Five major US technology companies have accumulated a staggering $1.8 trillion in hidden, off-balance-sheet AI debt driven by artificial intelligence ambitions, according to latest reports. This figure now exceeds their combined official liabilities and represents an eightfold increase over the last four years. The debt stems from future payment obligations including long-term operating leases for data centers not yet built, bulk GPU supply contracts, and joint ventures. Under current accounting standards, these obligations are not required to be recorded as liabilities until assets are delivered or facilities become operational. As per Business Standard, this surge in hidden debt is creating new financial challenges for investors, making it harder to judge companies' true financial exposure.
Despite early AI revenue returns, surging infrastructure spending is putting free cash flow under significant pressure across major tech companies. According to Reuters analysis of LSEG consensus estimates, the five hyperscalers are projected to spend more on capital expenditures than they generate in free cash flow by 2027. The estimates show companies will generate roughly $340 billion more in annual operating cash flow in 2027 compared to 2025, but capital expenditure is forecast to increase by around $534 billion over the same period. This implies approximately $1.57 in additional investment for every extra dollar of operating cash flow, with consensus estimates for 2026 capital expenditure climbing from roughly $485 billion in January to about $730 billion in July. The spending outlook has become significantly more aggressive as companies continue expanding AI capabilities.
The current earnings boom is masking a significant depreciation wave that will eventually impact profitability. According to Investing.com analysis, the five biggest hyperscalers spent about $412 billion on capex in 2025, with estimates running to roughly $760 billion for 2026. However, the AI capex depreciation and amortization that these companies expect to recognize against all that spending in 2026 is only about $211 billion. This creates a substantial gap where companies are spending $760 billion and expensing only $211 billion, with the other $549 billion sitting on the balance sheet as deferred costs. For 2026, combined free cash flow at these five companies is projected to fall 91% to about $16 billion, while net income is projected to rise 25% to roughly $506 billion. As Investing.com notes, this represents a situation where a business can report half a trillion dollars of profit and throw off almost no cash in the same year.
The AI buildout is transitioning from a cost center to a profitable asset class, with compute scarcity creating new monetization opportunities. According to Investing.com analysis, Meta sits on approximately 7GW of compute capacity, with plans to double to 14GW at 60-70% utilization. At conservative rental rates, a gigawatt facility costing $30 billion to build generates approximately $14.5 billion in net income annually, representing a two-year payback period. The depreciation concerns fail market testing, as four-year-old GPUs still command rising rental rates, contradicting assumptions that hardware becomes worthless in year five. Anthropic has contracted more than 11GW of compute across four deals, with three of four agreements with hyperscalers including Google, AWS, and Azure. This represents the second monetization engine beyond direct AI development, with the compute-resale opportunity becoming a fundamental shift in the AI ecosystem.
The profitability of AI development is accelerating rapidly, with Anthropic's revenue run rate increasing from approximately $9 billion at end-2025 to more than $47 billion by mid-2026, representing roughly 80x growth in one year. According to Investing.com analysis, inference costs have fallen 40x since early 2024 while revenue per token declined only 9x, swinging quarterly gross profit from -$55 million to an estimated $1 billion+ by Q3 2026 - marking the first profitable frontier lab. Anthropic's four chip deals across Google, AWS, Microsoft/Nvidia, and AMD total more than 11 gigawatts, with vendors funding their own buyers. The economics demonstrate that monetizing under 1% of a single 2GW deployment supports approximately $30 billion in revenue, with enterprise token consumption still early on its adoption curve. This profitability breakthrough validates the entire AI stack economics and reprices the entire cycle upward.
