Mark Zuckerberg, CEO of Meta Platforms, Inc.. If you are riding the wave of the AI boom, business is good. This could be because you supply components to Nvidia—as CEO Jensen Huang highlighted during his keynote at the company's GTC conference on Monday, that business is booming. Alternatively, you might be a small data center operator securing contracts from tech giants in urgent need of additional computing power.
Take the Dutch data center firm Nebius as an example. The company announced on Monday that Meta Platforms, Inc. has agreed to lease cloud service resources from it over the coming years, in a deal valued at up to $27 billion. Following the news, Nebius's stock price surged by 15%.
This agreement also signals that, despite Meta's plan to spend $125 billion in capital expenditures this year to expand its AI infrastructure, the company currently faces an immediate need for more computing capacity. This has led it to turn to external cloud providers to supplement its own facilities.
Meta's spending on external cloud services may warrant closer attention. This is not information the company has hidden: back in January, Meta warned that its operating expenses—which, unlike capital expenditures, directly impact the profit and loss statement—are expected to increase by 40% this year, reaching $169 billion.
Some of this increase stems from the costs of hiring AI talent and depreciation on owned servers, but the company explicitly cited "third-party cloud services spending" as a significant factor. (In simple terms, this refers to payments made by enterprises for using cloud services such as those provided by Nebius, Amazon, or Google Cloud.)
Similarly, in a securities filing at the end of January, Meta disclosed that, as of December 31st of last year, it had $131 billion in contractual commitments, "primarily related to third-party cloud compute agreements," in addition to its own investments in server and network infrastructure.
A year earlier, its total contractual commitments stood at just $32.8 billion, which were mainly tied to owned servers at that time.
Why should others besides Nebius shareholders pay attention to this shift?
Even if you are not a tech expert (though some knowledge helps), it is widely understood that the trade-off between building in-house data centers and using public cloud services has long been debated. The general consensus is that external cloud services are more expensive but avoid the massive upfront capital outlay required for building one's own data centers.
Shareholders of Meta Platforms, Inc. should clearly be concerned about how the company is allocating its AI-related spending.
Meta's operating margin is already on a downward trend—it fell to 41% in the fourth quarter of 2025 from 48% a year earlier. Data from S&P Global Market Intelligence shows that Wall Street analysts expect the full-year operating margin for 2026 to decline further to 34.8%. This context helps explain why reports emerged last week that Meta is considering cutting up to 20% of its workforce. The added cost of increased cloud services spending must be offset somewhere.
Nvidia's Trillion-Dollar Forecast
At last year's Nvidia GTC conference, Jensen Huang mentioned that Nvidia anticipated sales of its latest-generation AI chips, Blackwell and Rubin, would reach $500 billion between 2025 and 2026. This statement initially caused confusion among analysts and journalists.
For a company that reported annual revenue of just $130 billion for the fiscal year ending January 2025, this was an exceptionally large figure.
At the time, it was not entirely clear what exactly the $500 billion referred to or which period it covered. Morgan Stanley's report on the GTC event included a lengthy section dedicated to "interpreting the $500 billion statement."
It was eventually clarified that the figure covers the calendar years 2025 and 2026, even though Nvidia uses a fiscal year ending in January, rather than a calendar year.
Despite the earlier confusion, Huang offered another significant forecast during his Monday keynote at the GTC conference in San Jose, California.
In a speech lasting over two hours, he revisited the previous $500 billion projection and stated that Nvidia now expects Blackwell and Rubin chips to generate "at least $1 trillion" in revenue. An accompanying slide indicated that this figure covers the years 2025 to 2027 (assuming the same calendar-year basis as before). In other words, an additional $500 billion in scale is projected for this year alone.
Other Key Developments
According to Reuters, citing four informed sources, China's second-largest chipmaker, Huahong Group, has developed 7-nanometer chip manufacturing technology for use in AI chips.
Intuit stated in a regulatory filing on Monday that its founders and executive team have canceled future stock sale plans. Concurrently, Intuit plans to "significantly accelerate" the remaining $3.5 billion of its stock repurchase program, having already bought back $1.8 billion in shares over the past two quarters.