The Great Infrastructure Bottleneck: Why GenAI’s Next Phase is About Atoms, Not Bits
Or why genAI’s success now depends less on algorithm breakthroughs and more on solving fundamental infrastructure challenges. Companies and investors must rapidly evolve beyond traditional tech frameworks to succeed in this new reality where bits meet atoms at unprecedented scale.
In less than two years, Generative AI has toppled so much of the conventional wisdom around innovation, business, and finance that it can be hard to keep up. However, as some observers debate the short-term returns on GenAI and valuations, a far more sweeping transformation is happening around the globe that demonstrates just how profoundly this technology is redefining the analytical frameworks that investors have used to understand the world.
I made the case over the summer about the need to see the bigger picture and recognize that GenAI represents a profound moment of Discontinuity. There may be no starker example of this than the tidal wave of announcements involving tech giants placing massive bets on nuclear power.
In recent weeks, Alphabet signed a deal to purchase nuclear energy from multiple small modular reactors (SMR) to be developed by Kairos Power. Amazon announced a combination of contracts and investments to expand its access to nuclear power. And in a truly stunning move, Microsoft signed a 20-year power purchase agreement with Constellation Energy which will restart the Three Mile Island nuclear power plant — the first time a decommissioned nuclear power plant in the U.S. has been restarted.
These moves come just a few months after Alphabet and Microsoft disclosed that they had missed their decarbonization goals amid surges in power consumption at data centers. “As we further integrate AI into our products, reducing emissions may be challenging,” Alphabet said in its report.
When OpenAI caused a sensation in November 2022 with the public release of ChatGPT 3.5, who would have looked at this simple text prompt and guessed that it would turn the global conversation about climate goals on its head?
Two years later, here we are: GenAI and Climate — two of the single points of failure we track when it comes to testing companies’ resiliency — are now colliding.
While I have always been bullish on GenAI’s potential, there is no denying that implementing this technology on a massive scale will require a colossal increase in physical resources and infrastructure. This transformation will accelerate the battle to gain access to high-performance chips and more mundane assets like land, water, and energy.
All of these will become increasingly scarce. They create bottlenecks, driving up compute costs and making monetization even more urgent for GenAI companies (especially those building LLMs).
At the same time, this creates new investment opportunities that blend real estate, construction, energy optimization, advanced cooling technologies, and new clean energy sources, from solar to nuclear. As savvy investors try to seize this moment, GenAI is breaking down the silos between traditional asset classes and rewriting the rules for financing innovation in real-time.
Climate investors cannot ignore GenAI. GenAI investors cannot overlook its climate impact. To navigate this moment of Discontinuity, investors must rethink their frameworks for analyzing deals.
The Context
The investment frenzy around data centers is emblematic of the challenges and opportunities the GenAI arms race has unleashed.
Linklaters research firm reported that global data center investing during the first five months of 2024 reached $22 billion — a rate that should easily surpass the $36 billion invested in 2023. While North America accounts for 69% of the 2024 investment, Europe saw 29% of the deals — up from just 6% in 2022.
According to an Arizton research report, the construction market for data centers in the U.S. alone is projected to double in size over the next six years.
Why? A.I. Building on previous technological transitions such as mobile and cloud, GenAI represents the next era of potential demand growth and disruption for the data center industry.
Just within the data centers, there is need for innovation around cooling technologies, optimization for what is likely to be volatile demand, and new software and hardware that is most adapted to the needs of A.I. models.
This explosion in data centers has the secondary effect of causing a surge in energy consumption. Goldman Sachs projects that AI will cause power usage by data centers to increase by 160% by 2030. Not only will there be more data centers, but AI is more power intensive; a typical ChatGPT query uses 10 times the power of a Google search.
Training requirements are doubling every 6–12 months.
The impact of this trend reaches far beyond the tech industry. For almost two decades, annual electricity consumption in the U.S. was effectively flat, only growing 0.5 percent annually between 2001 and 2024. The U.S. Department of Energy forecasts that total energy demand could increase by 15% to 20% in the next decade.
There are some positive aspects to this crisis. To provide this power, data centers are pushing for the expansion of all types of energy sources — most notably climate-friendly sources, which in turn is rapidly expanding production of clean energy sources such as wind and solar and bringing down prices.
For instance, Amazon has backed Scottish Power’s wind farm project in the UK by agreeing to purchase its entire 50-megawatt output. When Microsoft announced this spring it would invest €4 billion in new centers in France, part of the deal involved Power Purchase Agreements for energy from alternative sources.
But it won’t be enough to feed GenAI’s appetite. There is an energy reality check.
With all of these pieces essential for building the infrastructure needed for GenAI, a new approach to investment in the different parts of this value chain is emerging.
“While investor interest in the AI revolution theme is not new, we believe downstream investment opportunities in utilities, renewable generation and industrials whose investment and products will be needed to support this growth are underappreciated,” according to Goldman Sachs analysts.
Data Center Deals
Data centers and the energy sector are now attracting a wide range of traditional and non-traditional investors, from growth capital to buyout to real estate to tech giants. These investors are scrambling to manage risks and return models that may differ from those of their classic core investment sectors.
“Increasingly complex structures are now being used on data center transactions to open up the market as widely as possible and attract even investors who have not historically invested in digital infrastructure,” wrote Linklaters Partner Rich Jones. “This ability for new types of investors to be involved in the data centre market in turn fuels further investment demand.”
Consider that PE giant Blackstone just entered the data center market three years ago with a $10 billion acquisition of data center operator QTS. During a Q1 2024 earnings call, Blackstone CEO Steve Schwarzman said the firm now has $50 billion worth of data centers in its portfolio or under construction and plans to acquire another $50 billion.
In December 2023, Blackstone created a joint venture with data center company Digital Realty to build four hyperscale campuses. More recently, Blackstone helped lead a $2 billion debt package for data center company Park Place Technologies.
Blackstone’s US energy consumption growth outlook is even more aggressive than federal agencies. Schwarzman said this summer that the need to provide power for these data centers will be a major contributor to a 40% increase in electricity demand in the US.
“We believe these explosive trends will lead to unprecedented investment opportunities for our firm,” Schwarzman said on a Q2 2024 earnings call. “I believe the consequences of AI are as profound as what occurred in 1880 when Thomas Edison patented the electric light bulb. While it took years to develop commercially viable products, the subsequent build-out of the electric grid over the following decades parallels the creation of data centers today to power the AI revolution…Blackstone is positioning itself to be the largest financial investor in AI infrastructure in the world as a resue to our platform, capital and expertise.”
Blackstone is far from alone in recognizing this opportunity. The rush to build AI infrastructure has catalyzed deals with atypical investment partners:
· There has been a big wave of data center M&A as various players seek to gain scale.
· KKR portfolio company CyrusOne, closed a $7.9bn warehouse credit facility for data center infrastructure and development projects.
· DigitalBridge Group, a digital infrastructure investor with $80bn AUM, partnered with Silver Lake in a $9.2bn investment in hyperscale developer Vantage Data Centers.
The Power Play
This massive upheaval has vast implications for society, politics, and the global economy. However, more immediately, it also creates even more pressure on GenAI companies.
Already, tokens — one of the main sources of revenue — are being commoditized. Players such as OpenAI have access to additional sources of investment capital. But that won’t be the case for everyone building LLMs. So the need to monetize these platforms will take on even greater urgency as issues like energy consumption lead to a spike in costs.
Meanwhile, the conversation about nuclear has shifted with stunning speed. Just a couple of years ago, the industry was largely a global pariah. Now, it is being embraced by governments and private investors with extraordinary gusto. Axios reported in August that private and public investment in U.S. nuclear energy companies had already reached $14 billion in 2024 — double the amount in 2023. Those private investments include Open AI CEO Sam Altman who has invested in nuclear startup Oklo.
On the public side, Taiwan has declared it will pursue an ambitious nuclear agenda. More than a decade after the Fukushima disaster soured the nation on nuclear power, Japan has done an about-face and plans to double the amount of power generated by nuclear power by 2030.
France, long one of the biggest producers of nuclear power, reversed course several years ago from shutting down nuclear plants to launching a €1bn nuclear innovation program.
Beyond power limitations, and data centers, there are still huge infrastructure constraints. Power grid capacity limits, aging transmission infrastructure and cooling technology limitations are just some of them.
Consider this timely analysis from S&P Global. The analysts note that surging demand for data centers is placing immense pressure on the U.S. power grid, with anticipated incremental demand of 150–250 terawatt hours by 2030 necessitating significant investments in both power generation and transmission infrastructure. Aging grid infrastructure and lengthy planning and permitting processes pose major challenges, while cooling technology limitations further constrain capacity.
So, who will pay for all of this? The cost of the necessary infrastructure, estimated at $60 billion for power generation and $15 billion for transmission, according to S&P Global, will likely be borne by a combination of investments from power generation companies and public utility commissions. Additionally, data center companies may contribute through long-term power purchase agreements (PPAs) and investments in co-located energy facilities, while consumers could face higher energy prices as these costs are passed down through grid tariffs and market rates.
Naturally, the tech industry is trying to find innovative ways out of this bind. For example, Crusoe Energy has placed itself at the nexus of these two bottlenecks: energy and data centers. The company has developed data centers that run on renewable energy sources like wind, solar and geo-thermal, as well as capturing wasted natural gas from oil fields.
Just last week, Crusoe Energy raised up to $500 million in deal led by Peter Thiel’s Founders Fund, valuing the company at approximately $3 billion. It is now at the forefront of a whole new industry dubbed “neocloud” that specializes in providing outsourced cloud computing infrastructure, particularly for AI.
Tech leaders argue that ultimately the tradeoffs between GenAI and Climate will be reconciled. “Artificial intelligence will make it easier to combat climate change,” Bill Gate said in an interview this summer. Former Google CEO Eric Schmidt, speaking at the Special Competitive Studies Project’s inaugural AI+Energy Summit in early October, echoed this sentiment and said regulations or roadblocks to building new energy infrastructure would be counterproductive to climate reduction goals.
“My own opinion is that we’re not going to hit the climate goals anyway because we’re not organized to do it,” Schmidt said. “I’d rather bet on AI solving the problem than constraining it.”
That view is bound to be controversial.
“For the love of all that is beautiful on this planet, I hope that we think extremely carefully before going literally all in with the Earth on an unproven technology,” author Gary Marcus wrote in a critique of Schmidt’s remarks. “Chatbots and movie synthesis machines (perhaps the most energetically costly systems) are fun, maybe even amazing — but I don’t think we should be risking our planet for them.”
This all adds up of a dizzying array of variables and complex tradeoffs for startups and investors to goes well beyond the balance sheet to include a fast landscape of factors such as policy that lay beyond their control.
Consider the scoop that Business Insider just published about Amazon. According to an internal memo BI obtained, Amazon’s aggressive data center plans “are increasingly running into physical constraints that are slowing the buildout…At Amazon, electricity shortages have been a major concern and a topic of almost every leadership meeting in recent years, according to the people familiar with the matter.”
Internal documents indicate that across its network of data centers, the company is facing power supply issues and shortfalls that could last through 2030. This has introduced new unpredictability into its financial and strategic planning. It has also given rise to what BI dubbed “zombie” data centers: facilities built but not operational due to a lack of power.
In the memo, Amazon insiders wrote: “Across the [Americas] region, we are experiencing headwinds in power, zoning and permitting, water, and workforce/labor that are providing challenges to our long-term capacity growth.”
If this is squeezing Amazon, a company with almost limitless resources, then how does this ripple across the entire genAI value chain? Understanding how all of these pieces now fit together and what they tell us about AI business models goes to the core of the work we’re doing at D’Ornano + Co.
Creating the right analysis is critical not only to helping investors improve returns but also to ensuring the necessary funding structures emerge to allow this revolutionary technology to scale and reach its full potential.
