Why Tech Companies Are Quietly Cancelling AI Data Centers

[00:00:00] Speaker 1: More than half of all AI data center construction projects worldwide are being canceled or delayed. [00:00:05] Speaker 2: Half of U.S. data centers planned for 2026 are expected to be delayed or canceled. [00:00:11] Speaker 3: Breaking news on the Bloomberg terminal, Oracle and OpenAI are apparently ending plans to expand that Texas data center site. [00:00:19] Speaker 1: Considering that in 2025, more money was spent building data centers than on housing construction, this is a big deal. [00:00:26] Speaker 2: Hey John, so Google is now planning to invest $40 billion in new data centers. Amazon is pumping $20 billion into Pennsylvania. $500 billion investment from one of the biggest tech companies. [00:00:37] Speaker 1: One of the reasons is that there is an incredible bottleneck in the electrical grid, since these centers already consume close to 2% of global electricity. [00:00:45] Speaker 4: Millions of Americans could suddenly lose their power this winter as the nation's power grid deals with an increased strain. This is thanks to surging numbers of data centers opening. [00:00:55] Speaker 1: Even so, most of these data centers obtain around 60% of their electricity from fossil fuels. And due to rising prices caused by the war in Iran, these costs are spiraling out of control. [00:01:05] Speaker 3: Putting right now, gas prices are rising again. The price at the pump has been jumping every day, driven by the war in Iran. Everyone feeling the pain. Here's where the numbers stand right now. In Pennsylvania, gallon or regular will cost you $4.58. That's up 8 cents from yesterday. [00:01:20] Speaker 1: In addition, the numbers do not add up. Experts estimate that NVIDIA, the largest producer of chips for this technology, is overestimating demand. [00:01:28] Speaker 5: What we're seeing too is supply right now is constrained. And I'm sure you've heard a bit of this. And what's happening, and it's going to get exacerbated over time. [00:01:37] Speaker 1: This has caused tech companies like Google, OpenAI, and Oracle to buy enormous quantities of chips and GPUs, assuming a future shortage that now appears to be false. [00:01:46] Speaker 6: For its first 30 years, NVIDIA wasn't a household name, unless you were a gamer. But now, some of its original fans feel left behind as the memory shortage pushes the world's most valuable company to prioritize its massively profitable AI chips over gaming GPUs. [00:02:02] Speaker 1: So, why tech companies are quietly canceling AI data centers? Since the public release of ChadGPT at the end of 2022, the tech industry entered an unprecedented race to build the infrastructure needed to sustain the AI boom. What started as a flashy innovation quickly became a strategic priority for the world's largest companies. Within months, companies like Microsoft, Google, Amazon, and Meta began announcing multi-billion dollar investment plans focused mainly on the construction and expansion of data centers. [00:02:34] Speaker 2: Meta continuing its massive investments in AI, this time in the form of compute power, announcing two new major AI data centers. Amazon investing another $15 billion into northern Indiana to build an AI data center campus. [00:02:48] Speaker 1: By 2023, global spending on AI-related infrastructure had already surpassed $200 billion. And just one year later, in 2024, that figure practically doubled. In 2025, it is estimated that the largest tech companies allocated nearly $400 billion in capital expenditures. Much of it directed toward data centers, specialized chips, and cooling systems. To put this into perspective, that annual investment exceeded the total amount spent on single-family home construction in the United States during the same period. A sector that has historically been one of the largest recipients of investment in the economy. But there is a problem. The vast majority of these investments in data centers were made based on simple overestimations that are now falling apart. [00:03:34] Speaker 7: As you alluded in your intro comments, like, is this a repeat of the Metaverse project, which I think a lot of investors, you know, never really kind of understood the, you know, the return on that. And again, I think investors do understand the return on AI, but it's such a level that I think it's for projects that, you know, don't even reflect the current business today. [00:03:57] Speaker 1: But to understand this, we need to focus on the role NVIDIA plays in all of this. Due to the AI boom, demand for GPUs has exploded. And NVIDIA, being the world's largest supplier of these tools, has also grown at an astonishing pace. In 2025, it reached a $5 trillion market valuation, making it the most valuable company in history. [00:04:17] Speaker 8: Move over Microsoft. Chip maker NVIDIA is now the most valuable company in the world, worth more than $3 trillion. Yes, that is trillion with a T. It surpassed Microsoft. That's why we keep saying move over Microsoft. NVIDIA, say, channels of NVIDIA, rather, are up a whopping 174% since the start of this year. [00:04:40] Speaker 1: But this also came with a major risk. Any significant change in its price could have implications for the global economy. One of the major problems emerged when suspicions began to arise that most of the investments and projects NVIDIA planned to carry out were heavily overestimated. In an interview with CNBC, Jensen Huang stated that the company was distributing around 10 gigawatts worth of GPUs in 2025 alone. He made this statement at the same time NVIDIA was announcing that it would invest more than $100 billion in infrastructure for OpenAI. [00:05:13] Speaker 9: I am here at NVIDIA headquarters in Santa Clara with the CEO of the world's most valuable company and the CEO and president of the world's most valuable private company, Jensen Huang of NVIDIA, Sam Altman, and Greg Brockman of OpenAI. So let's dive right in to the news. Jensen, NVIDIA is making a $100 billion investment in OpenAI. [00:05:35] Speaker 1: As a result, rumors about circular deals began circulating heavily in the news. However, the problem goes even further. An estimate from Goldman Sachs suggests that there are currently 7.7 gigawatts of data centers in operation worldwide, far fewer than what NVIDIA claimed to have distributed in 2025 alone. While some experts suggest that these estimates take into account projects currently under construction and future projects, even then, Jensen Huang's claims appear to be enormously exaggerated. And examples of this continue to come to light. [00:06:07] Speaker 10: According to Heatmap, 25 data contracts were canceled in the last quarter of 2025. And 99 of the 770 planned data centers it checked that year are now being contested. [00:06:18] Speaker 1: In April 2026, Bloomberg reported that Oracle and OpenAI's flagship data center in Texas had postponed its expansion plans. The fact is that even if all the data centers in the world replaced their current GPUs with new ones every year, NVIDIA's graphics card production would still far exceed demand. Now, adding the fact that a large portion of data center construction projects are being canceled or delayed, the estimates become even more dramatic. But it gets worse. Because even if all future construction projects are completed successfully and none are canceled, NVIDIA's production would still exceed demand. [00:06:54] Speaker 5: What we're seeing too is supply right now is constrained. And I'm sure you've heard a bit of this. And what's happening, and it's going to get exacerbated over time. So by 2030, our estimation is that that imbalance of supply demand could be up to 10 gigawatts. [00:07:10] Speaker 1: But perhaps the biggest problem NVIDIA and every tech company involved in data center construction projects are facing is the shortage of electrical components coming from China. According to Bloomberg, the biggest bottleneck facing data center construction is not necessarily the chips needed to run AI, but getting the electrical infrastructure required to support it. While components such as generators, power supplies, and connection cables represent less than 10% of the cost of building a data center, nothing can function until these components are installed. In addition, the price of essential components such as electrical transformers has doubled in just the last four years, causing projected construction costs to increase considerably. But it gets worse. Because the vast majority of these components generally come from countries like China, South Korea, or Mexico. Given uncertain tariff rates and the imposition of tariffs on most of these countries, this further increases previously estimated construction costs. [00:08:06] Speaker 11: Fears remain high of a possible escalation in President Trump's trade war after he threatened an additional 100% tariff on Chinese goods. [00:08:14] Speaker 1: Because of this, tech companies tried to stockpile as many of these components as possible before prices increased, even if they did not have the data centers built to install them in. The same thing is happening with other components such as cooling fans or power supply systems. Even if many of these companies do not currently need them, they buy them just in case. This largely explains the extraordinary spending these companies have made in recent years, even when many of these projects are not generating any kind of economic return. While Nvidia is benefiting enormously from this, since any product they release is guaranteed to find an AI company willing to buy it, all it would take is a small market adjustment to create a massive oversupply. But for now, this has been incredibly profitable for Nvidia. [00:08:57] Speaker 12: Nvidia's revenue reached another record of about $44 billion in the fiscal first quarter that ended in April. Their forecast was actually really good. That's a big reason why you're seeing the stock respond so strongly. [00:09:07] Speaker 1: The problem is that while Nvidia broke consecutive sales and profit records in 2025 on paper, there are also several figures raising alarms. Since 2024, Nvidia has quadrupled its inventory, going from close to 5 million units in inventory But if Nvidia is truly struggling to meet enormous demand, why is it storing so much product? This suggests that either Nvidia is finding it increasingly difficult to distribute its own chips and GPUs, or in a much more likely scenario, that even Nvidia is facing supply chain shortages and is therefore trying to secure as many components as possible in the hope that some AI company will buy them in the future. [00:09:51] Speaker 6: Industry reports suggest Nvidia has made plans to reduce production of its latest gaming GPUs [00:09:56] Speaker 1: And this is a major problem, because if Nvidia itself is anticipating supply shortages, then all those future data center construction projects will not have the necessary components to operate, and therefore, many of them will begin to be canceled or delayed, just as is happening today. [00:10:22] Speaker 7: We'll see you next time. At Economy Media, your opinion matters to us. Subscribe and let us know what you think in the comments below.