The race to build AI data centers is shifting away from America, with Gulf nations offering dramatically lower costs and sovereign wealth backing that U.S. locations simply cannot match. While Meta announced plans to sell excess computing capacity and Nvidia launched revenue-sharing deals with startups, industry observers are questioning whether building massive data centers in the United States makes economic sense at all.

Why Are U.S. Data Centers Becoming Less Attractive?

The economics of building AI infrastructure in America are deteriorating, according to technology strategists analyzing the market. Mark Douglas, president and CEO of connected-TV ad platform MNTN, argues that U.S. data center capacity "is not going to age well" because it remains one of the most expensive places globally to build such facilities, and many American communities actively oppose their construction.

The real competition, Douglas explained, is emerging from an unexpected direction: the Gulf region, particularly Saudi Arabia. "The Kingdom of Saudi Arabia is coming online with massive data center capacity, hosted out of Saudi Arabia, at significantly lower prices," he noted. The logic is straightforward: instead of extracting oil and shipping it overseas in tankers, Gulf nations can use that energy to power data centers locally, creating a cost advantage that's difficult for U.S. operators to overcome.

Douglas

Some Gulf data centers are being structured as legal extraterritorial zones, effectively functioning as data "embassies." This arrangement allows multinational clients with strict data-residency requirements to use these facilities without technically moving data outside their home country, solving a major regulatory hurdle.

What Are Tech Giants Actually Doing With Their Compute Power?

Meta's announcement that it plans to sell excess AI computing capacity to outside customers generated significant market enthusiasm, with the company's stock jumping more than 7% on the news. However, industry experts view this move with skepticism. Douglas argued that the strategy doesn't align well with Meta's core business of serving billions of social media users; pivoting to selling raw data center space to perhaps 10 major customers represents a dramatic shift in scale and complexity.

"Going from 3 or 4 billion social media app users to 10 customers buying data center capacity from them, that just doesn't seem like a good fit," said Mark Douglas, president and CEO of MNTN.

Mark Douglas, President and CEO at MNTN

Meanwhile, Nvidia is taking a different approach. The chipmaker announced revenue-sharing agreements with fast-growing startups, allowing them to swap access to computing power for a slice of future profits. Two initial Australian partners are providing the compute infrastructure: Sharon AI will deploy up to 40,000 Nvidia GPUs (graphics processing units, the specialized chips that power AI training), while Firmus Technologies is building a data center in Batam, Indonesia, expected to scale to 360 megawatts and house up to 170,000 Nvidia GPUs.

This approach reflects a broader trend in the AI industry. Startups face severe liquidity challenges and struggle to access scarce GPU capacity, which has become as critical as oil to their operations. By offering revenue-sharing deals, Nvidia positions itself as an intermediary helping startups gain direct access to full-stack computing powered by its chips, while also securing long-term revenue streams.

How Are Startups Adapting to the GPU Shortage?

• Revenue-Sharing Agreements: AI firms and model builders are increasingly entering into deals where they share both product and cloud revenue with chipmakers like Nvidia to secure access to scarce GPU capacity without requiring upfront capital.
• Equity and Investment Partnerships: Companies like OpenAI have inked multiple deals involving share purchases and investments from partners including Amazon and AMD to circumvent liquidity issues affecting the sector.
• Geographic Diversification: Startups are gaining access to computing infrastructure across multiple regions, including emerging data centers in Indonesia and other locations, reducing dependence on any single geographic market.

The GPU shortage has become so acute that computing power is now likened to oil and has even been tied to futures contracts as users grapple with cost fluctuations and availability constraints.

What Does This Mean for Meta's Cloud Strategy?

Douglas remains skeptical that Meta's cloud computing pivot represents a genuine strategic shift. He compared it to Meta's previous attempts to enter new markets, such as Threads, which struggled to gain traction. The fundamental challenge is talent: the engineers and leaders who know how to build new products from scratch typically work at startups, not at large established companies like Meta.

Despite his doubts about the cloud business specifically, Douglas expressed optimism about Meta's broader AI strategy. He highlighted the company's open-source AI models, particularly Llama, as a genuine competitive advantage. "I'm actually pretty bullish long-term on Meta in terms of AI models and Llama overall," he stated. Meta's massive data resources and motivation to make models available at attractive prices could benefit both its advertising business and the advertising industry more broadly.

Douglas

"I think Meta has the capital, the desire, and the ability to recruit to ultimately do quite well, especially since they're building and training those models on massive amounts of data and are motivated to make them available at attractive prices," Douglas explained.

Mark Douglas, President and CEO at MNTN

The broader implication is that Meta's compute-selling business, if it materializes, would likely remain a tactical move rather than a fundamental strategic pivot. Douglas suggested that "all tactical moves have a beginning and an end," and this one would probably follow the same pattern.

Douglas

As the AI infrastructure market matures, the competitive landscape is becoming increasingly complex. U.S. data centers face mounting pressure from cheaper alternatives abroad, while companies like Meta and Nvidia are experimenting with different models to monetize and distribute scarce computing resources. The winners in this space may not be those building the most data centers, but rather those who can most efficiently allocate and distribute the computing power that already exists.