Record Data Sheet
The AI infrastructure cycle is becoming a capital allocation race, with Amazon expecting approximately $200B of CapEx in 2026, Microsoft approximately $175B under its latest guidance, Alphabet $180B, $190B and Meta $130B, $145B, putting the combined midpoint at approximately $697.5B, although these figures represent corporate CapEx rather than AI-only spending. SpaceX is approaching the same infrastructure market from a different financial position: in June 2026, the company priced its inaugural bond offering at $25B, consisting of senior unsecured notes with maturities ranging from 2031 to 2056, creating a useful contrast between the scale of hyperscaler investment and the financing structure of a company building its own computing infrastructure.
The figures are not equivalent, and that distinction is central to the Budget War: Big Tech’s figure represents planned corporate capital expenditure, while SpaceX’s $25B represents debt raised, not an AI budget. The more relevant question is therefore not simply who is spending more, but who is putting capital at risk, how that capital is being financed and how quickly it can be converted into productive capacity and recurring revenue.
Big Tech’s Advantage: Scale
The hyperscalers enter the AI infrastructure cycle with operating businesses capable of generating substantial cash flow, allowing them to fund investment through a combination of internal cash generation, debt, leases and other financing arrangements while monetizing infrastructure across multiple businesses. Amazon expects approximately $200B of capital expenditures in 2026, citing AI, chips, robotics and other strategic opportunities among the principal investment areas, while the company generated $139.5B of operating cash flow in 2025 and said it expects strong long-term returns on its investments.
Microsoft’s latest disclosure puts its 2026 CapEx expectation at approximately $175B, following an accounting change affecting the treatment of certain future data-center leases, while Microsoft said the underlying investment expectation had not changed as a result of that accounting treatment. The company also reported that approximately two-thirds of its recent quarterly CapEx was directed toward short-lived assets, primarily CPUs and GPUs, illustrating how much of the current spending cycle is directly connected to computing capacity.
Alphabet expects $180B, $190B of CapEx in 2026, with the majority directed toward technical infrastructure, including servers, networking equipment and data centers, while Meta expects $130B, $145B, with its guidance including principal payments on finance leases. Taken together, the four companies represent an extraordinary concentration of capital expenditure, but the figures should not be described as $697.5B of AI spending because the companies use their CapEx for a broader range of infrastructure and strategic investments.
SpaceX’s Different Bet
SpaceX does not have the same financial model as the hyperscalers, and its $25B inaugural bond issuance should therefore be viewed as a balance-sheet financing operation rather than an AI-specific budget. The company issued $7B of notes due in 2031, $6B due in 2033, $6B due in 2036, $2.5B due in 2046 and $3.5B due in 2056, with coupons ranging from 5.350% to 6.650%, and SpaceX stated that the proceeds would be used to repay its bridge loan, cover associated fees and expenses and fund general corporate purposes.
That distinction matters because it would be incorrect to say that SpaceX raised $25B specifically to buy AI chips, but the financing sits within a broader expansion of the company’s computing infrastructure, and its SEC filings provide evidence that this infrastructure can also be monetized by external customers.
The Compute Revenue Test
SpaceX has disclosed agreements with Anthropic covering approximately 325,000 NVIDIA GPUs across COLOSSUS and COLOSSUS II, together with hyperscale-class CPUs, exabyte-scale storage and high-speed networking designed for AI workloads. Under the disclosed terms, Anthropic agreed to pay approximately $1.25B per month through May 2029, subject to the contractual ramp-up, termination and other provisions, giving SpaceX a direct commercial revenue stream associated with part of its computing infrastructure.
SpaceX has also disclosed an agreement with Google covering approximately 110,000 NVIDIA GPUs, together with CPUs, memory and other infrastructure, under which Google agreed to pay approximately $920M per month from October 2026 through June 2029, subject to the terms of the agreement.
For a CFO, the distinction is significant: infrastructure built exclusively for internal use generates returns indirectly through the products and services that consume the capacity, whereas infrastructure leased to external customers creates a contractual revenue stream directly associated with utilization. That does not eliminate technology risk, depreciation, power requirements, customer concentration or contractual termination risk, but it provides a more visible relationship between capital deployed, capacity installed and revenue contracted.
The Real Budget War
The most important difference between SpaceX and Big Tech is therefore not the headline amount of capital, but the capital structure behind it. Amazon, Microsoft, Alphabet and Meta can deploy hundreds of billions of dollars because their existing businesses generate cash and because the resulting infrastructure can support multiple revenue streams, including cloud services, advertising, software, enterprise products and AI applications, while SpaceX is building an infrastructure model in which external customers can directly pay for part of its compute capacity.
The two models can be summarized simply: Big Tech invests first and monetizes across a broad ecosystem, while SpaceX can invest in compute capacity and monetize part of that capacity directly through external contracts while retaining internal use. There is not enough public information to conclude that either model currently produces a superior return on invested capital, because the companies do not disclose comparable AI-specific ROI metrics, but the distinction will become increasingly important as AI infrastructure moves from experimental technology spending into multi-year capital allocation programs.
Who Is Taking the Bigger Risk?
On an absolute spending basis, Big Tech is committing vastly more capital, with the four companies’ current 2026 guidance implying approximately $697.5B at the midpoint, but absolute spending does not automatically equal financial risk because Amazon, Microsoft, Alphabet and Meta can spread infrastructure investment across enormous existing businesses and monetize the resulting capacity through several different channels. SpaceX has a narrower infrastructure monetization equation, but its disclosed compute contracts provide evidence that external customers are willing to commit substantial amounts of capital to its infrastructure, making the quality and durability of contracted demand an important variable for creditors and investors.
The Anthropic agreement alone represents approximately $1.25B of monthly payments under its stated terms, while the Google agreement provides for approximately $920M per month during its applicable payment period. These contracts should not be treated as guaranteed lifetime revenue, but they provide something particularly valuable in infrastructure finance: visibility into future cash flows tied to specific computing capacity.
The NVIDIA Position
There is another company sitting at the center of both sides of this Budget War: NVIDIA. Whether capital comes from Amazon’s operating cash flow, Microsoft’s balance sheet, Alphabet’s investment program, Meta’s CapEx or SpaceX’s debt-market access, a significant portion of the AI infrastructure cycle ultimately requires advanced compute hardware, placing NVIDIA in a position where the infrastructure buyers assume much of the utilization, depreciation and return-on-capital risk while NVIDIA captures revenue as the hardware and systems are sold.
NVIDIA’s Data Center business generated $89B of revenue in fiscal Q2 2027, while total quarterly revenue reached $96.2B, demonstrating the scale of the supplier economics at the center of the current infrastructure cycle. The significance for investors is straightforward: the companies financing and building AI infrastructure must ultimately prove that the capital generates sufficient economic returns, while the suppliers of critical computing capacity can monetize demand earlier in the investment cycle.
FAQ
Not specifically. SpaceX stated that the proceeds would primarily be used to repay its bridge loan, cover related expenses and fund general corporate purposes, so the $25B should not be classified as a $25B AI budget.
Amazon expects approximately $200B, Microsoft approximately $175B, Alphabet $180B, $190B and Meta $130B, $145B, producing a combined range of approximately $685B, $710B and a midpoint of approximately $697.5B. These are corporate CapEx figures rather than AI-only budgets.
Because the comparison is not only about size. SpaceX has developed a model in which external customers pay directly for access to its computing infrastructure, with its disclosed Anthropic agreement covering approximately 325,000 NVIDIA GPUs and providing for approximately $1.25B in monthly payments under the stated terms.
There is not enough public information to establish that. SpaceX’s contracts provide visibility into revenue associated with specific computing infrastructure, but the companies do not publish directly comparable AI-specific return-on-invested-capital figures.
Big Tech has the advantage in absolute financial capacity because its AI infrastructure programs are supported by large operating businesses and multiple financing sources, while SpaceX’s model places greater emphasis on converting infrastructure into contracted third-party revenue.
NVIDIA is one of the clearest beneficiaries because both hyperscalers and other AI infrastructure operators require its computing hardware, and its Data Center business generated $89B in fiscal Q2 2027.
Sources
Amazon, 2026 CapEx
https://ir.aboutamazon.com/news-release/news-release-details/2026/Amazon-com-Announces-Fourth-Quarter-Results/default.aspx
Microsoft, 2026 CapEx
https://www.microsoft.com/en-us/investor/events/fy-2026/earnings-fy-2026-q4
Alphabet, 2026 CapEx
https://www.sec.gov/Archives/edgar/data/1652044/000119312526251733/d160205dfwp.htm
Meta, 2026 CapEx
https://investor.atmeta.com/investor-news/press-release-details/2026/Meta-Reports-Second-Quarter-2026-Results/default.aspx
SpaceX, $25B Inaugural Bond Issuance
https://ir.spacex.com/updates/releases-details/2026/SpaceX-Announces-Pricing-of-25-Billion-Inaugural-Bond-Issuance-2026-33VwNgsx3O/default.aspx
SEC, SpaceX / Anthropic Compute Agreement
https://www.sec.gov/Archives/edgar/data/1181412/000162828026042639/spaceexplorationtechnologi.htm
SEC, SpaceX / Google Compute Agreement
https://www.sec.gov/Archives/edgar/data/1181412/000162828026040874/spacexukfwp.htm
NVIDIA, Q2 Fiscal 2027 Results
https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Announces-Financial-Results-for-Second-Quarter-Fiscal-2027/
NVIDIA, Q2 Fiscal 2027 Earnings Call Transcript
https://investor.nvidia.com/files/content_files/TRANSCRIPT_-NVIDIA-Corp-NVDA-US-Q2-2027-Earnings-Call-26-August-2026-5_00-PM-ET.pdf
The Bottom Line
The AI infrastructure cycle is no longer simply a technology spending race; it is a balance-sheet race, with Big Tech committing hundreds of billions of dollars of corporate CapEx while using cash generation, debt, leases and partnerships to expand capacity, and SpaceX accessing the bond market with a $25B senior unsecured issuance while developing a model in which external customers can pay directly for part of its compute infrastructure.
The critical metric over the next several years will not be who spends the most, but who converts capital into productive capacity, productive capacity into revenue and revenue into returns above the cost of capital. That is the real Budget War.