Leitura

At the 2026 Snapdragon Summit, Qualcomm announced that the Adreno GPU in the sixth-generation Snapdragon 8 Elite has, for the first time, dedicated cores for AI computation: Adreno Matrix Cores.

Built on that hardware, Adreno Neural Fusion combines neural processing, AI super-resolution and frame generation in a single rendering pipeline — rendering at lower resolution first, then using AI to restore detail and add frames, with the stated goal of improving image quality, frame rate and power draw together. Qualcomm says it already supports Unity and Unreal Engine 5, so developers need not build their own upscaling solutions.

All of this is vendor claims from the summit, with no independent measurements; how Nanite and Lumen actually behave on phones will only be clear once devices with the chip are tested.

At the 2026 Snapdragon Summit, Qualcomm announced that its sixth-generation Snapdragon 8 Elite uses a new Offset Package-on-Package design.

Conventional PoP stacks DRAM directly on top of the SoC, saving board space but blocking heat from escaping. The offset design physically shifts the hottest cores — CPU and GPU — so memory no longer fully covers the SoC, letting those areas contact vapor chambers or graphite sheets more directly.

Qualcomm says this sustains peak performance longer under gaming, AI inference and rendering loads. That remains a vendor claim; real sustained performance will only be verifiable through reviews and teardowns of shipping devices such as the Xiaomi 18 Pro.

At the 2026 Snapdragon Summit, Qualcomm announced it will bring its HBC (High Bandwidth Compute) architecture to Snapdragon chips, per on-site reporting by IT之家.

HBC is a near-memory computing architecture that places compute beneath stacked LPDDR DRAM, using 3D integration and through-silicon vias to shorten data movement. The architecture was first unveiled at Qualcomm's investor day in June 2026, at that point only for datacenter products.

Qualcomm's argument is that the bottleneck for on-device AI is no longer just NPU compute but the speed of moving data between memory and processor. If HBC ships, phones could run larger, longer-context models locally. Note this is an announced intention: no timeline, no specific chip, and no measured performance figures were given, so the claimed benefits remain unverified.

AMD rose nearly 10% to close at $615.52 on September 21, pushing its market capitalization above $1 trillion for the first time.

The stock is up about 30% over the past 30 days. Intel shares rose more than 12% the same day, and Reuters noted the move was a broad rally in AI-adjacent hardware stocks rather than an AMD-specific event.

Fast Company attributes the surge to investor optimism about AI infrastructure spending, citing the popularity of Meta's new AI app Muse. That causal story is the article's interpretation; AMD disclosed no new orders or results that day.

OpenAI has announced AGMAI, a nine-member advisory panel of mathematicians that will advise AI companies on how new mathematical results are reviewed and released. The group is hosted by the Institute for Advanced Study in Princeton.

OpenAI says members will not be paid by the company, can offer unsolicited advice and comment publicly, and will get early access to its research. Members come from Stanford, Harvard, Oxford and other institutions, with multiple Fields Medals among them.

The panel follows a reputational crisis over a string of mathematical claims. Simon Machado of ETH Zurich said the member selection seemed "shrouded in mystery"; member Martin Hairer wrote in a blog post that the group is independent of the AI labs, while conceding that nine people can hardly represent the whole mathematics community.

Amazon has blocked Meta's AI shopping agent Muse from shopping on Amazon.com. Users now see a warning that access by an unauthorized AI agent violates Amazon's terms of service.

Amazon's stated reasons: Muse does not identify itself as an AI while browsing and appears to store customer data, creating security risks. Meta had earlier said Muse has no access to passwords or payment details.

This extends Amazon's campaign against shopping agents, which has already targeted Perplexity, Google and OpenAI. Meta, notably, signed a billion-dollar deal for Amazon's cloud chips in April, so the two remain business partners. Muse launched on September 8 and became the most-downloaded free US App Store app within a week.

Amazon has blocked Meta's Muse AI agent from making purchases on its store.

Starting Sunday night, September 20, users trying to buy goods through Muse received an error stating that continued access by an unauthorized AI agent violates Amazon's Conditions of Use. TechCrunch reports the block was first spotted by GeekWire.

Amazon is under no legal obligation to open its doors to Muse, and if the agent places a bad order, the platform handles the fallout. Muse has one of the lower hallucination rates among AI models, but it is still far from zero.

Ireland's Data Protection Commission fined Google €403 million (about $462 million) on September 21.

The investigation began in early 2020 and found Google's location data processing violated the GDPR: users could have been unaware their location was used for ad targeting or interest inference, and data was retained longer than necessary. The commission also ordered Google to bring its processing into compliance within six months.

Google said the case concerns historical policies and that its practices have evolved significantly since 2019. The penalty is among the largest ever imposed by the Irish regulator, and whether the remediation lands on time is the thing to watch.

California Governor Gavin Newsom signed seven bills on September 21 requiring data centers to pay for upgrades to local power grids and water systems.

The package requires the California Public Utilities Commission to create a new rate classification for data centers, and new projects must disclose estimated water use, energy efficiency and drought planning to local governments.

The actual cost impact cannot be quantified until the CPUC writes the rate rules. According to The Verge, Newsom said the laws are meant to stop data centers from passing utility costs onto residents.

The Canadian province of British Columbia sued OpenAI in a California court on September 21, alleging the company failed to report violent ChatGPT activity by the perpetrator of a school shooting.

The case relates to the February shooting in Tumbler Ridge that killed eight people. Attorney General Niki Sharma said the suit seeks accountability and change; OpenAI had previously apologised for not flagging the account.

The suit was just filed. The specific claims and any damages are not yet visible in court documents, and whether OpenAI owes any reporting duty will be for the court to decide.

An MIT Technology Review investigation found that more than 1,050 people died within range of border surveillance towers between 2015 and early 2026, including more than 110 within range of Anduril's autonomous towers.

The team cross-referenced nearly 4,000 locations where human remains were found with nearly 600 tower locations identified by the Electronic Frontier Foundation. Deaths occurred within the advertised range of about two-thirds of the towers analyzed, and most did not happen in terrain blind spots. In April 2024, a man died 360 feet from the nearest Anduril tower in New Mexico; landfill workers, not Border Patrol, spotted him first.

Anduril responded that towers are operated by CBP once delivered, that actual coverage varies with terrain and CBP-set virtual boundaries, and that the reporting contains inaccuracies it did not specify. The investigation also notes it counted only deaths that left a trace.

404 Media reports that multiple contractors hired to improve OpenAI's models have been fired for using AI to do their own work. One contractor said that in a group of thousands, many have been caught, and that it is essentially the one offense that gets you removed immediately.

The contractors review real ChatGPT conversations and rate the model's responses; some projects involve more than ten thousand contractors. Internal documents bar reviewers from using AI themselves, including Grammarly and AI translation, and forbid AI-detection tools like GPTZero on the grounds that they are unreliable. Many contractors work through the AI-training company Mercor, whose spokesperson confirmed that contracts strictly prohibit using LLMs and that confirmed violators are removed from projects. OpenAI declined to comment.

The report also quotes one contractor who has trained models for several AI companies saying they sometimes deliberately chose the worst responses to sabotage training. The core facts rest on anonymous sources and internal documents, and the prevalence of firings cannot be independently verified; but if AI-generated content is entering the labeling pipeline, the reliability of human-feedback data itself is in question.