DX Today AI Daily Brief - Friday, September 4, 2026

0:00It's Friday, September 4, 2026. You're listening to the DX Today AI Daily Brief. Today, open AI ships its most powerful model yet and says the AGI era has begun. Nvidia buys hugging face for nearly $13 billion and Anthropic lines up 15 billion in credit ahead of a public offering. Let's get into it. Open AI began rolling out its next flagship model yesterday and the company is calling the moment a threshold. The model GPT-6 Astra is going first to a limited set of customers rather than the full-chat GPT user base. President Greg Brockman told reporters it is in his words not unreasonable to feel that we are now in the AGI era. Open AI says Astra came out of its largest training run to date more than 100,000 graphics processors at its Stargate site in Texas. It is also the first model the company has classified at its critical cybersecurity capability level,

1:02a designation that triggered additional internal safeguards before release. Now the deal of the week. Nvidia has agreed to buy hugging face, the platform where much of the open source AI world stores and shares its models. The price is $12.93 billion. One of the largest acquisitions Nvidia has ever made. Hugging face hosts millions of model checkpoints and data sets. And it has been effectively neutral ground in a market where chip loyalty runs deep. Nvidia says that neutrality survives the deal. In its announcement, the company said developers will still choose their own models, their own chips, and their own cloud platforms. And that Nvidia compute will not be required. Chief executive Jensen Huang framed the purchase as widening access to AI for developers and institutions worldwide. Next, a very large credit line.

2:03Anthropic is putting the financing in place for what could be one of the biggest public offerings in technology. Bloomberg reported yesterday that the company is close to finalizing an expansion of its revolving credit facility to $15 billion, up from 2.5 billion. Morgan Stanley is leading the arrangement with Goldman Sachs, JP Morgan, and City Group, also taking prominent roles. A credit line of that size is less about spending than about signaling. It clears a hurdle ahead of a public filing and the lineup of banks around a revolver typically overlaps with the lineup that ends up underwriting the offering itself. Anthropic has not commented publicly on the facility. Another valuation, another headline number. Miramarati's thinking machines lab is in talks to raise at least $1 billion at a valuation of at least $40 billion, according to reporting from the information and tech crunch. Axel is said to be in line to lead the round. The company's annual revenue run rate is reported at just over $100 million,

3:06which would put investors at roughly 400 times revenue. Nothing is signed. These are negotiations, and the terms could move. But the direction is unmistakable. Miramarati left the Chief Technology Officer job at OpenAI in 2024, and the lab she has built since has become one of the most richly valued private companies in the field. From Abu Dhabi, an open release. Abu Dhabi has published what it says is the largest fully open AI model family yet released. The Institute of Foundation models at Mohamed bin Zayed, University of Artificial Intelligence, launched a fleet called K2 Horizon yesterday. Six models ranging from 9 tenths of a billion parameters up to $375 billion. What separates this from most open weight releases is what came with it. The Institute published the weights, the code, the training data, and the methodology under an Apache 2 license. Most models described as open ship the weights alone

4:09and keep the recipe private. Releasing the data and the training process is what makes reproduction possible, and reproduction is what researchers have been asking for. Now, a product update. Google is putting a microphone in front of its productivity apps. The company began rolling out conversational voice modes for Gmail, Docs, and keep yesterday, branded Gmail live, Docs live, and keep live. The feature runs on a Gemini live integration, and it lets a user search a mailbox, draft a document or reorganize notes by talking, rather than typing into a side panel. Gemini can pull context from Gmail, drive, chat, and the open web, and it shows a recap of what it gathered before it generates anything. Availability is tiered. Gmail and keep go to AI Plus, Pro, and Ultra subscribers, while the Docs version is limited to Pro and Ultra. Turning now to China. DeepSeek is planning one of the largest known clusters of Chinese-made AI chips.

5:10Bloomberg reported this morning that the company intends to deploy at least 160,000 of Huawei's top accelerators at a data center it is building in a Mongolia. The report comes from people familiar with the plan. DeepSeek has not announced it, and Huawei has not confirmed the order. If it proceeds at that scale, it would mark a substantial move away from Nvidia hardware by one of China's most visible model developers. And it would be a real test of whether Huawei's accelerator supply can sustain a training cluster of that size. An unusual buyer of compute. The trading firm Jane Street is now buying computing power the way a hyper-scaler does. Crusoe, the AI data center company, has signed a five-year cloud contract with Jane Street valued at roughly $13 billion, according to Bloomberg. It is the highest profile customer Crusoe has landed for its cloud business, and it sits alongside contracts the company already holds

6:12with meta and oracle. What makes it notable is the buyer. Jane Street is a quantitative trading firm, not a model developer, and it has been steadily committing capital across the NeoCloud sector. Neither company has publicly confirmed the terms and the reporting rests on people familiar with the agreement. Now to enterprise software. Palantir and PWC are deepening a partnership that Wall Street clearly liked. The two announced an expanded strategic alliance yesterday, building an AI native deals platform on Palantir's Foundry and AIP products, aimed at mergers and acquisitions work, supply chain transformation, and enterprise resource planning migrations. The pitch is that transaction work can move up to 50% faster. PWC brings the industry practices and the client relationships. Palantir brings the data layer. Palantir shares rows about 7.5% on the day. For a company whose valuation has drawn persistent skepticism,

7:13a distribution deal through one of the big four consultancies is the kind of news that answers a specific objection. Earnings and the physical layer. Sienna's quarter is a reminder that the AI buildout runs on glass as much as it runs on silicon. The optical networking company reported record fiscal third quarter results yesterday. Revenue reached $1.67 billion. Up 37% from a year ago. Revenue from cloud providers grew more than 80%. The demand is coming from data center interconnect, the fiber that carries traffic between and across AI campuses that have grown too large to sit inside a single building. Sienna raised its outlook for the full fiscal year and adjusted profit per share climbed sharply. The company also flagged rising customer concentration, which is the standing risk in selling to a handful of very large buyers. From industry to the laboratory.

8:14Three Seattle research institutions are pulling money to teach AI the design rules of biology. The Allen Institute, the University of Washington and the Fred Hutch Cancer Center launched an accelerator called AI BioDesign yesterday, funded at $94.6 million over five years. The plan is a loop. Run large scale laboratory experiments. Feed the results into models. Use the models to propose proteins, genes, and biological systems that do not exist in nature. Then test those proposals back in the lab. The resulting data sets, models, assays, and tools are to be released openly. That last part is the unusual piece in a field where the most capable protein design work has been moving behind commercial walls. And finally, the labor market. A data point that cuts against the prevailing story about AI and jobs. Challenger Gray and Christmas reported yesterday that American employers announced 52,881 job cuts in August.

9:21That is the lowest August total since 2022. More striking is what employers gave as the reason. Restructuring led the list. Artificial intelligence fell to fourth the first month since February that AI was not the leading stated driver of announced layoffs. One month is not a trend. And stated reasons are not always the real ones. But after a year in which AI was named in nearly every layoff announcement, the August figures are worth marking. That's your briefing for Friday, September 4, 2026. For DX Today, stay curious.