggests that the process of workers moving from declining to expanding businesses is not as fluid and dynamic as it once was.  

Peters also explores the possible causes for this productivity decline.

The ten biggest tech companies by capitalisation in Nasdaq spent a total of $222 bn in R&D in 2022, of which Amazon alone spent $73.2 bn. 

Remarkably, these kinds of R&D expenditures have not been accompanied by job creation. As an illustration, even as Amazon, Microsoft, and Alphabet spent a record $139.3 bn on R&D, the three laid off 40,000 workers at the beginning of 2023. 

Germán Gutiérrez and Thomas Philippon measured the evolution of dominant firms in the US since 1960 (top 20 firms by global sales and top 4 firms in each 3-digit industry) and globally since 1990 (top 100 firms by global sales in a given year and the top 20 firms in 25 industries), and found that their contribution to aggregate productivity growth has fallen by more than one-third since 2000. 

In another paper Philippon writes

I estimate that markups in the United States have increased by about 12% since 2000. Such an increase in markups implies that wages and consumption are at least 10% below their potential… increasing markups in the United States have lowered labor income by about $1.44 trillion… the stars of the digital economy—Amazon, Google, Facebook, Apple, and Microsoft (GAFAMs for short)—are not as “special” as one might think… Along all qualitative dimensions, including profit margins and productivity, the stars of today are… they are smaller than market leaders of the past, and they matter less for overall GDP growth than General Motors, IBM, or AT&T did at their peak… rising market concentration since the early 2000s has produced market inefficiencies. Dominant firms have succeeded in erecting barriers to entry, which has resulted in lower investment, higher prices, and slower productivity growth.

Given the aforementioned facts, here are some observations:

1. Joel Mokyr has made the distinction of useful knowledge as one that promotes material progress. It consists of propositional knowledge (“what”) and prescriptive knowledge (“how”), with the former consisting of people who know things (savants) and the latter of people who make things (fabricants). He uses this distinction to explain why the Industrial Revolution originated in England and not in continental Europe. On the same lines, it may be useful to make the distinction within R&D investments and categorise some as useful R&D. Ditto with innovation. 

Are R&D expenditure and innovation increasing the economic output? Is it creating jobs? Or is it being used to create moats around the markets served by the big technology companies? Given the nature of the digital technology markets, with their network effects and the advantages conferred by access to large data, are the large R&D expenditures conferring an unbridgeable advantage to the large incumbents?

There’s a case for distinguishing between defensive and productive R&D, with the latter aimed at protecting the turf/market. More on this latter in the post.

2. In the last 2-3 years, there has been a surge in AI-related R&D investments. As the graph above shows, each of the US Big Tech firms - Amazon, Alphabet, Apple, Microsoft, Meta, Nvidia etc. - have been making AI-related investments in the tens of billions of dollars every year. So much so that the entire US equity market is now riding almost entirely on the AI investment boom. However several questions are being raised about the likely value generation from these investments. The vast majority of commentators view this boom as being driven by FOMO and disconnected from any value creation. 

Is it then the case that the returns from R&D investments, or their productivity, have declined? Or is it that the technology will take time to mature and start to show its benefits? Or more generally, are large firms being inefficient in their allocation of R&D expenditures? Or a combination of all three?

Ufuk Akcigit writes in the same issue of the IMF’s F&D magazine:

In earlier research, Harvard’s William Kerr and I found that small businesses are more innovative relative to their size, suggesting they use R&D resources more efficiently. As companies grow and dominate their markets, they often shift their focus from innovation to protecting their market position. In a more recent study, Salome Baslandze, Francesca Lotti, and I showed using Italian data that larger enterprises tend to innovate less and instead engage in activities that limit competition. One such activity is hiring local politicians. As businesses climb the ranks among the largest 20 players in their industry, they hire more politicians, while their patent production declines. This highlights what we call a leadership paradox, where leading companies plow resources into maintaining dominance rather than fostering innovation… As dominant players prioritize strategic moves over genuine innovation, the economy as a whole is almost certainly missing out on potential growth opportunities.

3. Most importantly, it’ll be useful to examine where these R&D investments are going. These are astronomical sums, larger than the entire R&D expenditures of several large countries combined, including those like India. Are they actually going into research and development? Or is it some accounting trick to benefit from tax rules?

The primary reason to doubt these numbers is a palpable absence of their signatures on the products and services being delivered. Has e-commerce, social media engagement experience, and internet search got so much better in say, the last five years, to justify even a small part of these expenditures? Does building and maintaining data centres (whose buildings and operations are outsourced to PE funds and their boring infrastructure contractors) demand such levels of R&D expenditures?

It’s hard to imagine or rationalise that e-commerce, social media, and internet search (and advertising) can consume anything even remotely close to the amounts being spent on R&D. One associates with R&D in digital technologies to some equipment and software costs, and considerable manpower costs. Even at the higher end of such expenditures, $73.2 bn (and similarly high numbers in earlier years) appears mind-boggling. So where is this likely coming from (or what’s it going into)? 

A comment on a discussion thread in Hacker News nails it:

This $73.2B figure was pulled directly from their 2022 10-K filing under an operating expenses line item labeled technology and content, which includes R&D and then some. The devil in the details is buried in a footnote on p. 26:

Technology and content costs include payroll and related expenses for employees involved in the research and development of new and existing products and services, development, design, and maintenance of our stores, curation and display of products and services made available in our online stores, and infrastructure costs. Infrastructure costs include servers, networking equipment, and data center related depreciation and amortization, rent, utilities, and other expenses necessary to support AWS and other Amazon businesses.

At face value, to handwave this figure as just R&D and purport that it's directly comparable to other publicly-traded companies who report disaggregated research and development strikes me as somewhere between shotgun analysis and hoodwinking.

Read this Forbes article which says the same things, and see this. 

Here is the table indicated above.

Amazon spends large amounts on its AWS data centres. All those expenditures are lumped together into the basket of technology and content. Even its prime Video content is classified under R&D. In short, it appears that Amazon lumps all its personnel, software and hardware costs under R&D. 

Is all this disingenuous accounting motivated by tax minimisation strategies adopted by these companies? Is Amazon benefiting from the greater tax benefits accorded to R&D expenses? Akcigit again

The Federal Reserve’s Sina Ates and I examined market competition trends in the US over the past several decades. Since the early 1980s, there’s been a noticeable increase in market concentration and a decline in business dynamism… This period aligns with the 1981 introduction of the R&D tax credit, a component of President Ronald Reagan’s sweeping Economic Recovery Tax Act. The credit was intended to encourage businesses to invest in research and development. Minnesota was the first state to adopt a similar state-level R&D tax credit, in 1982, and many other states followed, expecting to promote innovation and economic growth. Which companies are most likely to take advantage of the R&D tax credit? Our research with Goldschlag shows that large businesses are much more likely to benefit than smaller ones. The policy—perhaps unintentionally—favors big companies, encouraging them to dominate in R&D spending… Our research provides direct evidence that businesses actively claiming R&D tax credits are more likely to engage in stifling hiring practices. These enterprises often offer higher salaries to inventors, and the inventors become less innovative after joining.

It’s therefore clear that all the so-called R&D expenditures are mostly about the general operational and capital expenses! So after all, real R&D expenditures might not have gone up as much as we imagine, which also explains the apparent paradox of low productivity growth. 

4. In this context, it’s also worth making the distinction between the R&D expenditures of firms in digital technology and traditional economy sectors. Technology firms like those above are constantly iterating and improving their algorithms and user interfaces (and the logistics operations in the case of Amazon and advertising engine in the case of Google) as part of their regular operations. The nature of digital technologies (for example, with digital trails that serve as fuel for data analytics and AI-algorithms) ensures such iteration and refinement. 

There’s an indistinguishable line between the R&D expenditures of technology firms and their operational expenses. It’s different from the distinct role of R&D in the traditional economy where it’s about inventing new technologies, products, and services. The likes of Amazon, Meta, Alphabet, Apple etc., have not brought out any new product, but are only marginally refining their algorithms and at best moving into emerging adjacent markets (where too they have a head start due to the nature of their platform business models). They are, for all practical purposes, one trick ponies.

This raises the need for greater clarity in the accounting of R&D expenses of digital technology firms. 

5. One of the surprisingly less discussed but widely known features of technology markets today is the stifling grip exercised by the big firms. This hold covers hiring and retaining critical personnel, patent acquisition, protection of their own patents, and even protecting themselves from being gobbled up. Big Tech (and large incumbents generally) firms employ armies of lawyers to engage in the likes of patent trolling, squatting, and hoarding to copy, steal and intimidate startups. 

More from Ufuk Akcigit

Over the past two decades, there has been a notable reallocation of innovative resources toward large, established companies, Goldschlag and I documented in 2022. At the beginning of this century, roughly 48 percent of American inventors worked for these big incumbent companies—those that are more than 20 years old and employ more than 1,000 workers. By 2015, that figure had surged to 58 percent, marking a significant shift in where the nation’s innovative talent is concentrated… research shows a concerning trend: inventors that move to large firms become less innovative compared with inventors that move to young firms.

A specific practice identified in our research is innovation-stifling hiring. This occurs when big, established enterprises hire key employees from younger competitors, often by offering higher salaries. However, instead of using these new employees to drive innovation, the big businesses may place them in roles that do not fully leverage their skills. As a result, these individuals become less innovative, and the overall innovative capacity of the economy suffers. After 2000, there was a notable increase in the wage premium offered by established companies, compared with salaries paid by younger businesses. The pay differential widened by 20 percent, prompting many innovators to switch jobs and join larger, well-established companies. However, these inventors’ innovativeness dropped by 6 percent compared with that of their peers who joined younger employers… By hiring away top talent from rivals, these companies not only weaken their competitors but also prevent these individuals from contributing to potentially disruptive innovations elsewhere. This strategy may benefit the hiring business in the short term, but it poses a long-term risk to the economy’s overall innovation and growth. 

Given all these, startups stand little or no chance of competing with large incumbents. It’s an open secret that the Big Tech firms unleash their lawyers and financiers to intimidate and squeeze any startup trying to compete with them. I don’t know whether the libertarians and tech evangelists who wax eloquent about digital utopia and lose no opportunity to argue for deregulation and getting government out of the way are being ignorant or disingenuous or plain dishonest when they overlook the egregious manner in which Big Tech firms unleash their lawyers and corporate brokers to intimidate startups. 

Why are none of the economists in the big Economics Departments silent about one of the commonest market practices which is also the biggest threat to competition in digital markets?

I have written here (and here) about how Amazon uses anti-competitive practices to copy and forcibly buyout emerging startup competitors and then kill off their technologies or adopt them itself. 

The final word to Akcigit

The evidence suggests that while the US is investing more in R&D, the concentration of resources among large businesses has led to diminishing returns in terms of productivity growth. This outcome challenges the assumption that simply expanding R&D spending will automatically lead to economic growth. Instead, it highlights the need for a more nuanced approach to industrial policy—one that not only incentivizes R&D but also encourages the effective reallocation of resources. To foster a more dynamic and innovative economy, the US needs to design policies that support not just large incumbents but also smaller businesses and start-ups, which often have a greater capacity for disruptive innovation. This could include targeted tax credits for small businesses, grants for early-stage innovation, and policies that encourage competition and reduce barriers to entry for new players. 

Weekend reading links

1. On mangoes

India is the world’s biggest mango producer, with volumes greater than the next nine growers combined, according to Tridge, a data firm. Yet its share of the global export market by value is a meagre 7%. Mexico, which produces a tenth as much as India, accounts for a quarter. The reasons are many, writes Sopan Joshi in “Mangifera Indica”, a new book about mangoes. Chief among them are the delicate nature of the fruit, poor growing practices in India and strict standards in Western markets.

2. Interesting fact about the state where US companies are getting incorporated, Delaware stands out with over 70%.

3. Data does not bear out signatures of deglobalisation.

4. Some facts about Big Tech 

Since early 2019 the combined worth of the tech giants (Microsoft, Google, Amazon, Meta, and Apple) has more than tripled, to $11.8trn. Add in Nvidia, the only other American firm valued in the trillions, thanks to its pivotal role in generative artificial intelligence (ai), and they fetch more than one and a half times the value of America’s next 25 firms put together. That includes big oil (ExxonMobil and Chevron), big pharma (Eli Lilly and Johnson & Johnson), big finance (Berkshire Hathaway and JPMorgan Chase) and big retail (Walmart). In other words, while the tech illuminati have grown bigger and more powerful, the rest lag ever further behind...

Since 2019 the five tech giants and Nvidia have doubled their capital expenditures, to $169bn last year. Tot up the 25 next firms’ capex and it was just $135bn—up only 35%. As for brain power, over the same period, the big six added 1m jobs, doubling their headcount. No one can accuse them of resting on their laurels. They have invested in ai startups, ploughed fortunes into building large language models and, in Meta’s case, created open-source offerings that almost anyone can use. This year they are doubling down on their ai spending if only to protect their flanks.

5. Striking facts about the importance of immigrants to the US economy.

Immigrants are 14% of the population in America, 16% of inventors and directly produce over 23% of innovation, measured by patents, patent citations and the economic value of those patents, the authors estimate. Taking into account how they make their native-born collaborators more productive, they are responsible for a staggering 36% of total innovation.

6. Law of unintended consequences strikes at the Chinese government ban of tutoring centres for the gaokao entrance examinations.

Researchers at Peking University... analysed surveys of household spending before and after the tutoring ban and found that low- and middle-income families were, on average, spending less on after-school education. The richest families, though, were spending more. Tutors were still active. But because they were acting illegally, they charged more, pricing most people out.

7. Very good summary of the performance of State Bank of India under Mr Dinesh Khara who stepped down as CMD after four years. These are very impressive numbers

Between October 7, 2020 and last week, SBI stock has delivered a 328.36 per cent return to investors, compared to Bank Nifty’s 122 per cent return and Bankex’s 122.9 per cent. During this time, the Nifty returned 111.4 per cent, and the Sensex 103.3 per cent. The largest private bank, HDFC Bank Ltd, saw its stock rise 39.9 per cent, while ICICI Bank Ltd rose 214.6 per cent in this period. For SBI, the return on assets (RoA) moved from 0.43 in September 2020 to 1.10 in June 2024. During this period, return on equity (RoE) increased from 8.94 to 20.98, and earnings per share (EPS) from 19.59 to 76.56... While the bank’s assets have grown over the past three-and-a-half years at a compounded annual growth rate of 10.56 per cent, from Rs 43.58 trillion to Rs 61.91 trillion, its gross non-performing assets (NPAs), as a percentage of total assets, have more than halved, from 4.77 per cent to 2.21 per cent. After provisioning, net NPAs have decreased from 1.23 per cent to 0.57 per cent. Meanwhile, the net interest margin (NIM) — loosely the difference between what it spends on deposits and earns on loans — has risen marginally from 3.31 per cent to 3.35 per cent.

Who said public sector institutions must be inferior also rans compared to their private sector counterparts?

8. Since the pandemic, apart from the Bank of Japan and the Swiss Central Bank, the RBI has undertaken the least number of interest rate changes. 

It's also one of the very few that have not yet moved on reducing rates. 

9. Ruchir Sharma talks about a return of emerging market economies in the coming years. For a start, corporate earnings are growing faster in EMs than elsewhere.

This means that we could see a reversion of the trend since 2010.

10. Some numbers to put in perspective why the US equity markets stand alone globally. 

At the beginning of the 20th century, the US accounted for about 15 per cent of world market capitalisation, second only to the UK, which was at 24 per cent. By 1910, the US had crossed the UK to become the largest equity market in the world. It has since retained this title, unchallenged except for a brief period in the late 1980s when Japan held the top position. Japan peaked in 1989 at 40 per cent of world market capitalisation, while the US was second at 29 per cent. Today, the US remains unchallenged, accounting for more than 62 per cent of world market capitalisation. The next largest market is Japan at 6 per cent, followed by the UK at 3.7 per cent, and China at 2.8 per cent (all based on the FT World Index and free float adjusted). Just 12 markets, including India, account for 90 per cent of world equity market capitalisation. Looking at the longest available data series on equity market performance (1900-2023), spanning 124 years, we see that the US has delivered the best real returns, with an annualised rate of 6.5 per cent. The only market even close is Australia at 6.45 per cent in dollar terms, though there is no comparison in terms of size or absolute market capitalisation created. The UK has delivered 4.9 per cent real return, while Germany and France lag with return profiles of only 3.3 per cent and 3.16 per cent, respectively. Japan delivered 4.2 per cent (all in dollar terms). Compared to the 6.5 per cent real return of the US, the world ex-US, delivered 4.3 per cent, a gap of 2.2 percentage points compounded over 124 years. This leads to huge differences in terminal value. The US has undoubtedly been the right place to invest. If an investor had been exclusively invested in the US for the entire 124 years, their return would have turned one dollar into $2,443 in real terms. The same dollar invested in non-US markets would have grown to only $191, not even one-tenth of the US investment portfolio.

And about India

Over this 30-year period (ending July 30, 2024), MSCI India has delivered a nominal annualised return of 8.65 per cent in dollar terms, compared to 5.3 per cent for MSCI.

11. FT has a nice report from Panyu, a suburb in the southern Chinese city of Guangzhou, which is nicknamed the "Shein village" for its centrality in the retailer's business. The $66 bn valued firm, due for listing at the London Stock Exchange, has shaken up fast fashion with its $5 dresses and $2 T-shirts. 

The article captures the reasons for Shein's competitive advantage.

But going to the heartland of Shein’s supply chain, it was clear that its low prices are in spite of, not because of labour costs, which have been rising in China as the working-age population shrinks and young migrant workers shun factory jobs for the lower-paid service sector. Factory workers that source to Shein typically get paid between Rmb7,000 ($982) and Rmb12,000 monthly, depending on how many clothes they finish. By contrast, the average wage for other blue-collar workers in the area is between Rmb5,500 and Rmb6,500. Part of the reason the clothes are cheap is, well, because they are cheap. One factory manager held up a baggy dress — probably destined for the US or UK — and joked that she would never sell such low-quality clothes to a more discerning Chinese clientele. She says she uses cheaper fabrics for Shein orders than for Alibaba’s Taobao, because the domestic platform gives more money to the factories to cover their costs. 

Shein has also cut out expensive middlemen by shipping goods directly from warehouses in China to shoppers in the west — a model that has the added benefit of the great majority of its packages bypassing import duties. Panyu highlights the attraction of Chinese manufacturing. Like other manufacturing hubs specialising in anything from socks to sex toys to steel pans, it has the entire supply chain concentrated in one district. That means factories can within half an hour place an order, take delivery of fabric or get an engineer to fix sewing machines with components made nearby... China’s migrant worker population also brings it an edge. While in Vietnam and Bangladesh workers tend to return home to their families at night, the labourers in Panyu sleep in nearby dormitories, cutting down commuting time and meaning they can work longer hours if a large order arrives.

12.  Good graphic that shows how the markets over-react to economic news.

Robert Armstorng writes in Unhedged in FT.

Here is the futures market’s expectations for what the federal funds rate will be in December 2024, as well as the Fed’s projections from its quarterly summary of economic projections (the last SEP was released in early June)... One cannot help but notice the pattern of overreaction and correction on the market side. It’s like a car on an icy road. There is a whole sub-industry — Unhedged is part of it — that spends its time arguing about why the Fed is too loose or too tight. But in retrospect we probably overstate the importance of the current and expected level of rates. What matters is keeping expectations anchored on the one hand, and avoiding an unnecessary recession on the other.

13. BCG and Angola

BCG has admitted it paid millions of dollars in bribes to win business in Angola, and agreed to give up more than $14mn in profits from contracts it won with the country’s economy ministry and central bank. The consulting firm sent money to offshore accounts controlled by middlemen connected to Angolan officials and members of the ruling political party, according to a US Department of Justice investigation made public on Wednesday. The bribes were paid by BCG through its office in Lisbon, Portugal, between about 2011 and 2017, the DoJ said... BCG agreed to pay an agent with ties to Angolan officials between 20 per cent and 35 per cent of the value of the contracts it won, routing the money through three different offshore entities, the DoJ said... The period of the bribes coincided with the end of the rule of the late José Eduardo dos Santos, who stepped down in 2017 after 38 years in power... In total, BCG won 11 contracts with the Angolan ministry of economy and one with the National Bank of Angola over the years in question, bringing in $22.5mn in revenue. The firm will return the $14.4mn in profits that the contracts generated.

14. But KPMG and UK validates the adage that the more things change, more they remain the same

KPMG has won a UK government contract worth up to £223mn to train civil servants, the second-largest public sector contract awarded to the Big Four firm and agreed before the Treasury set out plans to drastically reduce Whitehall’s reliance on external consultants last month. Under the 14-month deal with the Cabinet Office, which commenced this month, the consulting firm will manage learning and development services across Whitehall, including overseeing courses on policymaking, communications and career development. The maximum value of the contract represents close to 8 per cent of KPMG’s annual UK revenues, making it the second-biggest public sector contract awarded to the firm, according to data provider Tussell. The most valuable piece of public sector work awarded to KPMG was a separate learning and development deal with the Cabinet Office worth £237mn, Tussell said. That four-year contract, which expires in October, involves the firm overseeing technical training for civil servants, such as professional qualifications. The lucrative contracts demonstrate a return to positive relations between the government and KPMG. The Big Four firm stopped bidding for UK government contracts in 2021 following a threat by the Cabinet Office to ban it from winning public sector work after its involvement in a series of scandals. It resumed bidding for public sector contracts in 2022. They also come as the Labour government has committed to halving Whitehall spending on consulting firms during this parliament, with chancellor Rachel Reeves last month ordering departments to stop all “non-essential spending” on external consultants. A government spokesperson said the KPMG contract was agreed before July’s general election. The Conservative party also pledged to halve Whitehall spending on external advisory firms in its election manifesto. The Treasury estimated in July that reducing the government’s reliance on advisory groups would save £550mn in the 2024-25 financial year and a further £680mn in 2025-26, when the policy to halve total spending on consultants came into force. The savings would, in part, help fund significant public sector pay rises, the chancellor said.

15. Very good article on how Nvidia is working to protect its domination of the high-end chips design market. 

The key issue is when the main focus in AI moves from training the large “foundation” models that underpin modern AI systems, to putting those models into widespread use in the applications used by large numbers of consumers and businesses. With their ability to handle multiple computations in parallel, Nvidia’s powerful graphical processing units, or GPUs, have maintained their dominance of data-intensive AI training. By contrast, running queries against these AI models — known as inference — is a less demanding activity that could provide an opening for makers of less powerful — and cheaper — chips... Nvidia’s lead in this newer market already looks formidable. Announcing its latest earnings on Thursday, it said more than 40 per cent of its data centre sales over the past 12 months were already tied to inference, accounting for more than $33bn in revenue... But how the inference market will develop from here is uncertain. Two questions will determine the outcome: whether the AI business continues to be dominated by a race to build ever larger AI models, and where most of the inference will take place. Nvidia’s fortunes have been heavily tied to the race for scale... Yet it is not clear whether ever-larger models will continue to dominate the market, or whether these will eventually hit a point of diminishing returns. At the same time, smaller models that promise many of the same benefits, as well as less capable models designed for narrower tasks, are already coming into vogue. 

Meta, for instance, recently claimed that its new Llama 3.1 could match the performance of the advanced models such as OpenAI’s GPT-4, despite being far smaller. Improved training techniques, often relying on larger amounts of high-quality data, have helped. Once trained, the biggest models can also be “distilled” in smaller versions. Such developments promise to bring more of the work of AI inference to smaller, or “edge”, data centres, and on to smartphones and PCs... The range of competitors with an eye on this nascent market has been growing rapidly... The data centre market, meanwhile, has attracted a wide array of would-be competitors, from start-ups like Cerebras and Groq to tech giants like Meta and Amazon, which have developed their own inference chips. It is inevitable that Nvidia will lose market share as AI inference moves to devices where it does not yet have a presence, and to the data centres of cloud companies that favour in-house chip designs. But to defend its turf, it is leaning heavily on the software strategy that has long acted as a moat around its hardware, with tools that make it easier for developers to put its chips to use.

16. Finally, Japanese startup scene is finally waking up after long drawn persistent efforts by the Government.  

The ambitions are charged with the faith that start-ups can drive GDP growth and productivity, rescue the country from a long-term innovative tailspin and channel its talent in the right — or at least less wrong — direction. It has a belated, even desperate feel to it, but start-ups now seem to be Japan’s core industrial policy. The extent of both central and local government backing is striking. In addition to the many subsidies now on offer, state-backed entities like the Japan External Trade Organization have been drafted into the effort by providing acceleration programmes and other services. The government-backed Japan Investment Corporation has invested close to $1bn into 32 private venture capital funds. Under heavy government pressure, Japan’s three biggest banks have recently begun offering start-ups loans backed against current and future cash flow, breaking their long, entrepreneurialism-crushing habit of only lending against hard collateral such as the property of a would-be start-up founder. By many metrics, all this is working. In 2013, said the Ministry of Economy, Trade and Industry in a recent paper, the total investment into start-ups in Japan was a minuscule $600mn; a decade later, that had risen to over $6bn. Between 2014 and 2023, the number of university start-ups more than doubled to 4,288, with METI research showing that roughly half of university students would prefer to start their careers at one.