The artificial intelligence hype has been very good for semiconductor companies, but it’s making the working capital structure powering the technology increasingly complicated.
Revenue among U.S. semiconductor and equipment companies jumped 32% in 2025, according to The Hackett Group's latest working capital survey. Artificial intelligence infrastructure and data center buildouts helped drive the highest revenue growth among all industries analyzed in the report, as advanced chip demand continues to rise.
But getting all of that growth to turn into cash is taking longer, even as the AI hype has created enormous value for the industry's biggest players. Nvidia became the world's first $4 trillion company last summer, a run that also pushed CFO Colette Kress' net worth past $1 billion.
But underneath those soaring valuations and equity packages, the industry's cash cycle has been moving in the other direction.
The semiconductor industry's cash conversion cycle increased eight days, or 6%, last year. Days sales outstanding increased three days, or 6%, while days payable outstanding fell 12 days, or 14%. Inventory days decreased seven days, or 5%, but remained at roughly 152 days. Over the past five years, CCC has increased 36.8%.
Semiconductor companies are still putting billions of dollars behind expectations that today's demand will last. Taiwan Semiconductor Manufacturing recently committed another $100 billion to its Arizona manufacturing footprint, bringing its total planned U.S. investment to $265 billion.
How demand is stretching cash cycles
Data center semiconductor and component revenue jumped roughly 44% year over year during the second quarter of 2025, according to Hackett, citing data from market research firm Dell’Oro Group. Demand for graphics processing units and high-bandwidth memory, researchers said, helped increase utilization of advanced manufacturing nodes and supported pricing power for suppliers.
They added that Nvidia's GPUs, particularly its Blackwell-based units, represented a large share of semiconductor wafer demand tied to AI processors, while Micron Technology delivered record data center and high-bandwidth memory sales.
Some of the industry's biggest customers also have considerable leverage over when semiconductor companies get paid due to their size and purchasing power.
Hackett points to Nvidia shipping large AI accelerator orders to cloud providers and hyperscalers on "extended invoicing cycles that can be tied to deployment milestones." Greater customer concentration and larger contract sizes have contributed to DSO rising 23.3% over the past five years.
Inventory has become another cash trap for chipmakers, too. Hackett attributed elevated inventory levels to longer production cycles and "tariff-driven buffer inventory" as companies make their supply chains more resilient. Inventory as a percentage of revenue did decline 8% last year, helped by data center construction and faster chip manufacturing cycles.
Demand also depends heavily on what kind of chip a company makes. Hackett said PC and smartphone markets remained weak or went through inventory corrections. Micron moved to exit the consumer memory business to focus on AI and data center memory, while Intel prioritized data center CPUs amid production constraints.
Those dynamics make capacity decisions more important because new fabrication facilities can take years to build, meaning the billions committed today are bets on where semiconductor demand will be years from now.
The accounting behind the AI money machine
All of this growing AI infrastructure raises another set of questions that have become the focal point of how AI developers account for their money.
David Leary, co-host of The Accounting Podcast, recently discussed a proposed $500 billion AI infrastructure financing arrangement. He and co-host Blake Oliver said the structure would use special-purpose vehicles to own data centers and buy the chips needed to operate them, with Nvidia reportedly guaranteeing a portion of the debt.
"If I sell them chips and they have no place for them to work, they're not going to buy my chips," Leary said. "So, I have a vested interest in data centers now."
Oliver also pointed to money moving between AI developers and the large technology companies providing their computing capacity.
"The money just goes in a circle, and it books revenue on both sides," he said. Oliver explicitly said he wasn't alleging fraud and noted the arrangements are permitted under existing accounting rules. But he also raised questions about depreciation assumptions for AI chips.
"Estimates are that these chips are only good for two to three years because everything's changing so fast," Oliver said, noting some useful-life assumptions have been pushed toward five or six years. Longer useful lives spread depreciation over more years and reduce the annual expense recognized by the companies buying the equipment.
CleanSpark President and CFO Gary Vecchiarelli, whose company is expanding from bitcoin mining into AI infrastructure, has seen reasons for caution, too.
"Whenever there's a gold rush, there are people who think it's easy money," Vecchiarelli told CFO.com in a recent interview. "The important thing is to stay focused on the long game and avoid getting caught up in the excitement of the moment."
Vecchiarelli said CleanSpark has encountered AI opportunities where some companies appeared to be taking a “fake-it-till-you-make-it approach.” His company recently secured a 20-year, $6.6 billion lease to develop an AI data center campus in Georgia.
Geopolitics adds pressure
Semiconductor companies are also making investment decisions in an industry where governments have an increasingly large influence over where chips are made.
Hackett said U.S.-China trade tensions, as well as tariffs and export controls, have "structurally altered" semiconductor supply chains. The push toward domestic manufacturing has improved long-term resilience while increasing short-term capital intensity and working capital drag.
Hackett said those pressures have accelerated Intel's shift toward localized supply chains, increasing costs and inventory levels. Advanced Micro Devices' reliance on external foundries has encouraged sourcing diversification and additional buffer inventory, while trade restrictions have increased input costs at Applied Materials.
Nvidia is dealing with its own version of the problem on its end, as export controls and AI demand are pushing the company to diversify its supply chain while greater customer concentration increases its receivables exposure, according to Hackett.
