01
VentureBeat ↗
Meta released Muse Glimmer, a 30 billion parameter open weight model built to run autonomous AI agents directly on consumer hardware rather than in the cloud, with weights available on Hugging Face now. It ships under the permissive Apache 2.0 license, Meta's first fully open release since it moved to the proprietary Muse Spark in April, and it drops the 700 million monthly user restriction that drew years of criticism of the Llama community license. The model is quantized to roughly 4 bits and uses block level speculative decoding so it runs fast enough inside a real agent loop on a single consumer GPU or a Mac, with Meta partnering with AMD, Arm, Dell, Intel and Nvidia on device optimization. The launch landed alongside a Mark Zuckerberg essay, The Future Is for Everyone, arguing that superintelligence controlled by a handful of companies or governments would produce worse outcomes for everyone else.
02
Bloomberg Technology ↗
Nvidia is partnering with six of the largest US investment firms, Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR, to set up independent compute financing platforms intended to mobilize more than $500 billion of third party capital for the AI buildout. The money would fund the full stack behind AI expansion, covering chips, power generation and data centers, as combined big tech capital spending heads past $730 billion this year. The structure follows a template Apollo and Blackstone used with Broadcom in June, when a $35 billion initial capital solution was designed to back more than 20 gigawatts of compute through 2028. Nvidia shares fell more than 3 percent in afternoon trading on the news.
03
Anthropic ↗
Anthropic asked an unreleased research version of Claude to attempt the Riemann hypothesis, one of the most famous unsolved problems in mathematics. It did not solve it, but it unexpectedly advanced a related result, raising the proven lower bound for the fraction of zeros of the Riemann zeta function satisfying the hypothesis from 41.6 percent to 67.2 percent. To get there the model coordinated 60 subagents, tested hundreds of ideas, searched the existing literature, attacked its own proof, and formalized the final result in Lean. Two mathematicians at Anthropic reviewed and validated the output and wrote an informal note stating the proof for other experts.
04
CNBC Technology ↗
Intel announced a $15 billion underwritten public offering of common stock, in what would be its first public share sale since the company listed in 1971. Proceeds are earmarked for general corporate purposes aimed at pushing Intel into new markets including applied AI and purpose built silicon, with the company citing a strong and sustainable demand environment driven by unprecedented AI compute investment. The deal includes a 30 day option for underwriters to buy an additional $2.25 billion in stock, with JPMorgan, Goldman Sachs, Morgan Stanley and Citigroup as joint book running managers. Intel stock has surged about 175 percent in 2026, but fell on the announcement as investors weighed the dilution.
05
Axios ↗
Senator Bernie Sanders sent a letter to Sam Altman, Dario Amodei and Mark Zuckerberg urging them to halt frontier AI development, writing that in the interest of humanity they should stand by their words, pause AI development, and stop building machines that humans cannot control. Sanders quoted each company's own prior commitments to stop if their systems became too risky to control safely, and pointed to recent incidents involving loss of human control and the creation of potentially dangerous viruses. He warned that if the companies do not act, he and his Senate colleagues will, and paired the letter with a broader policy package: a moratorium on new AI data centers, 50 percent government ownership of AI firms, and a one time 50 percent tax on their stock.
06
CNBC Technology ↗
OpenAI split its Daybreak cybersecurity program into two access tiers and released a new purpose trained model for security work. Daybreak Blue opens frontier general purpose models including GPT-5.6 Sol to approved defenders for routine security tasks, while Daybreak Red gates the new GPT-5.6-Cyber model behind tighter vetting for vulnerability research, exploit validation and security testing. GPT-5.6-Cyber is built on GPT-5.6 Sol and trained specifically to find zero days and build exploit chains; on OpenAI's internal advanced cybersecurity completion rate it answers 95 percent of advanced cyber prompts versus 1.5 percent for the standard Sol model. Under OpenAI's Preparedness Framework both models were rated High for cybersecurity capability but below the Critical threshold.
07
CNBC Technology ↗
TSMC reported July revenue of NT$467.58 billion, about $14.5 billion, up roughly 45 percent year over year and a record monthly high. High performance computing, the segment where TSMC books its AI chip sales, accounted for 66 percent of revenue in the second quarter, and the company now expects 2026 revenue to grow slightly above 40 percent in US dollar terms. TSMC also raised its 2026 capital expenditure projection to between $60 billion and $64 billion. Because TSMC manufactures for Nvidia, Apple and Google, its monthly figures are watched as a leading indicator of AI semiconductor demand, and European chip stocks including ASML rose on the report.
08
TechMeme ↗
Microsoft plans to unveil its next generation in house AI accelerator, the Maia 300, potentially as soon as September, according to The Information. The company is in talks with TSMC to secure manufacturing capacity for more than 300,000 units for delivery in 2027, and ultimately wants capacity for more than 1 million chips, though component supply and the ongoing TSMC negotiations could constrain those plans. Microsoft introduced the original Maia in November 2023 and unveiled the second generation Maia 200 in January on TSMC 3 nanometer technology, but it has lagged Alphabet and Amazon in scaling custom silicon to reduce its dependence on costly Nvidia processors. It is also trying to persuade major cloud customers such as Anthropic to adopt the chip.
09
VentureBeat ↗
Amazon Web Services announced at Black Hat USA 2026 that Continuum, its platform for finding and fixing code vulnerabilities, will plug directly into Anthropic's Claude Code and OpenAI's Codex alongside its own Kiro IDE. The move embeds AWS security tooling at the point where developers actually write code, regardless of which vendor's model they use, a bet that owning the security layer matters more than owning the model. AWS simultaneously expanded Security Hub Extended, its curated single bill security marketplace launched in February, with a tenth category focused on supply chain protection, adding Chainguard and Socket as partners. Together the announcements are the company's most sweeping attempt yet to become the default security control plane for enterprise software development in the AI era.
10
Unite.AI ↗
Dyna Robotics unveiled DYNA-2, a robot foundation model trained on roughly one million hours of ordinary human video instead of teleoperated robot demonstrations, the expensive bottleneck that has capped how far robot foundation models can scale. Its world action model architecture learns spatial reasoning and contact physics from video generation objectives first, then fine tunes to a specific robot platform in hours rather than weeks. Dyna reports manufacturing task success rates climbing from about 20 percent to between 80 and 90 percent, roughly 1.55 times more tasks completed than its previous DYNA-1 model, and just 13 minutes of data needed to teach robot hands to open bottle caps.