AI infrastructure investment is expanding beyond GPUs into power generation, fiber optics, memory, GPU leasing and chip connectivity. Nvidia (NVDA) has been the most direct beneficiary of the spending cycle in recent years, but as major technology companies continue to increase data-center budgets, other supply-chain bottlenecks are gaining pricing power.
The demand surge comes after decades of underinvestment in parts of the US and global physical infrastructure. In 1990, the US produced about 37% of the world's semiconductors; by 2020, that share had fallen to 12%. Power-generation assets, grid equipment, large-scale industrial manufacturing, advanced packaging and precision optics have also faced outsourcing or insufficient capacity. Production concentrated in Asia and other regions helped reduce costs, but the rapid expansion of AI infrastructure is exposing constraints across the supply chain.
Big Tech pushes capital spending higher
Combined capital spending by major technology companies on data centers and AI infrastructure has risen from about $142 billion in 2022 to roughly $600 billion in 2026, an increase of more than four times in four years. Alphabet raised its 2025 capital budget three times, from $52.5 billion in 2024 to as much as $93 billion. Microsoft's quarterly capital spending reached $34.9 billion, up 74% year over year, while Amazon expects 2025 capital spending of $125 billion, up 61%.
Goldman Sachs estimates that Meta Platforms, Microsoft, Amazon and Alphabet will spend a combined $5.3 trillion in capital expenditures from fiscal 2025 through fiscal 2030. PwC estimates that cumulative global data-center investment during the AI era could reach $31.6 trillion by 2050. Annual spending could rise from about $800 billion in 2026 to $1.1 trillion in 2030 and $1.8 trillion by mid-century.
Unlike some earlier infrastructure cycles, in which construction peaked before investment fell back, GPUs, servers, memory and networking equipment inside AI data centers generally need to be replaced every four to six years. That creates a recurring equipment-replacement market after the initial build-out, although the pace will depend on AI adoption, corporate budgets and the availability of components.
GE Vernova: power equipment becomes a constraint
AI data centers are constrained not only by chips but also by the time required to bring new power capacity online. GE Vernova (GEV) makes gas turbines and grid equipment used by utilities to expand generation and transmission capacity.
The company reported quarterly revenue of $11.1 billion. Organic orders increased 134% in its power business and 66% in electrification. Management expects its backlog, already above $160 billion, to exceed $200 billion by the end of 2027—one year earlier than the timetable given at the start of the year.
Microsoft has ordered seven GE Vernova gas turbines for a Texas data-center project developed with Chevron (CVX), and similar projects are increasing. GE Vernova's share price has already risen substantially, leaving investors focused not only on order growth but also on the pace of capacity expansion and how much future growth is reflected in the valuation.
Corning: fiber demand follows AI rack expansion
GPU servers in data centers need high-speed fiber connections. Corning (GLW), one of the world's major fiber-optics technology and manufacturing companies, reported core sales of $4.74 billion in its latest quarter, up 17% year over year. Operating margin increased by nearly two percentage points to 20.9%. The company's chief financial officer said sales directly linked to AI data centers had nearly doubled from the previous quarter.
Corning has expanded its fiber-manufacturing relationship with Amazon and signed a multiyear agreement with Meta to supply fiber connectivity for its next-generation AI data centers. The agreement has a potential value of up to $6 billion. Under the company's long-term plan, annualized sales could rise from about $20 billion in 2026 to $30 billion in 2028 and $40 billion in 2030. Delivery of those targets will depend on the pace of data-center construction, customer order execution and fiber-capacity expansion.
Micron: high-bandwidth memory remains tight
Memory chips are another part of the AI server supply chain where the balance between supply and demand has shifted sharply. Micron Technology (MU) has sold out its high-bandwidth memory capacity through the end of 2026. The company reported quarterly revenue of $41.5 billion, up 346% year over year, while gross margin rose from below 40% a year earlier to nearly 85%.
Samsung Electronics (SSNLF) and SK Hynix (SKHY) are Micron's two main global competitors in memory. Both expect broader memory tightness to persist through 2027 or later.
Micron shares rose nearly 340% in the first half of 2026 before falling more than 20% from their record high. The move indicates that expectations for the memory cycle are already substantial. Subsequent performance will depend on product pricing, new capacity, customer inventories and changes in AI server demand.
Nebius: renting GPU capacity to enter the market
Nebius Group (NBIS) does not manufacture chips or train AI models. Instead, it rents GPU clusters to customers through what the market often calls a “new cloud” service. The company reported quarterly revenue of $582 million, up 454% year over year, and posted a positive adjusted profit for the first time.
Nebius has signed a five-year infrastructure agreement with Meta Platforms with a potential total value of up to $27 billion. Nvidia has also invested $2 billion directly in the company, helping it secure priority access to future computing capacity.
For this business model, the main constraint is the speed of building computing clusters, not simply winning customers. Nebius must continue buying GPUs, building data centers and securing sufficient power before customer commitments can become revenue. As a result, investors are also watching capital spending, financing needs, equipment deliveries and customer concentration alongside the company's rapid revenue growth.
Astera Labs: data movement between chips becomes a bottleneck
As AI server racks carry more chips at higher speeds, moving data between those chips can become a limiting factor. Astera Labs (ALAB) develops connectivity and interconnect products for servers. Its latest quarterly revenue reached $392.4 million, up 104% year over year, with both revenue and profit above Wall Street expectations.
The company expects next-quarter revenue to be about 30% above analysts' previous forecast. Its latest-generation Scorpio X switch chip has entered volume production and could become the company's largest product line as soon as one quarter after launch.
After the earnings release, JPMorgan, Jefferies and Morgan Stanley each raised their price targets by at least 20%. Astera Labs is also the smallest and one of the more volatile companies among the group, leaving its valuation particularly sensitive to the ramp-up of new products and growth in customer orders.
Supply capacity will determine how demand translates into revenue
AI infrastructure investment has lifted demand across GPUs, servers, power, memory and fiber, but supply conditions differ from one segment to another. Nvidia's data-center revenue reached $194 billion in its latest fiscal year, up 68% year over year and roughly 13 times its level when ChatGPT was launched. Its gross margin remained above 75%. Server assemblers including Dell (DELL) and Super Micro (SMCI) have also benefited.
Nuclear power operator Constellation Energy (CEG) has signed long-term supply contracts at prices $30 to $40 per kilowatt-hour above prevailing regional rates. Adjusted earnings per share increased by more than 25% in the latest quarter. The contracts show how power constraints are beginning to reach energy suppliers, although new generation and grid equipment require lengthy construction timelines.
Share prices and valuations across these companies already reflect varying degrees of expected AI investment growth. Supply constraints should ease as new capacity comes online, while demand growth, replacement cycles, financing conditions and policy could all change the pace of capital spending. The market's focus is therefore shifting beyond the total size of AI investment to which supply-chain segments receive the next wave of funding—and whether companies can convert backlogs into revenue and cash flow.