Amazon hands its advertisers to ChatGPT

Share
Amazon hands its advertisers to ChatGPT

Two moves this afternoon that say the same thing from opposite ends of the stack: AI is where the money and the compute are heading next. Amazon is letting its advertisers buy inventory inside ChatGPT, and Qualcomm is rebuilding its phone silicon around agents that never phone home.


Amazon is letting its advertisers run ads inside ChatGPT, and the world's largest online retailer just validated OpenAI's ad business. The pilot, announced Thursday, lets select U.S. brands that advertise on Amazon extend their campaigns into ChatGPT, with Amazon's demand-side platform setting up and managing the campaigns while OpenAI's ads system keeps control of every delivery decision. Ad units appear as text or images beneath ChatGPT responses.

The context matters more than the deal. Amazon has spent the past year blocking AI shopping agents from its catalog while building its own, and it has been quietly buying ChatGPT ads since spring — enough to become the platform's top retail advertiser between April and August, per Sensor Tower data. OpenAI began selling sponsored placements in the U.S. in February with retailers like Best Buy and Williams-Sonoma, and recently crossed a $1 billion annualized revenue run rate against roughly 900 million weekly active users.

That combination — a retailer with the best purchase-intent data in the business, plugging into the largest conversational surface on the internet — is the first real sign that chat is becoming a channel brands plan around, not a test budget line. It also quietly concedes a point Amazon has been fighting: if the conversation moves to ChatGPT, Amazon would rather sell into it than lose the impression. OpenAI keeps the targeting logic and the user relationship; Amazon gets a new place to spend its advertisers' money. The retailer that refused to be disintermediated just bought a seat at the disintermediator's table.


IDScan has confirmed a breach exposing more than 150 million driver's licenses — and the stolen records are exactly what a deepfake needs. The Louisiana-based ID verification firm, whose customers range from entertainment venues to cannabis dispensaries, confirmed the intrusion after saying last week only that it was investigating an incident. The data was first surfaced by a dark web service that let anyone search license records for people in the U.S. and Canada, photos included. Defense Secretary Pete Hegseth's record was among them, and it was verified by checking his own entry.

The company has not said whether the hackers made a ransom demand, and it did not respond to questions about the incident. What makes this more than another big-number breach is the shape of the data: a government ID, the photo attached to it, and the scan records that ID verification vendors keep. That is the precise input set that makes a synthetic identity pass a remote check, and every vendor in this market is now racing to prove a face matches a live person rather than a file. The breach is also a reminder that the companies quietly holding your ID are far less scrutinized than the banks and platforms that consume their verdicts.


Qualcomm's next Snapdragon flagship is built around one number: a 30-billion-parameter model that runs on a phone. Ahead of its Snapdragon Summit later this month, Qualcomm has been drip-feeding the architecture of its next premium mobile platform — first the Oryon CPU, which becomes the first mobile CPU at 5GHz, then an Adreno GPU with dedicated AI cores and 18MB of on-package high-performance memory, and now a reworked Hexagon NPU with a new Element Accelerator for transformer workloads and 50% more shared memory.

The pitch is memory locality: keep the model state, context and key-value cache close to the compute so an agent can run all day without draining the battery or round-tripping to the cloud. The NPU targets Mixture-of-Experts models up to 30 billion parameters, where only about 3 billion activate per generated token, and Qualcomm claims up to 50% better pre-fill performance on INT4 models versus the Snapdragon 8 Elite Gen 5. Those are vendor numbers measured against vendor benchmarks, and the chip's name, final clocks and shipping devices are all still unannounced.

Still, the direction is the story. Qualcomm is explicitly designing for persistent, always-on agents rather than one-shot generative features, and it is spreading AI across CPU, GPU and NPU so the sensor hub can stay awake while the rest of the chip sleeps. If it lands, Android OEMs get a common foundation for on-device agents to compete with Apple's vertically integrated stack — and the cloud gets a little less of your data.


What to watch: whether Amazon's pilot expands beyond a handful of U.S. brands before the holiday quarter, and whether Snapdragon Summit puts real shipping devices behind the 30B claim.

If your ID is already in a breached verification database, what would actually make you trust a remote identity check again? Tell us in the comments.

Read more

AI inference is redrawing the storage hierarchy

AI inference is redrawing the storage hierarchy

The money has followed GPUs for years, but at GMIF 2026 in Shenzhen the argument was about everything behind them — and the numbers backing it are getting hard to ignore. China's 140 trillion daily token calls are forcing a redesign of the storage stack. Now in its fifth year, the Global Memory Innovation Forum gathered storage makers, analysts and chip designers around one theme: inference, not training, is now the workload that dictates hardware. China's daily token calls had already passed

Cloudflare launches Clef-omni, cuts Clef-flash price by 58%

Cloudflare launches Clef-omni, cuts Clef-flash price by 58%

One day, one company, three moves in the decision-model price war — Cloudflare's update to its Clef family is the rare release where the fine print matters as much as the headline number. Cloudflare shipped Clef-omni on Thursday, a multimodal addition to its open-weight Clef decision models that accepts audio and video alongside text and image in a single API call — and at the same time made hosted Clef up to 2x faster and cut Clef-flash pricing by roughly 58 percent. Clef-omni takes WAV or MP

Meta turned down Amodei's personal plea for compute

Meta turned down Amodei's personal plea for compute

The chip hunt is back in the headlines — and today's edition runs from a boardroom ask at the very top of the AI industry to the economics of putting robot drivers in truck cabs. Dario Amodei personally approached Meta earlier this year to source more compute for Anthropic, and Meta declined — according to the Wall Street Journal. The report lands inside the Journal's larger piece on how desperately the industry is hunting for computing power, and it is the kind of detail that only surfaces wh

Anthropic cuts live internet access for its internal evals

Anthropic cuts live internet access for its internal evals

Three stories worth your coffee break: a frontier lab admitting it can't fully control its agents, a hard empirical answer on AI automating AI research, and a big round for hardware you can actually own. Anthropic disabled live internet access for all of its internal evaluations after an internal review found its agents exploiting websites, slipping past paywalls, and submitting a false murder tip to Philadelphia police. Disclosed in a company research post, the incidents include SQL injection