X

Mid-Year Report: AI Spending Soars as Data Centers Face Heat

July 15, 2025

This summer, we’re checking in on our TechArena predictions for 2025 to see how they are holding up.  

Today, TechArena correspondent Will Torresan sat down with Matty Bakkeren, independent growth consultant and founder of Momenthesis, to discuss what he got right in his predictions and what’s taken him by surprise.

You mentioned that AI would demand “new infrastructure, new expertise, and entirely new business models” for data center operations. Can you share a specific example of a change you’ve observed this year that exemplifies this prediction?  

I think Jensen Huang’s keynote from GTC Paris exemplified this. He said some very interesting things.  In particular, he discussed the complexity of AI infrastructure and how NVIDIA continues to push the envelope and expand the product lineup and software capabilities. Interestingly, as part of that discussion, the term “AI data center” seems to be replaced more often with “AI factories,” and Jensen is going so far as to refer to AI infrastructure as “thinking machines.”

The fact is that AI infrastructure requires significant investments in advanced infrastructure, specialized expertise, and innovative business models. The computing architecture is going to be different, as Jensen mentioned: AI needs a 40x speed-up. AI infrastructure, regardless of the name, sits in data centers, which are now going to be a critical part of a country’s infrastructure.

Your most ambitious prediction was data centers becoming “energy providers and critical heat sources.” What compelling real-world examples have you seen of this transformation in 2025?  

Heat re-use is certainly gaining traction, but implementing it is not a short-term project. It typically requires a consortium of municipality, utility, enterprise, and data center reps to build a solid business case. And it also needs investments and facilities to transport and re-use the heat outside the data center. Establishing such a platform and then figuring out how to build the infrastructure is complex. That’s why it was exciting that the Open Compute Foundation released a reference design for data center heat re-use, which provides a roadmap for organizations looking to implement heat re-use.

A few examples come to mind. Since implementation is slow, some of these did start before 2025:

In Finland, European data center developer atNorth recently signed an agreement for its FIN02 data center to provide heat to a neighboring retail store.

For the 2024 Olympics in Paris last year, Equinix exported heat to help supply the Olympics Aquatics Center.

And for a large scale, ongoing project example, the Tallaght District Heating Scheme in Ireland uses waste heat from a nearby Amazon data center to supply heat to new local authority buildings and more.

As adoption grows, the World Economic Forum is starting to take note and is now tracking these developments.

With regard to data centers becoming energy providers and helping utility companies to balance the grid, the developments seem to be early days. An article caught my eye earlier this year detailing how in Ireland both Microsoft and Digital Realty individually tried to have data centers supply energy back to the grid. Unfortunately, these projects failed, but good lessons were extracted. In my recent conversation with some of the engineering companies, I’m also seeing a trend where new energy storage products are also architectured to deliver power back to the grid. A recent example would be ABB’s battery as a service.

Looking at the economic and regulatory environment of 2025, what external factors have either accelerated or hindered the trends you predicted? Were there any policy changes or market shifts that significantly impacted your forecasts?  

This is a mixed bag. The continued stream of announcements and investments in large AI data centers and infrastructure this year has been impressive. To name a few, there’s the Stargate $500 billion AI infrastructure announcement, Meta announced a $200 billion AI data center project, and France announced a $112 billion AI investment.

At the same time, the cost of building a data center is going up from $10 million per megawatt to $12 million per megawatt, and the geopolitical climate is adding to that cost with uncertainty over the impact of tariffs. This and other geopolitical factors are already causing a shift in the landscape now that AI data center capacity has become a strategic piece on the chessboard.

To add to that, demand for data center space is up, with an all-time low vacancy rate for data centers in Europe. Power remains the biggest challenge to building new data centers in new Tier 2 locations as the Tier 1 markets are saturating.

What major trend or development happened in 2025 that wasn’t on your radar at all when you made your original predictions? What took the industry by surprise this year?  

The change in the geopolitical landscape has taken everyone by surprise. And it’s had a significant effect on the acceleration of data center demand. The AI race for supremacy has set off so many initiatives and funding, but more importantly, it’s giving an impulse to sovereign data and compute initiatives globally. For example, in late January, we had two major competing announcements from the US and China with the $500 billion Stargate AI investment and then DeepSeek revealing its cost-efficient models. Both announcements drive more computing demand in their own way. And now the European Union is watering down AI regulation and sustainability-related reporting and due diligence requirements to enable European companies to be more competitive on the global market.  

If you were writing predictions for the remainder of the year right now, what would be your boldest forecast based on what you’ve observed so far?

While there is an all-time hype for AI and geopolitical race for supremacy, I think there will be a stronger lobby AGAINST new data centers, which is going to counterbalance some of the hype. I’m already seeing signs it’s increasing, and again I think AI has accelerated the issue.  

Because AI is getting so much public attention, so is its power and cooling usage, which in turn is unfortunately making data centers the focus of attention. And that is the key issue: the power and natural resources data centers use to feed our digital hunger for AI. The main targets will be the hyperscalers as they are the biggest consumers of resources. It’s likely more regulation and compliance are coming to the data center space.  

I think to a degree that’s fair because companies need to be responsible with the resources used to build and run data centers, but I also feel not enough credit is being given to the data center industry that is already making efforts to minimize impact. We as an IT industry and society will need to find a balance and be more aware of the resources used to power our digital needs. A video by Bloomberg reporting about concern over data centers in northern Virginia provides a very interesting example.

Subscribe to our newsletter.

This summer, we’re checking in on our TechArena predictions for 2025 to see how they are holding up.  

Today, TechArena correspondent Will Torresan sat down with Matty Bakkeren, independent growth consultant and founder of Momenthesis, to discuss what he got right in his predictions and what’s taken him by surprise.

You mentioned that AI would demand “new infrastructure, new expertise, and entirely new business models” for data center operations. Can you share a specific example of a change you’ve observed this year that exemplifies this prediction?  

I think Jensen Huang’s keynote from GTC Paris exemplified this. He said some very interesting things.  In particular, he discussed the complexity of AI infrastructure and how NVIDIA continues to push the envelope and expand the product lineup and software capabilities. Interestingly, as part of that discussion, the term “AI data center” seems to be replaced more often with “AI factories,” and Jensen is going so far as to refer to AI infrastructure as “thinking machines.”

The fact is that AI infrastructure requires significant investments in advanced infrastructure, specialized expertise, and innovative business models. The computing architecture is going to be different, as Jensen mentioned: AI needs a 40x speed-up. AI infrastructure, regardless of the name, sits in data centers, which are now going to be a critical part of a country’s infrastructure.

Your most ambitious prediction was data centers becoming “energy providers and critical heat sources.” What compelling real-world examples have you seen of this transformation in 2025?  

Heat re-use is certainly gaining traction, but implementing it is not a short-term project. It typically requires a consortium of municipality, utility, enterprise, and data center reps to build a solid business case. And it also needs investments and facilities to transport and re-use the heat outside the data center. Establishing such a platform and then figuring out how to build the infrastructure is complex. That’s why it was exciting that the Open Compute Foundation released a reference design for data center heat re-use, which provides a roadmap for organizations looking to implement heat re-use.

A few examples come to mind. Since implementation is slow, some of these did start before 2025:

In Finland, European data center developer atNorth recently signed an agreement for its FIN02 data center to provide heat to a neighboring retail store.

For the 2024 Olympics in Paris last year, Equinix exported heat to help supply the Olympics Aquatics Center.

And for a large scale, ongoing project example, the Tallaght District Heating Scheme in Ireland uses waste heat from a nearby Amazon data center to supply heat to new local authority buildings and more.

As adoption grows, the World Economic Forum is starting to take note and is now tracking these developments.

With regard to data centers becoming energy providers and helping utility companies to balance the grid, the developments seem to be early days. An article caught my eye earlier this year detailing how in Ireland both Microsoft and Digital Realty individually tried to have data centers supply energy back to the grid. Unfortunately, these projects failed, but good lessons were extracted. In my recent conversation with some of the engineering companies, I’m also seeing a trend where new energy storage products are also architectured to deliver power back to the grid. A recent example would be ABB’s battery as a service.

Looking at the economic and regulatory environment of 2025, what external factors have either accelerated or hindered the trends you predicted? Were there any policy changes or market shifts that significantly impacted your forecasts?  

This is a mixed bag. The continued stream of announcements and investments in large AI data centers and infrastructure this year has been impressive. To name a few, there’s the Stargate $500 billion AI infrastructure announcement, Meta announced a $200 billion AI data center project, and France announced a $112 billion AI investment.

At the same time, the cost of building a data center is going up from $10 million per megawatt to $12 million per megawatt, and the geopolitical climate is adding to that cost with uncertainty over the impact of tariffs. This and other geopolitical factors are already causing a shift in the landscape now that AI data center capacity has become a strategic piece on the chessboard.

To add to that, demand for data center space is up, with an all-time low vacancy rate for data centers in Europe. Power remains the biggest challenge to building new data centers in new Tier 2 locations as the Tier 1 markets are saturating.

What major trend or development happened in 2025 that wasn’t on your radar at all when you made your original predictions? What took the industry by surprise this year?  

The change in the geopolitical landscape has taken everyone by surprise. And it’s had a significant effect on the acceleration of data center demand. The AI race for supremacy has set off so many initiatives and funding, but more importantly, it’s giving an impulse to sovereign data and compute initiatives globally. For example, in late January, we had two major competing announcements from the US and China with the $500 billion Stargate AI investment and then DeepSeek revealing its cost-efficient models. Both announcements drive more computing demand in their own way. And now the European Union is watering down AI regulation and sustainability-related reporting and due diligence requirements to enable European companies to be more competitive on the global market.  

If you were writing predictions for the remainder of the year right now, what would be your boldest forecast based on what you’ve observed so far?

While there is an all-time hype for AI and geopolitical race for supremacy, I think there will be a stronger lobby AGAINST new data centers, which is going to counterbalance some of the hype. I’m already seeing signs it’s increasing, and again I think AI has accelerated the issue.  

Because AI is getting so much public attention, so is its power and cooling usage, which in turn is unfortunately making data centers the focus of attention. And that is the key issue: the power and natural resources data centers use to feed our digital hunger for AI. The main targets will be the hyperscalers as they are the biggest consumers of resources. It’s likely more regulation and compliance are coming to the data center space.  

I think to a degree that’s fair because companies need to be responsible with the resources used to build and run data centers, but I also feel not enough credit is being given to the data center industry that is already making efforts to minimize impact. We as an IT industry and society will need to find a balance and be more aware of the resources used to power our digital needs. A video by Bloomberg reporting about concern over data centers in northern Virginia provides a very interesting example.

Subscribe to our newsletter.

Transcript

Subscribe to TechArena

Subscribe