Amazon Ran Free Cash Flow Negative to Build AI: The CRE Read

Amazon.com released second quarter results on July 30, 2026, as Exhibit 99.1 to a Form 8-K. Net sales increased 20 percent to $200.6 billion against $167.7 billion a year earlier. The North America segment grew 16 percent to $116.2 billion, International grew 15 percent to $42.2 billion, and AWS grew 37 percent to $42.2 billion, which the company describes as its fastest growth in 18 quarters and a $169 billion annualized revenue run rate. Consolidated operating income rose to $27.5 billion from $19.2 billion, with AWS contributing $16.6 billion of that against $10.2 billion a year earlier. Net income was $62.6 billion, or $5.75 per diluted share, but the company states plainly that the figure includes $53.4 billion of non-operating pre-tax other income, primarily from its investments in Anthropic, so the headline earnings number is a mark on a stake rather than cash the business produced. Third quarter guidance puts net sales between $197.0 billion and $202.0 billion and operating income between $22.5 billion and $26.5 billion.

The cash flow statement is where the real estate story is. Operating cash flow increased 33 percent to $161.4 billion for the trailing twelve months. Free cash flow, which Amazon defines as operating cash flow less purchases of property and equipment net of proceeds from sales and incentives, decreased to an outflow of $7.6 billion for the trailing twelve months, against an inflow of $18.2 billion for the same period ended June 30, 2025. That is a swing of roughly $25.8 billion. Amazon gives the cause in its own words: the outflow was driven primarily by a year-over-year increase of $66.1 billion in purchases of property and equipment net of proceeds, and this increase “primarily reflects investments in artificial intelligence.” The underlying figures: gross purchases of property and equipment were $54.208 billion in the quarter and $173.028 billion over the trailing twelve months, and net of $1.132 billion of proceeds the quarterly figure computes to $53.076 billion. Trailing twelve month net purchases reached $169.007 billion against $102.953 billion a year earlier, which the supplemental table puts at 64 percent growth. The Form 10-Q filed July 31 carries the balance sheet behind it. Property and equipment, net stood at $446.046 billion as of June 30, 2026, against $357.025 billion at December 31, 2025. The segment table splits that $89.021 billion of growth: AWS went from $190.055 billion to $263.750 billion, North America from $122.043 billion to $135.013 billion, International from $30.632 billion to $32.879 billion, and Corporate from $14.295 billion to $14.404 billion. The commitments footnote adds the leased side: operating lease liabilities including interest of $116.350 billion, finance lease liabilities including interest of $16.660 billion, financing obligations including interest of $11.070 billion which Amazon notes are non-cancellable obligations for fulfillment network and data center facilities, and leases not yet commenced of $137.214 billion, of which $93.476 billion falls after five years. Those four rows compute to $281.294 billion of contracted lease and lease-like obligations, inside total commitments of $650.034 billion.

Strip out the earnings noise and the 10-Q says one thing an industrial landlord should read twice. Amazon added $89.0 billion of net property and equipment in six months, and AWS took $73.7 billion of it. The entire North America segment, which is to say the fulfillment centers, the sortation buildings, the delivery stations and every warehouse Amazon has ever been the marginal bidder for, added $12.97 billion. That is 82.8 percent of the six-month build going to the cloud segment, and AWS adding roughly 5.7 times what North America added. Measured as growth rates, AWS property and equipment expanded 38.8 percent in half a year while North America expanded 10.6 percent. For fifteen years the single most reliable demand assumption in American industrial real estate was that Amazon would keep taking space, and pro formas from Inland Empire to the I-78 corridor were underwritten against a tenant whose appetite looked structural. Amazon has not stopped taking industrial space, and 10.6 percent growth on a $122 billion base is not a retreat. But the capital inside the same company is now being allocated on a different priority order, and the number the market has spent two years watching, capital expenditure, is no longer mostly about warehouses. Anyone still treating hyperscaler capex headlines as a proxy for logistics demand is reading the wrong segment line.

The second read is about ownership, and it is sharpened by what Microsoft disclosed 24 hours earlier. Microsoft told investors on July 29 that it is extending datacenter useful lives to 25 years and that more of its future datacenter leases will shift from finance leases to operating leases, which is a company moving toward being a tenant, with $329.1 billion of leases signed but not yet commenced behind it. Amazon filed the day after with a different posture: $263.75 billion of AWS property and equipment on its own balance sheet, growing at 38.8 percent per half year, alongside $137.214 billion of its own leases not yet commenced. Both companies are committing enormous forward sums, but the mix differs, and the mix is the whole question for an owner. Microsoft’s pipeline is disproportionately a rent stream heading toward somebody’s balance sheet. Amazon’s is disproportionately capital it is spending itself, which is precisely why its free cash flow went negative while Microsoft’s reported capex figure fell. There is a useful historical rhyme here. Amazon ran negative or minimal free cash flow through the 2000s while it built the fulfillment network, and the operators who did best were not the ones who waited for the spending to prove out, they were the ones who owned the land and the power near where it was going. The difference this cycle is that the binding constraint moved from highway access and labor sheds to interconnection queues and substation capacity, and that a company willing to run $7.6 billion negative on free cash flow is a company that will pay for the constrained input. Who wins is whoever controls entitled, powered land that AWS can build on, or a shell it will sign long. Who is exposed is whoever underwrote speculative big-box industrial on the assumption that Amazon’s capital plan and Amazon’s warehouse plan are the same plan. As of this filing they are not.

Treat quarterly hyperscaler filings as a standing exposure feed rather than as news. The inputs are already public and almost nobody in a real estate shop reads them systematically: the 8-K earnings exhibit, the 10-Q segment tables for property and equipment, the commitments and leases footnotes, and, on your own side, the rent roll, the lease abstracts for every asset with a hyperscaler or hyperscaler-adjacent tenant, and the land bank with its interconnection status. An AI assistant pulls each new filing on the day it posts, extracts a fixed field set per company, and returns a cited row: property and equipment by segment this period and last, the period-over-period change, net purchases of property and equipment, free cash flow, operating and finance lease liabilities, leases signed but not yet commenced with the maturity split, and any language the filer itself uses to attribute the spending. Every cell carries the filing, the note and the page, and anything the filer does not disclose is marked unverified rather than estimated.

The output is a one-page quarterly exposure sheet per counterparty plus a delta view against last quarter, and the useful column is the one nobody builds: segment capital allocation, not headline capex. That is the column that would have flagged the AWS to North America divergence above. The reviewer is whoever owns the demand assumptions in your underwriting, usually a head of research or acquisitions, working with asset management for anything that touches an existing tenant. The approval gate is firm: no leasing assumption, land basis, or hold period changes on the strength of a filing summary until a person has opened the filing and confirmed the number in the note, because a segment table misread in either direction moves a land bank. The value is not knowing that Amazon spent a lot, which was on every wire on July 30. It is having the segment split, the lease pipeline and the free cash flow position on the same page, on the day, for every counterparty whose capital plan your portfolio quietly depends on.