Entrepreneurship and tech

Artificial intelligence. A global project

In a rapidly globalizing market economy, it was possible to develop artificial intelligence. Chance played a decisive role in this. However, governments and states are once again ready to bring it back under their control.


At the end of 2022, the publication of ChatGTP triggered a veritable gold rush in the world of artificial intelligence (AI). Visions of the future based on it range from cloudy utopias to dystopian scenarios, familiar from science fiction blockbusters. Yet AI is hardly a new technology; on the contrary, it has been with us since the 1950s. Back in the 1970s, scientists taught a computer to recognize pictures of cats. When the reigning world chess champion Garry Kasparov lost to the computer Deep Blue in 1996, AI experienced a renewed upswing, which quickly subsided.

With the constant collection of data from companies and citizens on the World Wide Web, the challenge began for Internet companies to analyze this data and turn it into profit. The areas of application are diverse, from self-driving cars to banking and online trading. While public perception is dominated by fears about jobs, politicians and industrialists are more concerned about maintaining the supply chains that underpin AI. After all, the constant supply of powerful hardware and data is the linchpin of current geopolitical tensions between world powers.

The uncomfortable truth, however, is that no country in the world can master this technology alone. Each country contributes only a part to the global AI industry. And the boundaries between consumers and producers are beginning to blur as the former generate the data that trains the AI models.

The development of AI is a story full of happy coincidences and coincidences that no one could have foreseen. Certainly, no civil servant could have planned it on the drawing board. AI is thus truly a product of global liberalized markets. As Milton Friedman, one of the most radical advocates of free markets, said: “The great merit of capitalism lies not in the accumulation of wealth, but in the multitude of opportunities it gives people to expand, develop and improve their capabilities.”

The production cycle of artificial intelligence: complex and fascinating

Artificial intelligence not only goes beyond the previous limits of human capabilities, it also turns the concept of traditional production chains on its head. Producers receive feedback from their customers via a price signal – if they are too expensive, their own products are left on the shelf. The production steps are strung together link by link until the end product finally reaches the consumer, where it is consumed or disposed of at the end of its useful life. Whether the sandwich in the stomach or the television at the scrap yard.

The AI production chain is more like a cycle or a spiral. It consists of feedback loops of mutually reinforcing interactions. The OECD describes the production of AI models as a broad bundle of mutually complementary inputs of a material and, above all, immaterial nature. It is based on three pillars: know-how, data and hardware.

Production cycle of artificial intelligence technologies. Inputs are Skills & Software, Data and computing Power fuelling AI systems to create outputs in the form of generative content, predictions/ decisions/ analytics. But also there is a feedback loop of data output flows back to the inputs which are used to train the model and hence creates another feedback by improving the performance of the output

AI production cycle


The process begins with IT experts who program an AI model to translate an unstructured collection of ones and zeros into a meaningful output. They then feed huge amounts of data into this model, from which the AI is supposed to learn what is expected as output. The inputs range from digitized texts and images to videos and music. Vehicles, washing machines and almost all other machines have also been collecting data for a long time. Sensors have long been scanning and measuring our waste for patterns in order to make automated waste separation more efficient. Other AI models help diagnose cancer, assess credit applications or learn animal languages. All based on ingested data.

Availability of data opens up opportunities for AI

I remember researching the anti-globalization protests in Seattle in 1999, when thousands flocked to the US West Coast to protest against globalized world trade. The situation escalated. However, the only evidence that remains of these riots are images by professional photographers, which are expensive to license. With this sparse selection of material, an AI could hardly generate meaningful images. Today, things are different. Smartphones have become our constant companions.

Bubble chart of smartphone units sold worldwide by regions in 2016 and unit price. Emerging asia 233 Million between 150 to 200 dollars. Middle East & Africa 177 Million units sold of 200-250 dollars. Central & Eastern Europe 85 Million units sold roughly 200 dollars. Latin America 116 Million units sold of about 300-350 Dollars. China 454 Million Units of 300 to 350 units. North America 201 Million units 400 to 450 dollars. Western Europe 125 Million dollars of about 400-600 dollars. Developed Asia 69 Million sold units of roughly 650 dollars on average

Sold smartphone units 2016


Thanks to smartphones, the same riots are now documented with images and sound from a wide variety of angles and uploaded to the internet in real time. Whether it’s vacation photos of the Eiffel Tower, recordings of traffic checks by potentially corrupt police officers in Russia or Africa or informal Facebook job exchanges in Latin America. Everything is documented, even in the remotest corners of the world. Even lower income groups are constantly recording their impressions in writing, images and sound. Mobile internet devices with GPS, integrated cameras and other apps are now even available to lower income groups. Apps collect movement data and usage behavior in the background. The volume of data is growing exponentially.

Exponential increase of data volulme between 2010 and 2025 from hardly above 1 Zetabyte in 2010 up to 180 Zetabytes in 2025.

Exponential increase of data volume


The data collected in this way feeds the machinery of countless AIs that scour the internet around the clock. With every bit uploaded, they learn more and become more precise. This blurs the boundary between the producers of the data and those who consume it in the digital sphere. The cheap mass production of end devices that make this possible takes place in China and Asia.

Quantity is a quality in itself

All of these devices contain chips of varying quality. The rise of China as the world’s extended workbench is also linked to Taiwan’s development into a chip superpower. Despite generations of bitter geopolitical rivalry, both countries benefit economically from each other. China has climbed the technological ladder step by step until it was able to develop increasingly complex electronic devices on its own. Taiwan, the small island on the east coast of the Middle Kingdom, supplied the brains in the form of microchips.

Export destinations of Taiwanese chip production. Despite the tensions in recent years, China is still the undisputed leader with close to 60 percent and roughly 35 into the other Asean countries

Export destinations of Taiwanese chip production. Despite the tensions in recent years, China is still the undisputed leader


Taiwan’s government recognized the importance of microelectronics as early as the 1970s. At that time still a poverty-stricken country itself, the foundations of its current market leadership were laid in the 1980s with the Hsinchu Science Park, conceptually based on Silicon Valley’s ecosystem of industry and research. Over time, the Taiwanese learned to produce faster and more precisely than their competitors. Foreign companies settled there, attracted by a well-educated and disciplined workforce. The partly state-owned Taiwan Semiconductor Manufacturing Company (TSMC) became a pioneer in the field and today produces over 13 million semiconductors a year. Its customers include all the big names in the industry, from NVIDIA and Microsoft to Apple. Taiwan is now home to over 60 percent of the world’s semiconductor production and over 90 percent of the most advanced chips that form the basis for AI computing.

Advanced chip technology, market shares (source: statista, 2023)

Advanced chip technology, market shares (source: statista, 2023)


Hardware for chip production

It is not only the mass, but also the ever smaller chips that are presenting manufacturers with new challenges. They have now shrunk to just a few nanometers (equivalent to a billionth of a meter). Only a few companies are able to manufacture machines that can work in these microscopic spheres. Surprisingly, these photolithographic machines have their origins in companies specializing in camera and photographic technology, particularly the Japanese Nikon and Canon.

The path how ASML in Netherlands became a monopoly in the chips procduction sector (Source: the-waves.org, 2022)

The path how ASML in Netherlands became a monopoly in the chips procduction sector (Source: the-waves.org, 2022)


On closer inspection, this even makes sense. Their experience in the manufacture of high-precision optical devices and their ability to manipulate light with extraordinary precision are crucial in the creation of the finest structures in semiconductor production. ASML Holding, founded in 1984 in Veldhoven, the Netherlands, has now achieved a monopoly position in this field. By using extremely ultraviolet radiation, its machines achieve greater precision than all its competitors. In turn, ASML relies on numerous suppliers – such as the long-established company Zeiss, founded in 1846 as a workshop for microscopes, with which ASML has had a strategic partnership since 1983.

The world of computer chips

Despite significant steps in research and development, Taiwan remains dependent on Western know-how for the design of its semiconductors. The US company NVIDIA in particular has made a name for itself in this area. Today, NVIDIA controls almost 95 percent of the market for highly specialized chips for AI applications. Within a year, its share price shot up by 450 percent, making NVIDIA the third most valuable company behind Microsoft and Apple overnight.

Originally, the company designed chips for excessive PC gamers who demanded ever higher levels of realism from their games. It was only over time that their particular suitability for mining cryptocurrencies and calculating AI algorithms became apparent. In the course of its optimizations, NVIDIA increased the performance of its high-end chips a thousandfold. Such a chip costs between 25,000 and 40,000 US dollars. Even during NVIDIA’s success outside the computer game scene, there were no plans to participate in AI.

From cryptos to artificial intelligence

This actually began in 2009 with the publication of the white paper “Bitcoin: A Peer-to-Peer Electronic Cash System” by Satoshi Nakamoto, who remains anonymous to this day. With the infamous Bitcoin, miners have to solve increasingly complex computing tasks to mine new coins (tokens) – a so-called proof of work. With increasing complexity, at some point standard computers were no longer able to do this.

Instead, server farms were set up in places where the often very conspiratorial, state-sceptical miners felt safe and energy was available at low cost – such as the deserted and natural gas-rich Kazakhstan. The competition for the next mathematical solution forced them to expand their capacities step by step. This resulted in a huge infrastructure of computing power, so-called GPU farms consisting of thousands of interconnected high-performance chips – predestined for the calculation of AI algorithms. Between 2012 and 2018, computing capacity increased three hundred thousand-fold, according to OpenAI, the development company behind Chat GTP. This corresponds to a doubling rate every three to four months.

The multiplication of computing power since the invention of the computer (Source: OpenAI, 2018)

The multiplication of computing power since the invention of the computer (Source: OpenAI, 2018)


Times have changed. Crypto miners increasingly relied on optimized ASICS technologies for their Bitcoin computing operations, although these are not compatible with GPU farms. Ethereum, a major competitor to Bitcoin, switched from creating money through a proof of work to a proof of stake process. Finally, in 2021, China, the largest farming location in the world, also took action against energy-intensive activities. Suddenly, huge capacities of electricity, cooling and computing power went unused. In the interplay of supply and demand, this meant a low-cost infrastructure as a playground for AI developers.

Shares of crypto mining per country. The Chinese leadership's crackdown on cryptocurrencies finally freed up large capacities (source: University of Cambridge, 2021)

Shares of crypto mining per country. The Chinese leadership’s crackdown on cryptocurrencies finally freed up large capacities (source: University of Cambridge, 2021)


AI talents: Architects of the virtual world

With the existing technologies, available data and a broadly rolled-out infrastructure, all that was needed were specialists such as programmers and data scientists to set up the corresponding AI models. They bring structure to the mountains of data and build complex algorithms to ultimately create meaningful products from ones and zeros. However, these talents are still few and far between and their training is lagging behind the opportunities that are opening up.

They are concentrated in just a few places around the world, such as San Francisco, but increasingly also in China’s capital Beijing. Large institutions and companies such as IBM, Microsoft, Google or the Indian Tata Consultancy are also able to secure most of the talent thanks to their high paying power. The USA and China are also by far the undisputed leaders in AI research. The United Kingdom, Germany and France are lagging behind. Surprisingly, Iran, not normally known as an innovative high-tech country, also holds a solid place in the midfield. With Azad University, the Islamic Republic is even home to a particularly outstanding research institution on the subject.

AI talents worldwide. (Source: China AI Development Report 2018)

AI talents worldwide. (Source: China AI Development Report 2018)


In the corporate world, IBM, Microsoft and Samsung are the leaders in terms of their research output in the form of scientific publications and patents. With 6,776 publications, the US Department of Defense even surpasses IBM with 5,105 papers. Other leading public institutions in this field are NASA and the National Institute of Health, as well as China’s state-owned electricity company.

However, this concentration on just a few players harbors risks. The AI scene is like a village, which makes collusion between companies very easy and thus makes it more difficult to found innovative start-ups and thus to compete. In addition, the rich and powerful countries attract skilled workers from poorer regions, which widens the gap between poor developing countries and rich high-tech countries.

An AI lumpenproletariat in the making?

Nevertheless, the AI supply chain also holds opportunities for the poor. Despite the flood of data contained in AI models, they remain prone to error. More importantly, not everything that people leave on the internet should be reproduced by AI. After all, if you really want to get to the bottom of a person’s character, you don’t look at their public social media feed, but at their confidential browser history. And this is exactly what millions of automated bots and crawlers do around the clock on homepages, comment columns and dark corners of the internet where people feel unobserved.

In doing so, they come across content ranging from the repulsive to the criminal, such as child pornography or glorification of violence. This flows unnoticed into the data sets that feed our AI applications and chatbots. These reflect this data in their outputs and responses, which must be corrected manually. AI needs human feedback on the correctness and accuracy of its statements. This is challenging given the unmanageable amount of data.

This opens up job opportunities for a young generation in the poorest regions of the world, such as sub-Saharan Africa or South Asia. They can integrate themselves into international value chains through so-called click work. With every click of the mouse, they balance AI in a healthy direction. The Australian company Appen employ s more than one million online workers from over 170 countries for this purpose. The company counts Google, Amazon, Facebook, Airbus and Siemens among its customers, who have outsourced the optimization of their AI models.

Appen breaks down the complex requirements into individual work steps, which it then assigns as gigs to its click workers. Their task is to select the best answers for the chatbot, check the results of search engines, highlight errors or report inappropriate language to an AI. As many people from different backgrounds work on the same tasks in parallel, the company’s own AI calculates the most optimal results from this collection of data.

Another agency, Sama, acquires customers in Europe, the USA and India and ultimately has the work done in Kenya or Uganda, where the internet infrastructure is now well developed. It employs several hundred people there in modern, well-equipped centers. They do not even have to have a school-leaving certificate to do their job and the company offers opportunities for advancement by performing more complex tasks, according to the Austrian Standard. Incidentally, this form of outsourcing also had its origins long before the great AI revolution. It began in 1980 with the laying of fiber optic cables through international waters, which enabled numerous US companies to outsource IT services to India, which made a name for itself as the back office of the world.

Capital remains the fuel for innovation in the age of AI

From AI talent to high-performance chips and energy-intensive computer farms, the immense financing requirements of AI systems are obvious. At the same time, their market success remains mostly uncertain, which makes research and development in the sector risky. Nonetheless, infrastructure, development and data collection must also be provided for these failures. Once again, it is China and the USA that are particularly prominent here.

Top Investoren in AI (Quelle: OECD.ai, 2024)

Top Investoren in AI (Quelle: OECD.ai, 2024)


While the USA, with its huge risk markets, leads the way in 2021 with 144 billion US dollars invested in venture capital, China follows with state-led investments of 48 billion dollars. Far behind are the 27 EU member states, the United Kingdom, Israel, India, Canada, South Korea and Singapore with amounts between 2 and 12 billion dollars.

Despite the general enthusiasm for generative AI around Chat GTP and Midjourney, the largest sums are still divided between autonomous driving technology, social media and marketing. During the height of the COVID pandemic, healthcare, drug development and biotechnology and underlying business processes were the focus of funding.

An unexpected journey: AI connects the world

The political landscape of AI is characterized by tensions between the major powers. All are striving to dominate artificial intelligence within their spheres of sovereignty. America and China have become embroiled in a trade war over Taiwanese high-performance chips and want to cut each other off from their respective supply chains. These power struggles overshadow the feat of globalized and unleashed markets that gave birth to AI. Just as Milton Friedman did when he triumphantly brandished a pencil and explained: “No one person is capable of producing a pencil like this alone. Thousands of people cooperate, speak different languages, practice different religions and may even hate each other.” And yet they all cooperate via a price system. This is even more the case with AI – as between Taiwan and China.

Companies such as Nikon or Canon or the microscope manufacturer Zeiss never intended to take part in an artificial intelligence mission when they were founded. Gamers unknowingly contributed to the fact that the manufacturer NVIDIA had to constantly evolve with their purchasing decisions. Around the turn of the millennium, an inconspicuous company called Google invented a revolutionary search algorithm and a young team of computer scientists founded the campus network Facebook. Just a few years later, both companies came to epitomize the hunger for data. The creation of critical computing infrastructure for AI is largely due to the boom in cryptocurrencies such as Bitcoin. The scene behind it is often a collection of conspiratorial, dogmatically anti-state types.

Global supply chains of artificial intelligence. USA provide the world with software, chips design and capital. Netherlands ASML builds machines for chips production, which is done in Taiwan and delivered to China. In China smartphones and electronic devices as well as infrastructure is produced and exported. Both US and China provide computing power to the world

The global supply chains of artificial intelligence


Only the Asian countries pursued a state-driven development policy. But even when these took shape in Taiwan in the 1970s or in China in 1990, the CEO of IBM had previously declared that demand on the global market would hardly allow for more than five computers. But these development strategies were also based on the exploitation of market economy rules of competition and pricing. Billions of users worldwide revealed their lives on the Internet. This data enabled the development of further new products. As with Friedman’s pencil, there was no commissioner in a central planning office who could plan this development.

On the contrary, countries around the world have their own ideas about what they want to use AI systems for: Controlling information through non-transparent moderation, influencing and restricting freedom of expression, taking over data corporations and interfering in their data policies, social credit systems and total surveillance – right up to the most brutal form, warfare with drones as is currently the case in Ukraine. It is no coincidence that the US Department of Defense and all US branches of the armed forces are among the largest research institutions in this field.

Only the signal of prices, which in Friedman’s pencil example controls the impersonal interactions of people on the global market, has given way to a different currency. In the digital age, many apps and products are free. Instead, they are paid for with data: clicks, dwell times, location data, likes, comments. What AI will generate from this information in the future remains uncertain. But markets, not states, are about to conquer this future for people.

Related Posts

Leave a Reply

Your email address will not be published. Required fields are marked *