Google raised its 2026 capital expenditure projection to as much as $205 billion in its late-July earnings report, up from a prior top-end estimate of $190 billion. The new floor of the range, $195 billion, exceeds the previous quarter's ceiling. As of July 2026, Google was spending more money than it was making. The Verge

The figures landed in a week already heavy with investor anxiety over AI infrastructure spending. On July 23, 2026, Wall Street indexes closed lower, with the Nasdaq sinking, as big tech earnings revived concerns about the scale and pace of AI capex — shorthand for capital expenditure, the money companies spend on physical assets like data centers and servers. Bloomberg's "The Close" program that same day was titled "Stocks Hit by AI & War Jitters." Two days earlier, Reuters reported that the AI investment boom was putting Big Tech's free cash flow — the cash left over after a company pays for its operating expenses and capital investments — under pressure. Reuters Reuters

Google's competitive environment compounded the pressure. During the July 2026 earnings period, the company faced pricing pressure to keep model costs low alongside competition from Chinese AI tools. Chinese startup Moonshot released a new model, Kimi K3, in late July 2026. China faces constraints on GPU access that at least theoretically limit its AI development capabilities, yet the timing of Moonshot's release shows that those constraints have not prevented model launches. The Verge

Meta, Amazon, and Microsoft were all scheduled to report earnings in the same week as Google. Expectations were high that each would also announce higher-than-expected data center spending. The Verge

The scale of spending across the largest cloud providers — often called hyperscalers because of the massive scale at which they operate — is staggering in aggregate. Bridgewater stated in February 2026 that Alphabet, Amazon, Meta, and Microsoft were expected to collectively invest about $650 billion in AI during the year. A separate projection placed total hyperscaler capital expenditure at $725 billion for 2026. Analysts expected Alphabet and Amazon to burn cash in 2026, and Meta's cash flow was projected to shrink 95.7% to just $1.85 billion. Reuters Reuters

Microsoft's own capex trajectory illustrates the pattern quarter by quarter. In FY2026 Q1 (ending September 2025), capital expenditures were $34.9 billion, driven by cloud and AI demand. The following quarter, Q2 (ending December 2025), capex reached $37.5 billion, with roughly two-thirds spent on short-lived assets, primarily GPUs and CPUs. In Q3 (ending March 2026), capex declined to $31.9 billion, which Microsoft attributed to normal variability from cloud infrastructure buildouts and equipment delivery timing. Microsoft's Q2 revenue was $81.3 billion, up 17% year-over-year, with diluted EPS of $1.02. For comparison, in FY2025 Q2, capital expenditures including finance leases were $22.6 billion, and cash paid for property, plant, and equipment was $15.8 billion. Microsoft Investor Relations

Amazon, for its part, expected to invest about $200 billion in capital expenditures company-wide in 2026, and had announced an investment of up to $50 billion to expand AI and supercomputing capabilities for AWS U.S. government customers. Amazon's Q1 2026 net sales increased 17% to $181.5 billion, and the company guided Q2 2026 operating income between $20.0 billion and $24.0 billion, compared with $19.2 billion in Q2 2025. Amazon Investor Relations

The strain extends beyond the hyperscalers themselves. Oracle's capital expenditure in fiscal 2026 came to 174% of operating cash flow, meaning the company spent more on infrastructure than it generated from its core business by a wide margin. Its credit risk hit a near 18-year high in July 2026 due to investor concern about its AI datacenter debt load. Oracle is considered the public market's stand-in for OpenAI, which is not itself publicly traded. SpaceX shares, meanwhile, had fallen to nearly half their peak value as of late July 2026. The Verge Reuters

Nvidia was engaged in rounds of deal talks worth a combined $750 billion as of July 2026. Among those, Nvidia guaranteed OpenAI's debt in a deal worth $250 billion. Billy Leung, Global X Management's tech sector investment strategist, characterized the guarantee as "as much a reminder of funding strain in the AI build-out as it is a demand signal." The Verge

Investor Michael Burry publicly highlighted rising AI spending by Big Tech, amid what was described as Wall Street's fading appetite for AI capex. Yahoo Finance

What makes this moment distinct from earlier infrastructure buildouts is the velocity of depreciation working against the spenders. Depreciation is the accounting process by which the cost of a physical asset is gradually expensed over its useful life. Microsoft disclosed that roughly two-thirds of its Q2 FY2026 capex went to short-lived assets, primarily GPUs and CPUs. These are not 20-year data center buildings or long-haul fiber. They are accelerator cards on roughly three- to five-year replacement cycles, being purchased at unprecedented scale. The depreciation curve means that the financial return on each tranche of capex must materialize faster than in any prior infrastructure cycle the industry has undertaken. When capex is financed with debt, as Oracle's 174% operating-cash-flow ratio makes plain, that compression becomes acute.

The competitive dynamics on the model layer add a second pressure vector. Google faces pricing pressure to keep model costs low at the same moment its infrastructure spending is accelerating. Chinese competitors like Moonshot are releasing capable models despite GPU access constraints, which limits the pricing power of any single provider. The combination of rising capex, falling or compressed free cash flow, shorter asset lifespans, and pricing pressure on inference — the process of running a trained AI model to generate outputs — creates a set of conditions where the gap between spending and revenue must close through AI-driven monetization that has not yet arrived at the required scale.

Not every signal pointed downward. Lockheed Martin's stock rallied on July 23, 2026, after the company lifted its 2026 forecasts, suggesting that investor capital is rotating toward defense and away from speculative AI infrastructure. And on the demand side, the sheer volume of Nvidia's deal pipeline, $750 billion in combined talks, indicates that the buildout is not slowing. The question is whether the customers on the other end of those deals can service the economics. Reuters The Verge

In my view, the industry is in a phase where capital allocation decisions made over the next two to three quarters will determine which companies emerge with durable AI-platform economics and which are left with depreciating GPU fleets and debt service obligations that outpace the revenue those assets generate. The technology is real and the demand is real. Whether the unit economics close in time is the open question, and Google's own cash-burn status suggests that even the best-positioned operators are not yet confident they have the answer.