Is AI About to Blow a Fuse?

AI’s explosive growth comes with a hidden cost: staggering capital expenditures, soaring energy demands, and concentrated data center risks. Economist Michael Normyle joins Jeff Praissman to break down whether artificial intelligence is powering progress — or overloading the system.

Summary – IBKR Podcasts Ep. 298

The following is a summary of a live audio recording and may contain errors in spelling or grammar. Although IBKR has edited for clarity no material changes have been made.

Jeff Praissman 

Hi everyone. This is Jeff Praissman with the Interactive Brokers Podcast. It’s my pleasure to welcome back, for our monthly podcast on the economy, NASDAQ’s U.S. economist, Michael Normyle. Hey Michael, how are you? 

Michael Normyle 

Doing well, thanks. Glad to be back. 

Jeff Praissman 

Oh, it’s always great to have you in the studio. And today we’re going to talk about something that’s really hot on everyone’s mind. It’s been a big topic for quite a while, but since you’re an economist, I think we’ll take a slightly different angle on it than many other podcasts. Today we’re going to talk about AI — but more specifically, the capital expenditures of AI, not really its capabilities. So again, we’ll have that economic twist to it. 

To kick it off: the so-called “hyperscalers” — Amazon, Meta, Microsoft, and Google, all household names — are projected to increase their annual capital expenditures from $150 billion in 2023, which already seems like a crazy number, to potentially $400 billion by next year. What does this unprecedented level of investment tell us about their confidence in AI’s future profitability, despite the current limited returns? 

Michael Normyle 

Yeah, I think first you could say it reflects that AI is a very capital-intensive industry, requiring chips, data centers, energy, and billion-dollar training runs for models. But more broadly, maybe it’s a reflection of a “winner-takes-all, or possibly winner-takes-most” mentality. Just the way we’ve seen dominant search engines, browsers, and social media platforms, I think there could be a first-mover advantage here. The first company with the best AI product could potentially dominate the market. That could be the view we’re seeing — maybe being the first to AGI, artificial general intelligence, meaning AI matches human intelligence, or at least the first to the best general model. That remains to be seen. 

But unlike browsers and search engines, the expectation is that consumers and especially businesses will be willing to pay for AI. Of course, the risk is that they’re wrong. A recent survey found that only 3% of consumers pay for AI at the moment. We’ll see if that changes over time. 

Jeff Praissman 

Yeah, that leads me perfectly into my next question. MIT recently reported that 95% of companies are seeing zero return on their AI investments so far. How should investors distinguish between companies that are likely to eventually monetize AI successfully versus those that might be caught in an unproductive spending cycle? 

Michael Normyle 

I’ll start by saying I’m not an investment advisor, so I can’t give real advice here, but I’ll do my best with my economist lens. It’s also important to note that the MIT study was small — just 300 AI projects. They found the biggest cause of failed deployment was actually a learning gap. Companies need to learn how to best integrate these tools, and people need to learn how to use them. So based on that, it’s hard to answer the question just yet. If I had to take a stab at it, it’s probably best to focus on the narrow areas where AI is already really effective — things like coding, writing, and even figuring out protein structures.

But as workers get more comfortable with AI, and as the technology improves, it might become more challenging to suss out which applications are likeliest to succeed as they broaden. That could be a real challenge for investors. 

Jeff Praissman 

About 15 years ago, tech and AI investment accounted for about 2.5% of U.S. GDP. It now accounts for about 6%. How sustainable is this trend, and what economic vulnerabilities might emerge if AI fails to deliver on the productivity promises we all hope it does? 

Michael Normyle 

I’m not too concerned about sustainability at the moment. I think this reflects the changing economy in large part. And this is a pretty broad group of investments we’re looking at: software, computer and communication equipment, power and communication structures — all the elements needed to power data centers and the like. So it’s not strictly AI; it’s “tech and AI.” That’s what makes up the 6% number. But of course, AI has become an increasing driver in recent years. 

If we drill down to information processing equipment — computers, servers, etc. — it was actually a bigger share in the 1980s and 1990s as PCs rolled out. It peaked at around 3% of GDP in the early 2000s during the internet boom. Right now, it’s about 2% of GDP. So there’s room for further investment relative to the size of the economy. 

Before worrying about sustainability, though, I think the bigger concern is productivity promises. At its core, GDP growth is labor force growth plus productivity growth. Like many countries, the U.S. has an aging population. We’ve also seen less immigration lately. So labor force growth is slowing. In fact, the CBO just moved forward the timeline for U.S. deaths exceeding births to 2031, from 2033. That means faster productivity growth is needed to keep the economy growing. Economic growth is also key to debt sustainability in the U.S. If AI doesn’t deliver productivity growth, we could run into issues. 

Jeff Praissman 

And just to emphasize how much AI capital expenditure matters — U.S. real GDP growth for the first half of this year would’ve been just 0.2% without it. We’ve talked in the past about different sectors. How concerned should we be about this dependency on a single sector for economic growth? 

Michael Normyle 

Honestly, I don’t see it so much as a concern as a comfort. Typically, consumer spending is the engine of the U.S. economy. But lately, we’ve seen a timely structural uptrend in AI investment that helps offset weakness elsewhere. 

For example, consumer spending faltered a little in the first half of the year as households were cautious, waiting to see how tariffs played out. And then you have a government purposefully reducing spending, which is still a drag on growth. In that context, I think it’s good that we have this multi-year trend in AI investment to help smooth out rough patches in the economy. 

Jeff Praissman 

You talked about the internet boom earlier, but what other parallels or differences do you see between today’s AI investment boom and previous investment cycles — like the dot-com era or even the smartphone revolution? 
It’s hard to imagine now that we didn’t always have smartphones in our hands, but there was a time not too long ago when they didn’t exist. 

Michael Normyle 

I think the internet is the closer parallel. To me, it presents more of a business case than smartphones. Like I mentioned before, during the first internet boom we saw tech equipment investment rise to 3% of GDP, which we haven’t seen since. But I think it’s realistic to think AI could get us back there. Smartphones, of course, have a business case for phone makers and app developers, but the end use is really a consumer story. Even though, like you said, everyone has a phone, it hasn’t been a big productivity driver. Hopefully that’s the difference with AI. 

Jeff Praissman 

The Congressional Budget Office has presented two dramatically different debt scenarios based on productivity growth. How realistic is the optimistic scenario where AI helps stabilize U.S. debt at 113% of GDP, rather than letting it balloon to 156%? 

Michael Normyle 

The 156% path you mentioned is their baseline. That assumes 1.1% annual productivity growth out to 2055. The optimistic case is that AI boosts productivity growth to 1.6% per year. In that scenario, debt stabilizes at 113%. Debt is about 100% now, so it would rise to 113% and plateau there. 

But there’s also a downside scenario: productivity growth at just 0.6% per year, which would double the debt-to-GDP ratio to 200% by 2055. For context, over the last 10 years the U.S. has averaged 1.1% total factor productivity growth. So the CBO is essentially assuming that continues. Given that, I do think it’s realistic we land somewhere between the baseline and the upside scenario. Even though there aren’t signs yet of AI materially impacting productivity growth, recent research suggests a wide range of outcomes. On the low end, MIT’s Daron Acemoglu suggests a 1% boost to productivity in 10 years. On the high end, a Brookings study estimates a 20% gain in 10 years. 

Given the learning curves we talked about earlier, and still-low adoption rates — only 8–12% of companies are using AI according to recent surveys — it will likely take a few years before we see a noticeable productivity boost. But I think there’s a reasonable case that we could get closer to that optimistic path, if not all the way there. 

Jeff Praissman 

You also mentioned consumer spending earlier, and we’re seeing that slow while AI capital expenditures rise. That’s a shift in economic drivers. How does it affect employment patterns, wage growth, and even wealth distribution in the coming years? 

Michael Normyle 

I think the slowdown in consumer spending is temporary. We’ve seen tariffs pushing through a little, which is raising inflation. Core goods inflation is showing that. But it looks like a temporary boost to inflation from tariffs. That could also cause a temporary drop in real wage growth to near zero. But early next year, tax cuts will help offset that drag, and the tariff effect on inflation will fade. So we could start to see positive momentum for spending. Early next year, there’s a scenario where both consumer spending and AI CapEx grow together. 

Given that, I don’t think AI will massively impact employment patterns or wage growth in the next couple of years. There is some research from Stanford economists showing that employment for young workers in AI-exposed industries — like software engineering and customer service representatives — has been falling. Even after accounting for other factors, like the Fed’s 2022 rate hikes cooling the labor market, there’s still an element that seems related to AI. At the same time, the research also found faster job growth in industries augmented by AI. One key question is whether an industry is automated or augmented by AI. Automated industries could see less hiring and shifts in work scope, which may impact wage growth. But in augmented industries, employment growth could pick up, and we might even see wage premiums develop. 

Jeff Praissman 

So basically, either AI helps you be more efficient — or it takes you out of the picture for your job. 

Michael Normyle 

Yeah, possibly. 

Jeff Praissman 

I want to pivot here. With so much capital being directed to AI infrastructure, are we potentially putting too many eggs in one basket and missing out on other critical areas of the economy — or even on technological developments we can’t see yet because of this tunnel vision on AI? 

Michael Normyle 

The bulk of the investment is in private industry, so it’s not clear that means we’re underinvesting elsewhere. A lot of infrastructure investment, for example, is public. On that, you can look at the American Society of Civil Engineers’ infrastructure report card for the U.S., which shows things moving in the right direction, though with room for improvement. AI is also forcing upgrades to energy infrastructure, which has been underinvested in. Given AI’s power demands, it requires a comprehensive approach to energy generation. We’re seeing revivals of nuclear facilities and new investment in small modular nuclear reactors. There are also advancements in energy technology more broadly. 

So by necessity, AI may drive positives in modernizing energy infrastructure and energy production. The challenge will be balancing AI’s energy demands with environmental concerns — and that’s where green energy investments could play a big role. 

Jeff Praissman 

Yeah, I was going to say, one of the things with AI — and with a lot of other technologies that flies under the radar — is just the pure amount of energy consumption these advances require. You really just touched on my next question, about the environmental concerns, energy constraints, and potential new investments in cleaner and greener energy technologies going forward. 

As AI expenditures continue to rise, we’re also seeing significant regional concentration in data center development. How might this geographical clustering impact grid stability in those areas, as well as real estate markets and local economies? I know Nevada and Virginia are hotspots, probably along with some others. But that’s something the average person isn’t necessarily thinking about, right? They just see what AI does, not the practical aspects of making it work. 

Michael Normyle 

I think if you’re outside those markets, it’s probably not front of mind. But there are already over 600 data centers in Virginia — double the number in Texas and California, which are in second and third place. From the research I’ve done, grid stability is definitely a concern. And it’s not just data centers. They’re coming online at the same time EVs are becoming more popular, and there’s growing demand for air conditioning. Multiple factors are colliding to impact grids. So this is something that needs to be carefully planned for when constructing data centers, especially large clusters of them. 

In terms of real estate, like you mentioned, Virginia and Nevada are pretty different states in terms of density. On the coasts, where areas are more densely populated, data centers are often built on formerly rural land, which can increase property prices. In Nevada, where they tend to be more remote, the impact might be smaller. For local economies, data centers aren’t very labor-intensive to operate. You get a boost to construction jobs during the build — potentially a thousand workers or more, since they’re large structures — but once completed, they typically employ only 100–200 people. So, not a huge long-term driver of jobs, though there is some creation. Recent research from Carnegie Mellon and NC State estimated that electricity bills could rise as much as 25% by 2030 in data center hotspots, compared to 8% nationwide. On the flip side, data centers are a big revenue source. One county in Virginia reported that for every $1 invested in data centers, it receives $26 in tax revenues, helping fund schools and other public interests. 

Jeff Praissman 

Michael, I just want to thank you for stopping by the IBKR Podcast Studio. For our listeners, you can find more from Michael and Nasdaq on our website under education, webinars, podcasts, and articles. You can also visit nasdaq.com to see their own education materials and great articles by Michael and other economists. Thanks again, Michael — appreciate you stopping by.

Michael Normyle 

Thanks.