The Watson Weekly is sponsored by Avalara — the agentic AI platform automating global tax and compliance for leading eCommerce brands with simple plans, transparent pricing, and complete coverage across Shopify, Stripe, WooCommerce, BigCommerce, and more.

Amazon's $62.6B in net income is mostly an accounting event. The release puts $53.4B of non-operating pre-tax other income in the quarter, primarily from the Anthropic stake. Operating income was $27.5B, up 43% against $19.2B last year. That is the figure to work from.

Free cash flow is now an outflow of $7.6B on a trailing twelve month basis, against an $18.2B inflow a year ago. Net purchases of property and equipment ran $169B over those twelve months, up 64%. Q2 alone consumed $53.1B. Jassy raised the 2026 cash capex plan to roughly $220B from about $200B and attributed the increase to memory prices, not additional demand.

The funding is showing up on the balance sheet. Long-term debt went from $65.6B at year end to $128.9B at June 30, with $67B of long-term debt issued in six months. Colin Sebastian asked Olsavsky where the capital comes from next. Olsavsky said the options are open and there was nothing to share on the earnings call.

AWS margin of 39.4% is being read as evidence that AI workloads carry their weight. It was 39.5% in Q1 2025, before the build accelerated, so the quarter restores an old level rather than setting a new one. Olsavsky adjusted his own number on the call, 520 basis points of expansion instead of 650 once you remove the energy derivative gain. Roughly $1.2B of consolidated operating income came from two items he flagged: $600M of tariff refunds in North America and $600M of fair value movement on energy contracts. Q3 guidance assumes no further derivative remeasurement.

Retail is quieter and more useful. Paid units grew 17%. Worldwide shipping costs grew 19% to $27.9B, so the network is still absorbing fuel and line haul inflation despite the FBA surcharge implemented in April. Advertising hit $19.8B, up 26%. Physical stores grew 4% to $5.8B, which sits oddly beside the grocery narrative.

The grocery detail is what operators should mark. Same-day perishables now ship from same-day facilities in 2,300 U.S. cities. Jassy says nine of the top ten sellers in those cities are perishables, monthly active perishables customers are up more than 50% since January, and orders containing perishables carry three times the units. Basket size was always the argument for grocery. The 4% at physical stores says the volume is not moving through Whole Foods.

Two things from the Q&A. Jassy said Amazon can build a wildly successful business without a frontier model, then confirmed Amazon is building one for cost control, prioritization and speed, and expects at least half a dozen comparable models within a few years with one of them Amazon's. He also said there is a real chance Amazon sells Trainium to third-party data centers, which would make it a merchant silicon vendor bidding for supply it already cannot get enough of.

His demand framing was more candid than the headline. Adoption is barbelled: AI labs and a few runaway applications at one end, enterprise cost-avoidance projects at the other, and the entire middle of existing enterprise production workloads still mostly not running inference. Jassy said he does not know whether that middle climbs at the same slope.

Backlog is $496B. The trillion-dollar AWS number and the 2028 demand behind it are commitments, not cash.

Quince told WWD it crossed $2B in sales over the trailing twelve months, roughly double the figure it claimed for 2025. There is no earnings release behind that, and nobody has said whether the number means gross merchandise value or net revenue after returns. For a company selling $50 cashmere sweaters and furniture out of the same catalog, the distance between those two definitions is not small.

The March round priced the business at $10.1 billion with Iconiq leading. Against $2 billion in reported sales, that is about five times revenue for a company whose stated edge is demand forecasting at the size and SKU level and short production runs it scales once orders come in. Matt Lippert, chief merchandising officer, points to the $50 sweater holding its price through seven iterations since 2019 as evidence the model works.

Holding a price flat for seven years across tariff changes and freight swings means somebody absorbed the difference. The factories did, or Quince did, or the sweater is not the sweater it was in 2019.

Reformation filed in June. Tailored Brands filed this month. Lippert declined to comment on an IPO in March, which is what you say when the window is opening.