Amazon's Q2 2026 earnings dropped last week, and if you skimmed the headlines you probably saw the Wall Street version: net sales up 20% to $200.6 billion, operating income up 43% to $27.5 billion, AWS growing 37%, its fastest clip in 18 quarters.

But buried in that release is a bigger story for sellers. Amazon is no longer a retailer with a cloud business bolted on. As Gennaro Cuofano of The Business Engineer put it in his excellent deep dive, Amazon has reorganized itself into "The Everything AI Factory.” It’s one giant, interconnected AI machine where the store, the cloud, the ads, the robots, and the delivery network all run on the same engine.

And you, the seller, are operating inside that factory. Here's what the numbers say about where this is going, and what to do about it.

The proof it's an AI company now

Follow the money. Amazon spent $173 billion (up 64%) on capital expenditures over the trailing twelve months and free cash flow actually went negative ($7.6 billion outflow) because of it. Amazon explicitly says the increase "primarily reflects investments in artificial intelligence." The company nearly doubled its long-term debt on purpose to build faster.

AWS's AI business and its custom chips business (Trainium, Graviton) each blew past $25 billion annual run rates, both growing triple-digits. Anthropic and OpenAI, the two leading AI labs in the world, made multi-year, multi-gigawatt commitments to Amazon's chips. And Amazon's stake in Anthropic delivered a $53.4 billion pre-tax gain in a single quarter, pushing net income to $62.6 billion.

When a company redirects that much capital into one thing, that thing becomes the company. The retail side isn't separate from this, it's the showcase for it.

What the AI factory means on the seller side of the glass

Here's where it hits your business directly:

1. AI shopping assistants are going mainstream — fast. Amazon merged Rufus and Alexa+ into "Alexa for Shopping," an agentic assistant that compares products, tracks price history, and can automatically buy through Price Alerts and Auto-Buy.

Active users nearly doubled and interactions are up 5x year-over-year. Even better: customers who shop through Alexa spend over 40% more per order. Translation: an increasing share of your sales will be decided by an AI reading your listing, not a human scrolling past your main image.

Structured, specific, benefit-rich listing content isn't optional anymore. It's how you get recommended by the machine.

2. Auto-Buy changes the repeat-purchase game. If a customer sets Auto-Buy on a competitor's electrolyte powder, you don't get a second chance at that buy box. Winning the first AI-assisted purchase now locks in a subscription-like revenue stream. Price competitiveness and review velocity matter more than ever, because the agent is watching both.

3. Ads are getting cheaper for sellers who use Amazon's AI, and pricier for those who don't. Advertising grew 26% year-over-year (a $70B+ annual business now, bigger than any software company on earth). Amazon expanded Ads Agent to 11 new countries this year, and advertisers using it see 8% lower cost-per-impression and 6% lower cost-per-acquisition. When your competitors' AI is optimizing campaigns in minutes, running manual campaigns is a self-imposed tax.

4. Speed is the new table stakes. Prime members got 40%+ more items same-day or overnight in the first half, and Amazon Now (30-minute delivery) added 80 U.S. cities with 80% quarter-over-quarter sales growth. Grocery and Everyday Essentials are growing meaningfully faster than the rest of the business.

If you sell replenishable, everyday-use products, this is your tailwind. If your inventory placement can't support fast delivery promises, you're increasingly invisible.

5. The robots are coming for your fees (in a good way, maybe). The next-gen Proteus robot moves 1,300-pound loads and now takes plain-language voice commands from warehouse workers. Amazon also opened its entire logistics network to outside companies via Amazon Supply Chain Services. P&G, 3M, and American Eagle are already in. Amazon is turning fulfillment itself into a product, powered by the same AI stack.

6. Big brands keep flooding in. Amazon added 700,000+ products from names like Rabanne, Bobbi Brown, and Ted Baker. The AI factory attracts premium brands, which means more competition, and higher customer expectations, in nearly every category.

Where this is headed

Cuofano's framing nails it: Amazon is the only company that owns both "the wafer and the doorstep." When a customer asks Alexa to reorder detergent, the request runs on Amazon's models, on Amazon's chips, in Amazon's data centers, ships through Amazon's trucks, and monetizes through Amazon's ads. Every step of the loop belongs to Amazon.

For sellers, the strategic takeaway is simple: the customer journey is being rebuilt around AI agents, and every Amazon tool you touch, listings like ads, fulfillment, pricing is being rebuilt with it. The sellers who win the next 24 months will be the ones who optimize for how machines shop, not just how humans browse.

The old Amazon was two companies. The new Amazon is one factory. Make sure your products are built for the assembly line.

This piece is based on analysis from the team at Incrementum Digital. Credit to them for the original breakdown — we've adapted it here for BDSN readers.

Amazon just started requiring sellers to label AI-generated people in their listings, and much of the ecommerce world read it as a crackdown on AI imagery in general.

But look at what the rule actually covers, and a more useful picture emerges: it targets photorealistic synthetic humans, and nothing else. Most of the imagery that actually helps a shopper buy doesn't need a person in it at all.

What the rule actually says

Amazon now requires a metadata tag on any photorealistic AI-generated person appearing in images, videos, or A+ content, and shoppers see an indicator when one is present. This isn't Amazon being difficult. It follows New York's new synthetic-performer disclosure law, and similar rules are spreading to other states. In other words, this is compliance, and it's likely just the beginning.

Here's the line the rule draws. AI imagery with no person in it — the product in a setting, an object, a benefit shown visually — is completely untouched. AI imagery with a synthetic person in it is the only kind that gets labeled. Almost everything that helps a shopper sits on the safe side of that line.

What you can generate freely

People buy what they can picture owning, and that takes a scene: the product in a believable moment close to the shopper's own life. A real lifestyle photo shoot is still the gold standard, but many brands can only afford one or two.

This is where AI earns its place cleanly. You can render your product on a gym floor, a work desk, a stroller cup holder, or a nightstand without booking a shoot and without a single synthetic person on screen. No person, no label, nothing to disclose.

The same logic applies to benefits a camera can't easily capture: quiet enough for a nursery, packs flat in a carry-on, fits a standard cup holder. Amazon's new 125-character "Item highlights" field is built for exactly these claims, so pair each one with an image that shows it.

The one move to avoid

The rule is aimed squarely at generating fake people. It might be tempting for a brand that can only afford one or two models to generate a whole diverse cast so every shopper sees someone like themselves. But a synthetic person on your listing implies a real customer, and manufacturing that stages social proof that doesn't exist. No metadata tag makes an invented customer real, and shoppers can feel the pretense.

There is a clean exception: an AI-generated person is fine when it works as information, not testimony. A synthetic hand holding the product for scale, a body showing proportion or fit, a figure demonstrating assembly read as specs, not customers. The key is presenting them as demonstration, never as a gallery of happy buyers.

One more wrinkle worth knowing: the rule exempts images of real people even when AI was used to alter them. So AI retouching, lighting fixes, and background cleanup on real photography don't trigger the tag.

Shoot a consenting model for real, then let AI handle the polish. Just don't generate that real person into scenes they never shot. That's fabrication, requires their consent, and can land you back on the labeled side.

The bottom line

Generate people-free imagery freely. Extra use-scenes and benefit visuals need no label and give a small catalog reach it couldn't otherwise afford. Put real people in the people slots: your models, your testimonials, your customer content. And never fabricate customers. A synthetic face standing in for a real buyer is the one use that's both labeled and deceptive.

The label reads like a restriction on AI imagery, but it's really a line drawn around one use of it: people. Stay on the near side of that line and AI is a clean, cheap way to give buyers what makes them buy. And with Q4 approaching, this quiet stretch is the right time to decide what to generate and what to shoot.

Ever post a TikTok video that gets a few hundred views and then just dies? According to Stuart Baddiley of Optimise Your Marketing, that's not bad luck, it's a test you failed. And in 2026, that test has gotten noticeably harder to pass.

Every video starts with a test

When you post a new video, TikTok doesn't show it to the world. It shows it to a small sample of your existing followers first as a trial. How that small group reacts decides everything. Do they watch to the end? Do they share it or save it? If the answer is no, the video stalls out, and it almost never recovers later, no matter how good the idea was.

The completion rate bar has moved sharply

Here's the biggest change: to have a real shot at going viral, a video now needs a completion rate above 70%, meaning 70% of viewers watch all the way to the end. Back in 2024, roughly 50% was enough. That's a major tightening. Videos that would have gone wide two years ago now die in the test phase. Every second of footage that doesn't need to be there is now a liability.

There's also a strong secondary signal: rewatches. If 15–20% of viewers watch your video a second time, TikTok reads that as a mark of real quality. As Stuart puts it, TikTok isn't asking whether people watched your video once. It's asking whether they watched to the end and then watched it again. That's a much higher bar than most content is built for.

Consistency beats viral moments

The other big shift: posting three to five times a week now outperforms chasing one big viral hit and going quiet for weeks. TikTok's system rewards accounts it can test and learn from regularly. An account that posts sporadically gives the algorithm less data to work with, which quietly caps its reach, even when a video does perform well.

Knowing the rules is the easy part. The hard part is doing the work like editing videos tightly enough to hit that 70% bar, posting several times a week without fail, and resisting the urge to swing for one big viral moment. In Stuart's 18 years of experience, the accounts that grow on TikTok are never the ones with the biggest single hit. They're the ones that show up consistently enough for the algorithm to trust them.

So if your videos keep stalling after a few hundred views, the real bottleneck may not be your content quality, it may be your completion rate.

Amazon quietly added a powerful new feature to Product Opportunity Explorer, one of the most underrated tools in Seller Central: Validate a Product Idea.

You enter a product title, short description, a few bullets, and your target price, and Amazon tells you whether you'd have an advantage or disadvantage in that market.

Isaac Gross of IG PPC tested it with sugar-free electrolyte powder packets at $30 and got back: a feature-gap analysis showing where his concept trailed the competition (mineral breadth, added vitamins, certifications, sweetener transparency), the top brands owning the segment (FlavCity at 27%), customer demographics by income bracket, 12 months of seasonal click trends, and specific listing recommendations — like the fact that 7 of 10 benchmark titles explicitly name 2+ electrolyte minerals. It even confirmed his $30 price was well-positioned against the $29.22 category benchmark.

The kicker: this is 1P data straight from Amazon's own search, purchase, and review data, and not scraped third-party estimates.

Find it under Seller Central → Growth → Product Opportunity Explorer → Validate a new product idea. Worth running on your next idea, or your current products.

✌🏼 See you again Thursday …

The answer to today’s STUMP BEZOS is
6% of Gemini’s conversations have purchase intent