NVIDIA used the GTC Berlin keynote to introduce DSX POD, a reference design for partner-built Vera Rubin AI factories in modular containers, and the OpenShell Reference Architecture for governing agents. OpenAI opened a private preview of GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna in Google Cloud's Model Garden. The lab also disclosed critical PHP flaws found with GPT-5.6 Sol and pointed operators to a Codex checker plus maintainer patches.

In the news

NVIDIA DSX POD and OpenShell at GTC Berlin

On Oct 21 at GTC Berlin, NVIDIA introduced DSX POD, a reference design for complete Vera Rubin AI factories that partners build inside modular containers, integrate, and validate before delivery. The units arrive ready to connect to a prepared site's power, cooling, and networking. NVIDIA says onsite deployment shrinks from six months to as little as 14 days once those utilities are ready, and that partners plan to deliver modular factories worldwide in early 2027. Named partners include Bull, Dell, HPE, Lenovo, Schneider Electric, Vertiv, Crusoe, Anduril, and Armada.1

The DSX platform, NVIDIA says, uses up to 100 percent less water for cooling and up to 30 percent less power for the same AI output. Jensen Huang told the Tempodrom audience that Vera Rubin is in full production.2 3

The same keynote launched the OpenShell Reference Architecture, a full-stack design for building, containing, observing, and enforcing safe agents. NVIDIA puts Vera Rubin on inference, Vera CPUs on the agent harness, and BlueField-4 DPUs running NVIDIA Sentry for out-of-band monitoring and in-silicon policy enforcement. The company claims NVL72 delivers 30 times higher throughput per megawatt on agentic workloads, Vera CPUs run at twice x86, Sentry quarantines in milliseconds, and the stack adds up to 20 percent more agent capacity. NVIDIA lists the docs as now available, and names the design as runnable at the edge on RTX Spark, DGX Spark, and DGX Station, plus Canonical, SUSE, Red Hat, Microsoft Windows with MXC, AWS, Google Cloud, and Azure. ServiceNow announced Arc as generally available, with actions governed by OpenShell.4

OpenAI models on Google Cloud Model Garden

OpenAI posted on Oct 21 that its frontier models and Codex agentic workflows are coming into Google Cloud's Model Garden. Customers purchase through Google Cloud Marketplace, and enterprises with GCP spend commitments can draw them down. Google also says usage counts toward Flexible Savings Plans across model providers.5

Google Cloud's matching post names GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna in private preview for select enterprise customers. The models are not generally available. Google claims it is now the only hyperscale cloud offering Gemini, Claude, and GPT, plus more than 200 models on one platform. Prompts and outputs stay in the customer perimeter and are not used to train foundation models. Zero Data Retention and Confidential Compute are coming soon.6

OpenAI says general availability is planned in the coming weeks, and that the models will soon be available in Google Antigravity and the Gemini Agent. Microsoft remains OpenAI's primary cloud partner, the company wrote. A request-access form is live.7

Patch the Planet PHP disclosure

OpenAI's Oct 21 Security post says that over the past two months the lab led a coalition to address critical vulnerabilities in PHP. Calif identified a critical flaw using GPT-5.6 Sol as part of Patch the Planet, and Wiz independently discovered and reported both vulnerabilities. PHP maintainers led the fixes and release, with Wiz, Cloudflare, and others contributing research and coordination. OpenAI is sharing a PHP version checker as a Codex skill. The post does not list CVE identifiers or PHP version numbers. Operators are pointed to maintainer patches and the checker.8 9

OpenAI says that, to its knowledge, frontier AI found the bugs before attackers did, opening what it calls a Defender's Window. The same post commits $1 billion in subsidized Daybreak access through Daybreak for Frontline Defenders, aimed at organizations with limited security resources that protect essential services. The proposed response runs through five stages: validate and fix, give organizations time, coordinate with government partners including the White House Super Intelligence Force, ONCD, and CISA, prepare updates across products, then publish guidance. Public guidance sits on the Patch the Planet site.10 11