Why are AI data centers so expensive to build and operate?

The cost of AI data centers has significantly increased, from $10-15 billion per gigawatt in 2020 to $40-50 billion now. This, combined with the need for constant hardware replacement and the relatively small number of jobs created, makes many AI businesses financially unviable.

How does technology, like AI, affect the economy?

Technology is a deflationary force, making automated tasks cheaper over time. This could lead to the overall economy shrinking, but not necessarily negatively, as reduced costs for goods mean people can buy more with the same income.

What is the 'K-shaped economy' and how does it relate to AI?

A K-shaped economy describes a scenario where capital owners gain increasing wealth while wage labor stagnates or declines. The speaker argues that AI's current trajectory, particularly with large tech IPOs, is exacerbating this trend, leading to growing inequality.

What can be done to ensure AI benefits humanity?

To avoid a negative future, AI must be treated as a public good, regulated for safety, and its development slowed to allow societal adaptation. It's crucial to combat greed and ensure AI development serves everyone, not just a few.

What is the '3.5% rule' and how does it offer hope?

Erica Chenoweth's research shows that nonviolent movements achieving just 3.5% participation can lead to significant societal or political change. This suggests that collective action and protest can influence the direction of AI development and regulation.

What are the proposed solutions for managing AI's societal impact?

The speaker proposes treating AI as a public good, regulating it for safety and societal adaptation, and creating social safety nets like a 'GI Bill for AI' or a 'Marshall Plan for AI' to ensure global benefit and mitigate negative impacts.

Key Moments

What happens if the AI market crashes? | Alvin Wang Graylin | TEDxBerlin

TED

Nonprofits & Activism6 min read22 min video

Aug 7, 2026|4,271 views|159|12

AI English Future Global issues TEDxTalks Technology [TEDxEID:68282]

Save to Pod

Want to know something specific about what's covered?

We've already dissected every moment. Ask and we will deliver (with timestamps).

Key Moments

TL;DR

The AI market is poised for a crash due to unsustainable spending and inflated IPOs, potentially creating a K-shaped economy where capital gains vastly outpace wage growth.

Key Insights

AI companies are projecting revenue streams that dwarf global software revenue, aiming to capture the entire white-collar labor payroll ($20-30 trillion).

Data center costs have quadrupled from $10-15 billion per gigawatt in 2020 to $40-50 billion, with Financial Times reporting four out of five companies would be loss-making even without operating costs.

Chinese AI models now account for 30-50% of AI traffic, up from 1% a year ago, due to their significantly lower cost compared to US investments.

Economists predict 2-4% global GDP growth over the next 20 years, a stark contrast to Elon Musk's projection of 10x GDP growth, which would require 26% annual growth.

The upcoming IPOs for major AI labs like SpaceX (non-space revenue 93%), Anthropic, and OpenAI are structured to create artificial scarcity and demand, with significantly reduced lock-up periods for early investors.

Unemployment for youth in the US is 9%, with a 42% underemployment rate, indicating that more than half of young Americans are in jobs below their education level, a trend exacerbated by AI's impact on the job market.

The AI market's unsustainable race for profit

The AI industry is currently engaged in a high-stakes, zero-sum race, driven by the prospect of a massive $10 trillion windfall. This rush is fueled by ambitious financial projections, such as those from SpaceX, where 93% of its projected revenue comes from AI-related businesses, aiming to capture the global white-collar payroll of $20-30 trillion. This strategy is financially precarious, as evidenced by soaring data center costs, which have increased from $10-15 billion per gigawatt in 2020 to $40-50 billion currently. A study by the Financial Times indicated that four out of five AI companies would remain unprofitable even if data center operating costs were eliminated. Furthermore, the current model struggles with job creation, with data center deployments generating only about 1% in payroll revenue. The notion of space-based data centers to save energy is also questioned, given that energy costs are only 1.5% of the total, and the logistical challenges of maintenance in orbit are immense, with 10-15% of devices failing annually.

The cost advantage of Chinese AI models

Despite significant investment in the US, Chinese AI models are rapidly gaining market share due to their drastically lower costs. A year ago, Chinese models accounted for only 1% of AI traffic, but this has surged to 30-50%. This shift is driven by the commoditization of AI traffic and the development of more affordable alternatives. While US AI investments are in the hundreds of billions (projected at $1.1 trillion next year), China spends approximately one-tenth of that. The performance gap between US and Chinese AI is narrowing, with only a 2-3% difference in performance and a 2-3 month lag, yet the cost of AI inference can be 10 to 50 times higher for US-based solutions. This economic reality is pushing developers and users towards cheaper, more accessible Chinese models.

Economic growth projections versus AI's deflationary impact

The narrative driving AI investment is a prediction of unprecedented economic growth. Elon Musk, for instance, forecasts a tenfold GDP increase within 10 years. However, this projection is highly optimistic and contradicts historical economic trends and expert opinions. Over the past 40 years, global GDP growth has averaged only 2-4%, with current rates around 2%. Even economists within AI companies project only 2-4% growth over the next two decades. A critical factor often overlooked is that technology, particularly automation, is a deflationary force. As AI automates tasks, the cost of goods and services decreases. This trend suggests that overall economic growth might be slower, or even that the economy could shrink in real terms if productivity gains outpace demand. However, this is not necessarily negative; if incomes remain stable while costs decrease, people can afford more, leading to an improved quality of life. This deflationary pressure on AI inference, decreasing by 9x to 900x annually, poses a significant challenge for AI labs reliant on AI as their primary revenue source.

The Jevons paradox and the limits of AI compute demand

The concept of the Jevons paradox suggests that as a resource becomes cheaper, its consumption increases, potentially leading to greater overall use. While AI usage is indeed rising, this paradox has limitations. Historically, even as coal became cheaper, energy consumption eventually plateaued because there's only so much energy a society can utilize. Similarly, there is a finite limit to the amount of compute power people will need, regardless of its cost. This principle, related to Kuznets's work on GDP measurement, indicates that AI compute demand may peak. The idea that AI will drive infinite growth is therefore questionable.

The geopolitical race and the AGI disruption

A significant driver behind the current AI spending is a geopolitical race, primarily between the US and China, to achieve Artificial General Intelligence (AGI). AGI is defined as technology capable of replacing white-collar workers, and its advent would represent a profound disruption to capitalism as we know it. This race is characterized by differing national strengths: China boasts strong infrastructure and energy resources, while the US prioritizes speed. Europe, though slightly behind, focuses on social safety nets and regulation. However, this race is not without global consequences; interconnected systems mean that a major market collapse in one region could trigger a worldwide downturn. The 'finish line' of AGI is not a point of triumph but a potential catalyst for widespread economic and societal upheaval.

The investor-driven AI IPO bubble

The intense activity in the AI market is largely propelled by the desire for an 'exit' for early investors and venture capitalists. Companies are racing towards Initial Public Offerings (IPOs) to cash out, with upcoming IPOs for major AI labs like OpenAI and Anthropic expected to be in the trillions of dollars. These IPOs are structured with features like artificial scarcity (low share availability) and expedited index inclusion to create artificial demand, potentially forcing trillions in capital into these stocks. The short lock-up periods (e.g., 60 days) allow early investors to sell quickly, often leaving average investors with less favorable terms. Control also remains concentrated, with founders often retaining a majority of voting rights, limiting external influence.

The K-shaped economy and its impact on employment

The current trajectory of AI development is creating a 'K-shaped' economy, where capital gains increasingly benefit those who already possess wealth, while wage labor sees diminishing returns. This is visible in the stock market, where rising markets are no longer correlated with increased job openings. Since the launch of ChatGPT, stock markets have climbed, yet job listings have declined. This trend is particularly concerning for young people and early-career professionals. In the US, youth unemployment stands at 9%, with an underemployment rate of 42%, meaning over half of young Americans are in jobs below their qualifications. This widening gap between capital and labor is seen as unhealthy and unsustainable for societal well-being.

Call to action: Regulating AI for societal benefit

Despite the potentially bleak outlook, there is hope for a positive AI future if society intervenes. Drawing on Erica Chenoweth's '3.5% rule,' which shows that significant societal change can occur when 3.5% of a population actively protests, collective action is possible. The speaker urges treating AI as a public good, shifting from a zero-sum to a positive-sum model, and implementing regulations for safety and societal adaptation. The rapid pace of AI development, expected to unfold in 5-10 years compared to 40-80 years for previous industrial revolutions, requires proactive measures. Key actions include demanding AI serve humanity, ensuring equitable distribution of benefits (a 'GI Bill for AI' or 'Marshall Plan for AI'), and establishing global institutions to prevent misuse. The ultimate goal should be the betterment of humanity, aligning with Nikola Tesla's vision of science serving the greater good.

Mentioned in This Episode

●Software & Apps

●Companies

●Organizations

●Concepts

●People Referenced