The limits of progress: Why AI development could slow down

​Tech giants are investing hundreds of billions of dollars in artificial intelligence. But the race to build increasingly powerful models is facing rising costs, resource shortages, and growing investor skepticism. Is AI approaching the limits of its capabilities?

Billions that may not be enough

Broadcom CEO Hock Tan warned that a shortage of capital could become the biggest obstacle to AI development. Speaking at a conference in Singapore, he noted that shortages of chips, memory, and electricity could ease over time as production capacity and infrastructure expand. However, access to capital could become the main constraint on the industry's future growth.

View post on X Loading...

The tweet was deleted by the author.

But we saved everything 🙂.

The largest technology companies continue to increase spending on AI infrastructure. In the second quarter of 2026, combined capital expenditures by Microsoft, Alphabet, Amazon, and Meta reached $165 billion, up 87% from a year earlier. A significant portion of this spending goes toward building data centers and purchasing equipment needed to train and run AI models.

Against this backdrop of massive spending, Intel CEO Lip-Bu Tan highlighted another problem: the cost of capital and the time needed to recoup investments. Rising bond yields make borrowing more expensive, increasing the cost of building data centers and purchasing equipment. Meanwhile, investors want to know when enormous investments in AI will begin generating profits.

Nevertheless, neither executive expects a sharp slowdown in the AI industry over the next few years. Hock Tan predicts that demand for computing capacity will exceed supply through 2029 and possibly into 2030. At the same time, the Broadcom CEO criticized investors who pour money into AI without understanding the industry's prospects. "A lot of investors have no clue what they're investing in, but set very extreme expectations, only to find out that it's almost like they are in a casino gambling," he said.

When AI hits physical limits

Even with sufficient funding, AI development could slow down because of electricity shortages. According to the International Energy Agency (IEA), data centers worldwide consumed 485 TWh of electricity in 2025, up 17% from the previous year. Electricity consumption by AI-focused data centers surged by 50%. By 2030, total data center electricity consumption could reach 950 TWh, or around 3% of global electricity demand.

The problem is not just the amount of electricity available, but also the ability to connect new facilities to power grids. The IEA estimates that around 20% of planned data center capacity could face delays due to grid infrastructure limitations. For example, building new transmission lines in developed countries can take four to eight years. Meanwhile, delivery times for transformers and cables have doubled over the past three years.

Another constraint is the production of advanced chips. Dutch company ASML, which manufactures equipment used to produce cutting-edge semiconductors, expects demand from the AI industry to exceed supply until at least 2028. The company is considering expanding production capacity as customers are already inquiring about equipment deliveries for 2029–2030.

View post on X Loading...

The tweet was deleted by the author.

But we saved everything 🙂.

At the same time, the demands placed on data centers are increasing. According to IEA estimates, power consumption by AI servers increased elevenfold between 2020 and 2025 and could quadruple again by 2027. A single server rack the size of a large refrigerator could potentially consume as much electricity as 65 households. This places greater demands on cooling systems, power supplies, and equipment that must withstand sudden changes in workload when running AI models.

Where are the limits of progress?

Shortages of electricity, chips, and data centers are not the only obstacles to further technological development. Even if these problems are resolved, another question remains: will new models improve enough to justify rising costs? Hock Tan acknowledges that demand for computing capacity could slow if new generations of AI models stop significantly outperforming their predecessors. However, there are currently no signs that progress has stopped.

At the same time, increasing computing power does not guarantee that AI will perform equally well across all tasks. For example, advanced models can already solve complex mathematical problems but still struggle to work independently with computers. In the OSWorld benchmark, which tests AI's ability to perform tasks in computer applications, the best systems successfully completed approximately two-thirds of the tasks.

The practical benefits of AI for businesses are equally important. In August 2026, McKinsey published the results of a survey of 1,719 company representatives across 97 countries. Around 80% of respondents said AI had improved their personal productivity. However, only 37% reported a positive impact on their organizations' operating profits.

A new reality for the technology race

There is currently no reason to conclude that artificial intelligence is approaching the limits of its capabilities. However, maintaining the current pace of development could become too expensive even for the largest technology companies. If spending continues to grow faster than revenue, investors may have to reconsider their expectations, while developers may need to reduce the costs of creating and operating new models.

This is more likely to lead to a shift in priorities than an end to technological progress. Instead of racing to build the most powerful models, companies may focus on cheaper, more efficient solutions capable of generating real profits. In that case, the future of AI development will depend less on the size of investments and more on the technology's ability to solve practical problems for businesses and everyday users.