ThredUp's annual report: the US secondhand market grew nearly four times faster than new clothing retail last year and is on track to hit $78 billion by 2030. The RealReal's CMO Samantha McCandless did a soft profile in Modern Retail today that mostly discusses her personal shopping habits. Nice branding. But the bigger story is infrastructure.

AI is rewriting resale discovery, and nobody has solved it cleanly yet. Poshmark bet its entire redesign on AI curation over inventory expansion. ThredUp's numbers confirm the market is big enough that whoever cracks search and discovery at scale wins something enormous. The catalog problem hits resale harder than any other category — every item is unique and most of them were never properly described. $78 billion is the prize. Solving unique-item AI discovery is the entry fee.

Meta is testing native product tagging directly inside Instagram Reels across five markets, and a Meta executive actually said the words out loud: the era of link in bio is finally over. This is not a feature announcement. This is the formal burial of the entire decade-long DTC playbook — build audience, funnel to link, convert on your own site.

For fashion brands, the arithmetic changes immediately. Native checkout on Reels means less friction and more impulse. It also means you now live and die inside Meta's algorithm rather than your own storefront. We watched this play out in Southeast Asia already: video commerce went from under 5% to 25% of GMV in just a few years — all platform-side, not brand-side.

The TikTok Shop AI discovery layer is already mature. Now Meta is chasing in earnest. Fragrance won TikTok Shop precisely because it converts on emotion and impulse — categories that thrive when the checkout is embedded in the scroll. Fashion is next, and the brands without a native Reels commerce strategy are now, officially, behind.

Chinese financial press is reporting that Gap Group has announced checkout functionality inside Google Gemini, described as the first mainstream fashion brand to enable AI shopping this way. We covered Gap's dual AI bet at Shoptalk — Bold Metrics sizing and Google's Universal Commerce Protocol. The Gemini checkout is the next domino.

The question isn't whether Gap moves meaningful volume through Gemini in 2026. It's whether fashion brands that don't establish AI checkout presence now will find themselves structurally absent from a discovery layer that matters in two years. Google Search is already becoming a retail interface — Gemini checkout is that thesis made transactional.

The liability shadow: when Gemini recommends the wrong size and the return rate spikes, whose problem is that? We answered this for ChatGPT storefronts — the brand gets blamed regardless of where the recommendation came from. Same answer here.

JD Retail has publicly surfaced its ecommerce AI architecture system for the first time, calling it Oxygen. The name is deliberate: infrastructure everything else breathes. Chinese ecommerce giants do not typically open-source their architectural thinking — when one does, read it simultaneously as a recruiting signal, a competitive flex, and a statement of intent about where they think the platform is going.

Put this alongside Meituan's approach of treating every product as a language token — also out today — and you have two very different philosophical bets on how AI ecommerce should be built. JD's model is architecturally layered; Meituan's is semantically unified. Both are operating at a level of abstraction that most Western platforms have not yet reached. The gap is not closing on its own.

极睿科技 (Jiruike Technology) has completed a 100 million yuan Series A led by Sequoia Capital China. The pitch: AIGC full-chain ecommerce content generation for fashion retail — AI-generated product imagery, copy, and campaign assets in one pipeline. The company's CTO gave a conference talk on how large models are reshaping the entire content generation engine from product shoot to published listing.

This is the second Sequoia China–led funding into AI fashion retail infrastructure in the same news cycle. That is not coincidence. ByteDance's Seedance benchmark set a floor: one person, one day, full campaign. Now the infrastructure to hit that floor at ecommerce scale — thousands of SKUs, not one hero campaign — is getting funded. The market Sequoia is betting on is not campaign production. It's catalog production. That is a much larger and more defensible surface.

Microsoft's terms of service for Copilot include the phrase for entertainment purposes only — the same disclaimer you see on psychic hotlines. TechCrunch flagged it. I am not going to let it pass.

This is the gap between what AI vendors sell at conferences and what they will defend in court. Fashion retailers deploying AI assistants for sizing guidance, return policy clarification, or product recommendations are building customer relationships on top of something the vendor officially classifies as entertainment. Only two of seven chatbot quality scores predicted a sale in the arXiv study we covered. Now we know part of why the other five aren't being fixed: reliability is not a contractual requirement.

The liability logic runs straight into what happens when the AI gets the product wrong. The brand takes the return. The vendor has a ToS. This asymmetry should be showing up in procurement conversations. The AI marketing stack problem — out today — is related: brands are building on infrastructure whose vendors have quietly capped their own accountability at zero.

Franziska Bell joins Home Depot as CTO with an explicit AI mandate focused on customer experience and operational efficiency. We argued in March that Home Depot is building AI tools that only work because of its physical footprint — the store is not being disrupted by AI, it is being amplified by it. Now they're institutionalizing that thesis at C-suite level.

The connection to the store as AI operating system is direct. Home Depot's advantage is that its AI has something real to operate on: physical aisles, inventory, local demand signals. Fashion retailers who read this and think it doesn't apply to them should explain what their AI is operating on instead.

Saks Global has secured $500 million in financing and expects to exit bankruptcy this summer. Over 650 brands now shipping merchandise — up from 500 a month ago. That's a meaningful vendor confidence signal, or a sign that brands have few alternatives in the luxury department store channel and are taking what's available. Probably both.

The real question is what Saks looks like post-exit. The luxury department store model needs a reason to exist beyond real estate and brand relationships. As checkout surfaces multiply — AI storefronts, Reels commerce, Gemini checkout — department stores that survive will need to function as discovery and curation engines, not just distribution nodes. Saks has the brand access. Whether the post-bankruptcy ownership has the technology conviction is the only question that matters for its next decade.

The Ministry of Commerce plus five other Chinese government departments issued a joint guidance document on ecommerce high-quality development. The operative language: develop AI+ecommerce, guide companies to strengthen AI large model research and application, optimize consumer experience, reduce operating costs.

Six-ministry alignment in China is not casual coordination. When this language appears in a joint directive, the policy environment for AI ecommerce investment is formally supportive at the national level. Compare to Europe, where retail volumes fell but AI budgets held — the difference is that European AI investment is happening despite market pressure. In China, it is now backed by explicit government mandate.

This is the structural context behind Alibaba's willingness to absorb a ¥10 billion profit drop — also out today — as the cost of AI infrastructure investment. The trade is not just a corporate strategy. It is government-endorsed.

TechCrunch reports that Japan is moving physical AI from pilot programs into real-world deployment — driven not by displacement but by demographic vacancy. The labor shortage in Japanese logistics is structural, not cyclical. Nobody is being pushed out; the positions were empty to begin with.

This is a direct fashion supply chain story. Physical Intelligence's second billion-dollar round in four months is chasing exactly this deployment window. Fashion 3PLs in Japan — and increasingly in Korea and parts of Europe — are running against the same demographic wall. The robot filling a fulfillment slot nobody applied for is not a threat model; it's an answer to a problem that was going to sink delivery SLAs either way.

Here is the connection that should worry Western fashion logistics: Japan is proving deployment-ready physical AI works in production. China's compute infrastructure is scaling to train the next generation of models. The supply chain is already fragile. The East is solving this faster than the West is discussing it.

Per OpenRouter data: last week, Chinese AI large models logged 12.96 trillion token calls, up 31.48% week-over-week. US models: 3.03 trillion. Five consecutive weeks where Chinese AI infrastructure outpaces America by raw volume — and the gap is widening, not narrowing.

This is a retail story, not an abstract geopolitics story. That compute is running ecommerce recommendation engines, AIGC content pipelines for fashion listings, and customer service stacks. Alibaba Cloud raised AI compute prices 34% in March and usage still surged. That is a demand curve that does not flinch at price — which tells you the value being extracted at the application layer is real.

红果短剧 (Hongguodj) issued a formal announcement: Q1 saw 1,718 AI drama titles removed for platform violations. In one targeted enforcement sweep, 15,000 works were fully audited and 670 removed specifically for unauthorized use of AI-generated materials — meaning the training data, the generated footage, or the character likenesses couldn't be cleared.

We called this in March: the AI short drama glut had arrived on schedule and the compliance problems would follow. They have. For fashion brands that adopted AI short drama as a campaign production channel — and several did, attracted by the ByteDance Seedance cost floor — this is the moment to audit your content rights chain. The IP problems that tanked these drama producers will surface in campaign content if nobody is checking provenance. The platform is checking now. Regulators will be next.

Modern Retail's Shoptalk recap confirms: the pressure to demonstrate actual AI results — not announce plans, not share roadmaps — is now the defining expectation at retail events. This is not a vibe shift. It is a structural change in how the industry evaluates AI investment.

We wrote from the floor that Shoptalk Spring marked the inflection point where AI's permission slip expired. The evidence base is now large enough that the vague announcement generates skepticism, not applause. Renner's six-year AI buildout produced 9.2% revenue growth on 3% less inventory. Algolia's survey ranked AI search as the top digital investment priority with concrete ROI attached. The industry has the data. The era of the AI press release as proof of competence is over.

Spangle, a six-person AI ecommerce startup, has raised $15 million at a $100 million valuation for what they're calling AI-mediated product pages — pages that adapt dynamically rather than presenting a fixed catalog layout. Chinese tech press is covering this as a US startup story worth watching.

The pitch is essentially that the product grid is structurally broken. Gen Z doesn't browse — Neritus Vale's piece today makes the demand-side case — which means the page optimized for browsers is the wrong artifact for a generation that didn't arrive intending to look around. Spangle is betting the future is a page that already knows what you want before you reach it.

Here is the connection worth flagging: Meta just killed the link in bio and Spangle is betting the product page itself is next. These are two separate companies making the same underlying wager — that the static, browse-optimized commerce interface has maybe 18 months of relevance left. Pinterest starting with mood rather than query is a third bet in the same direction. When three different companies independently decide the query-and-grid model is broken, it probably is.