AI is the current hot topic. You’ve probably heard it been mentioned on countless podcasts, shows, on social media and on the news. However, one thing has been painful evident, to me at least, from many of the conversations I’ve heard people engaging in about the topic.
Most people don’t have the slightest idea of what the technology is about, and what a future dominated by AI would look like.
At the moment, there are inflated expectations of what the technology is capable of. This has fueled the growing speculations about AI, and has led to massive investment into AI technology and the companies behind it. Much of the stock market is now propped up by what is basically a big bet on AI, as everyone clamors to put invest their money on this ‘futuristic’ technology.
Nonetheless, this doesn’t dilute the potential that the technology carries. Despite the massive flaws, there’s definitely a ton of potential about what it can be used to achieve for humanity — if only it is implemented in the right way.
But before we get there, we may have to wake up to the reality of AI and how it will eventually shape the future of humanity.
Table of Contents
AI still sucks at getting better.
Learning can be defined as the process of acquiring new understanding, knowledge, behaviors, skills, values, attitudes, and preferences.
Learning is what makes us better because we constantly make improvements to ourselves by acquiring something new each time. Learning is engrained into our existence, and is a key constituent to what makes us human. And it is not limited to just humans; most other living things exhibit too.
Learning can be divided into two main categories; associative and non-associative. Non-associative learning pertains to response patterns acquired by changes in exposure to a stimulus. Basically, it is how a living thing adapts to sub-consciously react when presented with continued changes in the environment. Depending on the stimuli, this may trigger a heightened (habituation) or a reduced (sensitization) reaction.
Associative learning derives from drawing connections between two independent situations/events. Associative learning is itself part of a larger form of Active learning, in where there’s a conscious participation of the subject in the process. Active learning is in itself collectivized from many other interconnected forms of learning: play, enculturation, episodic learning, rote learning, deep learning, evidence-based learning as well as both formal and informal learning.
TLDR: Learning is incredibly complex, and vastly interconnected. Yet we humans have found ways to inculcate it into the very fabric of our existence.
Something else, learning isn’t just about acquiring information. It is also about acquiring the ability to make improvements and adjustments through a process of self-correction. Here’s where the vast chasms in learning differences between humans and AI come into play.
Learning in Humans vs AI
Today, computational systems have achieved the capacity to convincingly imitate human thought and reasoning using linear calculations, largely thanks to advancements in neural networks, deep learning, and reinforcement learning. These technologies grant AI the essential ability to self-correct and iterate, which is critical for demonstrating ‘human-like intelligence’. Many of these systems, including popular Generative AI models like the LLMs (e.g., ChatGPT, Gemini, Llama, and Grok), are fundamentally built on Natural Language Processing (NLP) to understand and respond using human language.
Humans engage in reinforcement learning by default, gaining context and knowledge iteratively through a single, unique stream of multi-faceted experiences (episodic memory). AI models, conversely, rely on digesting millions of separate data points from a huge initial dataset.
Learning from a coherent, nuanced life narrative versus learning from rigid, fragmented data is what gives human thought its distinct adaptability and depth.
Human learning is an evolutionary advantage, providing us with an innate context that AI models must artificially acquire from vast datasets. We don’t need a huge initial data dump. Instead, we build knowledge and nuance by living through a continuous stream of experiences ranging from social interactions to professional life. This episodic learning allows our brains to extrapolate far more information from limited, real-world interactions than AI can from its entire rigid training history.
More or less, AI is basically built as an extrapolated version of human thinking, reasoning and importantly learning. However, as we have since established, human thought and learning is complex as is, and is something we’ve not even fully wrapped our heads around. But as it stands, AI is still far off to coming close to acquiring the complexity behind how we humans think and learn.
That said, there still lies an existential danger with the technology. As the technology continues to evolve, it may acquire learning patterns entirely separate from our own, faster than we may realize or adapt to. At this point, would it be sealed off into a Black Box, holding on to it’s reasoning patterns that may or may not surpass our own.
Yeah, so AI still kinda sucks at learning, but no one knows exactly how it would look like if it were to get better at reasoning than us. And that’s the scary part.
But before we get there….
AI’s computing + energy resources debacle
AI has a gnarly problem at the moment.
But to understand it, we must first unravel the architecture that makes AI actually possible.
AI, which as you probably know stands for Artificial Intelligence, is an augmented form of machine learning that leverages on advancements in computing technology that have allowed for the development of ‘thought-like patterns’.
Basically, AI is built on computing infrastructure which is part of its value chain and output. In other words, AI simply wouldn’t be possible without the existence of computers, and its development has been fueled by the subsequent growth of computing technology.
Think of AI as an Operating System i.e. Windows or Mac; reliant on human input to output results, only more complex and with greater intricacies, functionality and ‘autonomy’.
Whenever you prompt an AI model (say ChatGPT, etc.), it has to run your question or instructions through its massive data set of information to extract (or more accurately, predict) what it thinks your answer may be.
Now, sifting through a massive dataset (we are talking about billions if not trillions or more various data points) require a ton of computing and processing power. And as anyone with a gaming PC might reliably inform you, the more the computing power, the more the energy resources consumed.
The demand on computing for many of these PCs necessitates the existence of separate processing units, other than the CPUs, to handle the graphics for most highly sophisticated video games. Consequently, a Graphics Processing Unit (GPU) is a mainstay in any gaming rig.
Scaling this up to the level of AI software, with it’s maze of highly complicated technological demands reveals one thing; a need for sophisticated computing technology.
It is at the height of this debacle that Nvidia, the world’s largest GPU manufacturer, moved in to seal this gap. Before the rise of AI, the company was largely known for it’s GeForce RTX models that were widely popular with gamers. Now, Nvidia offers a wide range of adapted GPU models optimized to handle compute for large-scale AI tasks.
Oh, and it also became the world’s most valuable company while doing so. But let’s save that for later.
Speaking of gaming, a large gaming PC consumes an upwards of 300–500 watts of energy compared to a typical laptop that consumes about 30–70 watts.
An AI model runs a lot more computational processes, which as you would guess, consumes a lot more energy.
How much is ‘a lot more energy’?
What the energy?
To understand the energy demands, we must first come to terms about a crucial cog in the entire AI ecosystem; data centers.
You’ve probably heard about these a lot more recently.
Data centers are vast, industrial-sized operations that house the computing resources on which AI systems are run. And in recent days, building and managing these AI data centers has become big business, given the huge investment into AI at the moment.
But as we have since established, all this AI functionality is not powered by some otherworldly force. AI data centers are the heartbeat that makes the entirety of this technology possible. And for these data centers to remain alive, they need a ton of energy.
Well, what does all this come down to, you ask?
These data centers are basically huge ‘server factories’. For those who may not be aware, servers are huge, powerful computers that provide services, data and information to other computers within the network. Whenever you connect to the internet to, say, access a certain website, your phone/computer connects to one of these servers to request the information on that website that is stored on it.
Because the internet is a global service that is used 24/7, these servers have to remain operational at all times, handling multiple requests from multiple devices at a time. So, in other words, servers are built to be high performance computing models, with a continued and an unending supply of power to remain functional.
Let’s tie this back to our arguement on AI’s massive computational requirements. Most of these AI companies lack the physical infrastructure needed to set-up and run complex computing machines in-house, so they outsource these to data centers, built specificially for AI. These AI data centers aren’t built like your typical server, taking in simple requests for information or simply storing data. They are meant to handle the actual compute for these AI models.
That is; taking in training data, performing complex computational processes that define the working patterns of these AI models, taking in output generation requests when you and I ‘infer’ on the model by giving it prompts, and using the input data we give it to learn and improve on itself.
At most of these data centers, AI models are run on special processing computing chips; i.e. CPUs and the aforementioned Nvidia GPUs, that are molded together onto computing servers. This makes things easier as all modelling, training and inferencing is done in one place. While a single model may only need a couple of GPUs working together, larger models may need more processing and may end up being housed in data centers with 10,000+ GPUs. Alternatively, these models may be placed into thousands of clusters of GPUs in multiple data centers across the world.
(Source: ABC News – Australia)
All these GPUs don’t just run on goodwill. They require energy.
Now, let’s crunch some numbers.
A lot more energy
Two crucial factors play a huge role in determining how much energy a model uses; the size of the model and the size of the prompt.
It perhaps follows that larger models use a lot more energy as they are run on more chips, which consume more energy. Furthermore, larger prompts require more processing power which also runs up the energy usage of any model.
Text-based models are built on an architecture of predicting and generating words based on pre-acquired patterns of speech in relation to the input given. Larger models with more parameters (learned data points that are assigned numerical value to determine the model’s knowledge and understanding) will require more processing as well. Mark you most mainstream models e.g. Chat GPT, Llama, Deepseek etc. have parameters raging in the hundreds of billions to trillions.
Comparing a model with several billion parameters to one with hundreds of billions shows a significant increase in energy consumption demands. The model with 50 times more parameters ends up consuming about as many times more energy.
If accounting for the size of the prompt too, more complex prompts (writing entire essays) consume as much as 9 times more energy as simpler prompts (generating a couple of jokes).
So the correlation here is quite evident. The bigger model and the more complicated the prompt, the more the energy consumed; for text based generated outputs, at least.
Image and video generation models (e.g. Open AI’s Sora, Google’s Veo) use a slightly different method to generate output known as diffusion. Here, the model learns how to generate an image of say, a car, from noise (random variations in brightness and color) by following perceived contours and shapes that a car is made of.
For images and video, the complexity of the prompt doesn’t affect the amount of energy required as much. The key factors for energy consumption are the size of the model (the number of diffusion points) and the resolution quality of the image/video it produces.
Surprisingly, a lot of the image generation models end up using less energy than the more advanced text generation models. The reason for this is because the diffusion method uses less parameters, hence less ‘reasoning instances’ for the model with less energy consumed.
Now let’s scale this up to the commercial level. Remember that all these models are now deployed in mass and are used globally at every second. How much does all this consumption add up to?
(Photo by Eli Hiller/For The Washington Post via Getty Images)
According to the International Energy Agency (IEA), data centers gobbled up 415 TWh of energy as of 2024, with a projected 12% increase year on year since 2017. However, given the massive adoption of AI across 2025, these energy numbers could be much higher than projected.
By 2028, the researchers estimate, the power going to AI-specific purposes will rise to between 165 and 326 terawatt-hours per year. That’s more than all electricity currently used by US data centers for all purposes; it’s enough to power 22% of US households each year. That could generate the same emissions as driving over 300 billion miles—over 1,600 round trips to the sun from Earth.
MIT Technology Review; We did the math on AI’s energy footprint. Here’s the story you haven’t heard. by James O’Donnell and Casey Crownhart; Published May 20, 2025
Initiatives such as the Stargate Project are pledging to invest up to the tune of $500 billion over the next four years, to build even more data centers. These new data centers will by all means, ramp up energy consumption. Therefore, the main AI stakeholders: Open AI, Oracle and SoftBank have been on the hunt, not just for new data center locations, but also for alternative energy sources.
Current plans for these data centers are heavily dependent on natural gas energy sources. However, there are plans in place to harness the high energy output of nuclear power to run the huge power demands of these AI data centers.
Tech companies such as Google, Amazon and Meta have pledged to address this energy issue by pledging to triple the output of nuclear power by 2050.
Despite an insistence on using clean energy sources, the reality is normally much more far-fetched for many of these data centers. Running data centers on entirely clean energy sources is a tall order – especially given the fact that these data centers are expected to remain operational 24/7, 365. For example, what happens when the sun goes down and a data center that uses solar power needs to remain operational throughout the night? Most times, the most viable option would be for the center to switch over to ‘dirtier’ energy sources such as coal or natural gas.
Much of the world, and the US especially, is far behind on the adoption of clean, renewable energy sources. Therefore, it means that the most accessible energy source for a typical data center build up is often a more carbon-intensive resource.
Consequently, it should come as no surprise that the carbon intensity of electricity used by data centers was 48% higher than the US average.
A typical AI data center uses 100,000 times more energy than a normal household with the larger centers using up to 20 times more energy. Furthermore, a lot of these centers use up more gallons of water to run their cooling systems. According to the Environmental and Energy Study Institute (EESI), a large data center uses up about 5 million gallons (19 million liters) of water in a day – the equivalent of a town of 10,000 to 50,000 people. Added up, that’s about 1.8 billion gallons (7 trillion liters) of water in a year.
If that’s too huge of a number for you to think about, think of it this way. Generating a 100-word AI prompt is estimated to use about 1 bottle of water (0.5 liters). Now scale that up for all the AI models and the usage across the world.
Mark you, these numbers don’t fully accout for the future use of AI. For most people, AI may just be a novelty on a chatbot or an image/video generation tool at best. However, a lot changes when AI becomes starts running agents that perfom automated workflows with minimal input, powering humanoid robots that perform household tasks for us, performing complex research tasks and generating reports using deep research and reasoning models.
As AI becomes embedded and personalized into our lives more and more, and is expected to perform even more and more complex tasks, its energy demands will consequently increase. The more complex it gets, the more the energy it demands – and as we’d earlier established, running complex AI functionalities increases demand on energy sources by up to 50 times.
The energy debacle is especially worrysome at the moment. The global energy grids are disjointed and heavily dependent on carbon intensive energy resources which, as countless research has shown, is a direct contributor to an increase in global warming. Global warming is a direct threat to the existence of our species and the planet at large as it disorients ecosystems and threatens the continued existence of life in it’s current form.
In other words, the hyper-scaling of AI technology places a huge expense on energy resources, and could undo most of the gains made to streamline energy use away from carbon-intensive sources.
But before we get there…..
The elephant-shaped bubble in the room
If 2021 was the year of NFTs, 2022 the year of climate-led investment, 2023 the year of crypto and 2024 the year of commodities investment, 2025 has been the year of AI investment.
Record levels of investment has gone into AI-led or AI-based technologies, projects and start-ups.
A huge reason for the rapid increase in AI spending has been the technology’s inculcation into a lot of services and products. Every website now has an AI assisted chatbot. Every app has an inbuilt AI feature. Entire customer support and service teams are being replaced by AI agents. If it connects to the internet, it sure has an AI-enabled feature connected to it.
Heck, even Google might end up replacing it’s search engine architecture with the AI overviews feature that it’s been shoving down our throats for so long.
AI is (and rightly so) marketed as the technology of the future. Every company and investor alike doesn’t wish to be left behind as the technology takes off. And the market has the numbers to back this up.
Estimates by Gartner in September 2025 put projections on cumulative AI spending at a mouthwatering $1.5 trillion.
Of the projected spending figures on AI in 2025, $300-400 billion are funneled into building AI infrastructure. A huge chunk of that came from the $100 billion Nvidia – Open AI deal announced in September 2025. The deal attracted a ton of backlash and was pointed to by critics as evidence to an AI Circular Economy and a plausible indication for a larger AI Spending Bubble.
(Credit: Grant Hindsley for The New York Times)
Nvidia seemingly recognized the demand for power for it’s GPU chips, and pledged to cover this by investing into data center infrastructure through it’s largest clientelle, Open AI. Basically, Nvidia pledged to gradually invest the $100 billion into Open AI as the company buys more and more from it; all of which would be channeled into building AI data centers, of course.
For most critics, the deal presents a massive conflict of interest — mostly for the absurdity of ‘collusion investment’. In layman terms, Nvidia seems to be grooming Open AI into buying more from them, after which Nvidia would turn around and pump more money into Open AI. The nature of this deal is especially worrysome if you consider that much of the AI economy is built on such ‘quid pro quo’ financial arrangements.
The ‘scratch my back I scratch yours‘ engagements extend into debt arrangements between these companies, as they borrow money from each other to then invest it into each other, then buy each others’ products and services.
Crazy, I know.
It is through these financial arrangements that AI companies can justify their massive valuations.
However, this AI spending trend should come as no surprise as it is representative of a larger macro-economic trend.
The global economy has grown tremendously over the past decade alone, and in fact, has continued to grown even when all other metrics suggest otherwise. All this despite an uptake in unemployment rates and increase in basic commodity prices (which are normally an indication of an economic go-slow).
Part of this is due to the AI hype that has attracted a ton of economic mobility as investors pump money into AI companies, projects and start-ups.
A Financial Times article based on an MIT report found that the US economy only grew at an annualized rate of 0.6% for the first half of 2025, while not accounting for AI spending. In fact, 92% of GDP growth over that period came from AI spending and 80% of gains in the stock market for 2025 came from investment into AI.
More than 20% of the S&P 500 index (which tracks the US’s most valuable companies) is dominated by three companies; Nvidia, Microsoft and Apple – two of which are basically big bets on AI.
AI has become the only part of the US economy that seems to be doing well – and this has had ripple effects on the larger tech industry. Tech companies are now the most valuable companies in the world, and by a long shot, and their billionaire founders and CEOs now have unprecedented power — from Silicon Valley to State House.
Trump and his government know this. The AI and Tech economic pillar may be the only saving grace of the current disaster class that his presidency has been so far. Therefore, he has handed tech companies the proverbial ‘keys to the kingdom’, letting them run loose with little to no regulations.
Foul Economics
However, the current AI investment trajectory is testament to a larger economic trend in the market that is propped up by two main factors; private equity and debt financing.
There is a third, even more arduous factor, but for the interest of keeping this (already really long article short), I’ll save that for another day.
Now, where we we?
Private equity has grown tremendously over the past couple of decades and has become a mainstay in the current economic system. It offers a plausible alternative to the typical public markets, which despite having a wider investment reach, remain exposed to market forces and heavy-handed regulation.
Private equity provides several advantages which has made an attractive investment vehicle for investors:
- Higher Returns: It historically provides a premium return over public markets, attracting large institutional investors like pension and sovereign wealth funds.
- Operational Value Creation: Modern private equity firms actively enhance acquired businesses through specialized expertise, implementing long-term strategic and operational improvements (e.g., better governance, technology upgrades).
- The Lure of Privacy: The number of public companies is declining because many businesses prefer to avoid the high costs of increased regulation, compliance, and pressure to meet short-term quarterly earnings mandates.
- Deep Capital Access: The private capital market offers expansive pools of financing which include leveraged debt in low-interest-rate environments) that allow firms to execute larger deals and magnify returns.
- Long-Term Focus: The illiquid, long-term nature of private equity capital of typically 5-10 years allows management to focus on significant transformations without the immediate scrutiny of public investors.
- Diversification Benefit: Private equity investments offer investors uncorrelated returns compared to public markets, providing valuable stability and diversification to investment portfolios.
It therefore no surprise that private equity has become a lucrative investment vehicle for investors (particularly wealthy investors), companies and investment collectives. In fact, it estimated that 20% of US corporate equity is tied up in private equity as of 2021, compared a figure of 4% in 2000.
Private equity-backed venture capital firms have become the new investment juggernauts in the economy and have played a particularly important role in AI funding and investment.
Of the $252.33 billion of corporate AI investment in 2024, $150.79 billion came from private investment – a 44.5% increase from the previous year. In fact, the AI private investment space has increased thirteenfold since 2014.
It is at this point that I should mention that most AI startups and companies; e.g. Open AI, are held up in private equity.
And what’s the big deal with that, you ask?
While private equity is a great alternative to trading in the public market, it presents significant risks that cannot be ignored. Privately held companies tend to be less transparent, and it can be hard to gain information on their financial well being. Remember, these companies aren’t obliged to publicly disclose information on their financials on a quarterly basis.
Just ask any researcher who has sought any information from Open AI, especially on their financials and operations.
The risks increase tremendously when these companies are funded by publicly traded companies. For example, Microsoft owns a significant 27% of Open AI stock (which is worth about $135 billion). If Open AI were to go under (which would be less apparent now than if it were publicly traded), this would wipe out a pretty significant share of Microsoft valuation — affecting Microsoft’s public investors who were otherwise not invested in Open AI.
Microsoft and its compatriot pack of the ‘Magnificent Seven’ (Alphabet, Amazon, Apple, Microsoft, Nvidia, Meta and Tesla) have huge multi-national reach, power and influence. So huge that if their stability faltered, it would affect the vast majority of the global population – directly or indirectly.
Now, let’s talk debt financing.
According to the IMF, global debt (both public and private) as a share of GDP stands at a staggering 235%.
Basically, the current global economy is debt-heavy and we owe more than we produce. This debt burden has far-reaching effects, as it ties up both public and corporate revenue through ever-rising interest rates in debt repayments and severely hindering economic growth.
An elevated debt burden is a teotale indicator of global economic strain. Much of this arises from the COVID pandemic of 2020 as many governments and businesses alike had to find ways to borrow to sustain themselves in the absence of economic productivity.
While the evidence of the debt burden is much easier to document in the public domain, it makes for a plausible arguement that similar issues are evident in the corporate investment i.e. AI startups and companies.
Many AI companies are running with loads of investor-fueled debt on their books with little if any profits to show for it. The revenue they generate is not enough to keep up with the rapid scale in technology and infrastructure that the AI-fueled requires at the moment.
It takes only a critical look at the books for any of the AI companies to establish that these companies are running firmly in the red.
CoreWeave, a major data center operator, expects to bring a revenue of $5 billion while spending $14 billion just this year. Open AI already has $96 billion debt from it’s biggest backers; Oracle, Softbank, CoreWeave and other venture capital backed firms and banks. Open AI has also made commitments to take up $1.4 trillion in financing to cover its energy, compute and operational commitments in the future. All while making only about just $20 billion in expected revenues this year.
A lot of this debt has been acquired through the high interest rates set by the private investment market that these companies are based in, and is due to mature over the next couple of years. Revenues are nowhere enough to cover the cost of repayment, let alone keeps these companies businesses going.
What happens when the payments are due? Many of these companies ma be forced to take up even more debt on higher interest rates and worse repayment terms, which would drive them even deeper into debt.
Even worse, it may force their publicly traded, private investors deeper into their pockets. Before the AI boom, the financing needed to keep these AI initiatives, projects and start-ups going came straight from the pockets of Amazon, Alphabet (Google), Microsoft and Meta.
However, according to Bank of America data, the five largest technology giants—Amazon, Google, Meta, Microsoft, and Oracle—have collectively issued $121 billion in new debt this year specifically to fund their AI initiatives. This extraordinary sum is four times the average debt of $28 billion which these same companies took on over the preceding five-year period.
The current AI economy is treading on a high-wire tread rope that could snap at any moment, likely when the debt and credit defaults kick in, which could be way sooner than most realize.
The debt-fueled AI hype reminds a lot of the 2008 Global Financial Crisis; which occured when the inflated housing market collapsed brought down by mortgage defaults. As more and more people defaulted on their mortgages, it emerged that a lot more of the global market had bought in to the housing market through financial products which were basically bets hedged on these mortages and loans.
It took one peace of the domino to fall to expose a deeper, decaying rot within the global market built on hype on an uproven ‘innovation’. While the circumstances for this bubble could be quite different than in 2008, they share enough parallels to get me, and many like me, worried that we are headed for the same eventuality.
Will these AI companies grow fast enough to keep up with their debt and financial obligations? What happens if they don’t?
Would we have a repeat of 2008 where the banks (insert AI and tech companies for today) are too big to fail, and the burden of their financial miscalculations is placed on all of us?
Geo-Political Implications
China
The hyperscale AI investment is now felt across the world. AI has become an important geopolitical tool and has inserted itself as a dimension of the US-China trade war.
China has been quietly developing it’s own AI models and technology, aiming to surpass the United States and the west. This is not a newly-found plan for them either. In 2016, the Chinese Communist Party (CCP) announced its five-year plan in which it laid down a strategic vision to become a global leader in AI by 2030.
The United States recognized the threat of AI advancement in China, alongside other crucial technologies and in 2022, it restricted China’s access to advanced computing chips and technology. In 2023, Biden’s administration expanded to limit investment into Chinese companies from the US, citing rising concerns in the national security.
However, that has not slowed down progress in China. To cope with the shortage in compute, China optimized it’s models for efficiency. When DeepSeek (dubbed as China’s Chat GPT) released in 2024, the world was astounded by its performance — which rivaled those of its American rivals despite using a fraction of the computing resource.
All in all, the rapid advancement of China’s AI landscape has been flying under the radar as the much of the attention has gone to American AI companies and startups. Even with the advent of Chat GPT in late 2022, Chinese companies had been leading the pack, releasing their own versions of various LLMs.
China’s Beijing Academy of Artificial Intelligence released the world’s largest pre-trained language model (WuDao), as early back as May 2021.
While the country’s AI landscape is valued at just about $170 billion (which may seem like a miniscule figure compared to the US), it is also more regulated and institutionalized.
China implemented crucial regulatory and diplomatic moves concerning Artificial Intelligence during the latter half of 2023.
In July 2023, the nation formally introduced its foundational rules for the management of Generative AI offerings, titled the “Interim Measures for the Administration of Generative Artificial Intelligence Services”. This initial regulatory step was followed in October 2023 by a proposed set of minimum safety standards for fundamental generative AI services. This draft outlined specific prerequisites for processes such as data gathering and the subsequent training of AI models.
Coinciding with this regulatory work, the Chinese government unveiled a significant diplomatic effort in October 2023: the “Global AI Governance Initiative”. This initiative positions China’s AI strategy within its broader vision of a “Community of Common Destiny,” aiming to foster international conversations about AI policy, particularly with nations in the developing world.
A key feature of this global framework is its emphasis on mitigating risks associated with AI, specifically citing worries about the misuse of data and the potential deployment of AI by extremist groups.
The Chinese government has also taken a proactive role in leading AI development, deployment and research. It actively funds peer reviewed papers on AI and supports AI initiatives through its Guidance Funds: a public-private investment initiative meant to pursue industrial and economic growth goals. Therefore, it is no surprise that China is home to most of the world’s top researchers on AI, is the biggest publisher of patents related to Generative AI and has the most open and publicly available AI models.
However, given the strict nature of authoritarianism and censorship by the Chinese government, experts have called into question the true nature behind Chinese proactiveness into AI investment. Despite imposing huge restrictions on tech companies in the AI forefront, the Chinese government has been accused of harnessing AI to further enforce censorship and scrutiny through advancement of facial recognition and speech recognition technology.
China has also been suspected to use AI as a tool of interference, especially in elections for countries such as the US, India and South Korea.
The Chinese Military is a key interest party in the development and use of AI technology. China is set up quite differently to most western countries, with less distinction between government and non-governmental actors (commercial companies and university laboratories and thinktanks). Therefore, the government has a lot of say on what technology is developed and at times has unlimited access to it.
China has been pretty brazen about its mission to build a military heavily dependent on unmanned technology and AI. Much of these arise from fears of falling behind to similar military advancements in the US and NATO. A key constituent of building this intelligence comes from gathering data (especially from it’s perceived geo-political rivals).
Reporting in December 2024 by ABC News found that a Chinese hacking and espionage campaign stole the data of than 1 million US customers by exploting weaknesses in the telecommunications infrastructure.
Data privacy in the age of AI is a key tool of warfare for both China and the US. In 2022, China passed the Data Security Law of the People’s Republic of China which not placed obligations on data use for tech companies, but was part of a wider geo-political framework on data collection and storage for both local and non-local tech companies. It mandates that data transfer to foreign law enforcement or judicial agencies requires official approval.
The act was in retaliation extraterritorial reach of the US’s CLOUD Act passed in 2018 which grants US-based law enforcement agencies the authority to compel technology companies headquartered in the United States to hand over customer data, regardless of where that data is physically stored around the world.
European Union
AI companies and startups in Europe have also seemed a tremendous amount of growth, albeit at a smaller scale and pace compared to the US.
The European AI market size was valued at about $53.03 billion in 2024 and was projected to grow to $65.48 billion in 2025. AI investment in the EU has mostly been clustered in the UK, France and Germany. France startups led in 2024, receiving $1.6 billion of investment in 2024, which was about half of all investment into AI that year.
More recently however, more AI unicorns have been emerging from other countries in the Netherlands, Sweden, Switzerland and Norway. AI startups within countries in Central and Eastern Europe have also received more investment. Poland has been a top destination for AI funding in the region closely followed by Greece and Croatia, all of whom had received $120 million in AI-related investment just as of March 2025.
What Europe lacks in building massive scaling it makes up for in building robust regulatory frameworks. The EU has created a strong emphasis for ethical transparency, hence positioning itself as a responsible leader in the adoption of AI. Nonetheless, AI has also seen widespread adoption across multiple industries in the public and private sectors — much as is the case globally.
A key avenue of AI adoption has been in regulating the cybersecurity space. The European Commission and the High Representative of the Union for Foreign Affairs and Security Policy adopted The EU’s Cybersecurity Strategy for the Digital Decade in 2020. This document outlines measures to build cyber resilience and guarantee trustworthy, unfettered access to digital services and technology. Consequently, this was a foundational framework for the adoption of the European Union Agency for Cybersecurity (ENISA).
(Source: European Parliament)
A key enabling factor has been the Proposal for a Regulation of the European Parliament and of the Council laying down harmonized rules on Artificial Intelligence (Artificial Intelligence Act – AI Act). The AI Act, which came into force on August 1, 2024, is the world’s first law that comprehensively lays out guidelines for AI deployment and usage. It is a risk-based regulation that categorizes AI systems (unacceptable risk banned, high-risk with strict rules, low-risk largely unregulated) to ensure safety and trust.
The Act sets specific legal requirements for AI technologies across several key areas:
- Data and Governance: Rules for how data is handled and managed.
- Transparency and Documentation: Mandates for clear records, thorough documentation, and providing users with necessary information.
- Performance: Requirements for AI system robustness, accuracy, and cybersecurity, along with the necessity of human oversight.
For providers of high-risk AI systems, the proposal imposes additional burdens, requiring them to:
- Establish a written quality management system which should include defined policies, procedures, and instructions.
Finally, the document also defines obligations for importers and distributors and details a conformity assessment procedure designed to keep the burden on businesses and economic operators minimal.
All in all, the European Union has chosen to take the path of ‘most resistance’, sacrificing rapid short-term growth for longer term safety and sustainability. Perhaps as the hype dies and the reality of AI’s ability comes to light, this may emerge as the rightful decision after all.
Only time will tell…..
But wait, you say, you didn’t mention anything about the developing world! Well, there’s a lot to more to cover about that — which I think is warranting of it’s own article.
MEME TIME……
Leave a Reply