Apple's 512-gigabyte Mac + OpenAI's own inference chip

Apple introduced the M6 and M5 Ultra chips alongside a new Mac Studio and Mac mini, and said plainly that the top configuration can run large language models with hundreds of billions of parameters entirely on device. The M5 Ultra pairs up to 512 gigabytes of unified memory with 1.2 terabytes per second of bandwidth, about 50 percent more than the previous generation, and Apple claims 4.5 times the peak graphics compute for AI against the M3 Ultra. Mac Studio starts at $2,499 with the M5 Max and $5,499 with the M5 Ultra, shipping September 22, with the largest memory configuration held back to late October.

OpenAI presented Jalapeño at the Hot Chips conference and invited SemiAnalysis into its lab to run the InferenceX benchmark suite against it. The chip was designed with Broadcom from a blank sheet for language-model inference, went from initial hiring to tape-out in roughly 16 months, and uses HBM4 memory. SemiAnalysis measured more than 700 tokens per second per user on DeepSeek R1 at a concurrency of one, with no speculative decoding. Two caveats belong beside the headline, and SemiAnalysis states both: OpenAI supplied the numbers, and the fair comparison is Nvidia's shipping Rubin generation rather than Blackwell.

OpenAI said it removed a cluster of accounts operating from Russia through virtual private networks that used ChatGPT to draft social media posts and then to strip the linguistic tells of their origin. Much of the output promoted the International Burke Institute, a self-described expert community whose site falsely listed scholars including Francis Fukuyama and Noam Chomsky, and which published a sovereignty index rating countries in ways favorable to Moscow. Of thirty-six articles attributed to the site's experts, OpenAI found thirty-four had been copied from elsewhere. The company said reach was limited and the manufactured authority was the real product.

Louisiana Economic Development said SpaceX will build its largest launch site on 125,000 acres of former Exxon land in Vermilion Parish, with five launch complexes of two pads each, its own propellant production and power generation, and capacity for thousands of launches a year. Construction begins in 2027 and the first Starship flight is targeted for 2029, against 3,000 direct jobs over a decade at an average salary of $92,600. The number that matters for AI readers is launch cadence: SpaceX has filed for as many as one million compute satellites, and no version of that plan works at the flight rate it can manage today.

Alibaba's Qwen team said it will release Qwen3.8-Flash-Next, a multimodal mixture-of-experts (MoE) model with 125 billion total parameters that activates roughly 6 billion of them per token, at 11 in the evening Beijing time on Wednesday. The framing is the unusual part. This is a technology preview of the architecture that will carry the full Qwen 4 family rather than a flagship in its own right, so developers get to build against the new design months ahead of the models that will use it. At the time of writing the weights were not yet on Hugging Face, no benchmark scores had been published, and no license had been named.

David Buchanan published a walk-through showing that the hardware-backed key attestation underpinning the C2PA content-provenance standard on Android cameras does not survive a rooted device. Extracting the private key is not necessary: with root, an attacker can simply ask the secure element to sign whatever data they choose, and he produced AI-generated images and video that validated as camera-captured on Pixel 8a and 9a hardware. He demonstrated a one-click software root against a current patch level and a low-cost memory fault-injection attack that no patch can reach. Content Credentials are increasingly offered as the answer to synthetic media, which makes the distance between the promise and the device worth reading closely.

Stability AI raised $76 million in a Series B whose investor list is the story: Universal Music Group, Sony Music Group, Warner Music Group, Electronic Arts, AMD Ventures and Pacific Alliance Ventures. That brings the company to $232 million raised under chief executive Prem Akkaraju, who joined in 2024. The money goes to a creative production suite and a professional services arm, with the labels positioned as co-developers rather than licensees, which is a markedly different posture from the copyright litigation running elsewhere between rights holders and generative media companies.

Generalist, founded by two former Google DeepMind researchers and a former Boston Dynamics engineer, has added about $200 million led by 8VC, taking its Series B to $600 million and its valuation to $3 billion, according to sources cited by TechCrunch. The company was valued at $2 billion in June. Its pitch is a robotics foundation model that teaches machines new tasks from video demonstrations as short as three to twelve seconds, and its earlier backers include Nvidia, Bezos Expeditions and the researcher Fei-Fei Li. It has no relation to the similarly named General Intuition, a different robotics startup that raised at a $6 billion valuation a day earlier.