The U.S. Census Bureau's 3 November full M3 report estimated September 2026 seasonally adjusted new manufacturing orders at $677.247 billion, up 1.8% from revised August. Orders excluding transportation were $557.359 billion, up 1.3%. Shipments were $668.454 billion, up 1.3%; unfilled orders were $1,623.818 billion at month-end, up 0.9%; and inventories were $976.044 billion, up 0.4%. The total manufacturing inventory-to-shipments ratio eased to 1.46 from 1.47. Orders and shipments measure activity during the month, while backlog and inventory are end-of-month stocks, so they are not interchangeable. The September estimates are preliminary, subject to revision and not adjusted for price changes. Census also says the survey panel is not a probability sample, so sampling error cannot be calculated.
On 21 October, NVIDIA announced capacity agreements and planned facilities across several stages of its supply chain. It says substrate agreements with IBIDEN, Samsung Electro-Mechanics and Unimicron should raise aggregate capacity 2.5 times over three years. Amkor and NVIDIA target more than a 300% increase in interposer-wafer capacity by the end of 2028. NVIDIA expects total semiconductor-testing capacity to double from 2026 to 2027; a $1.1 billion reservation advance is intended to support KYEC testing facilities in the United States, with high-volume production targeted for April 2027. SPIL targets a 15-fold increase in optical-engine capacity by mid-2027 and aims to double it again by the end of 2028. These are company targets and forward-looking statements. They are not completed capacity, independently audited output, guaranteed allocation, delivery dates or customer pricing.
Microsoft and AMD announced different rack-scale AI infrastructure approaches on 12 October. Microsoft says a Maia 300 rack combines 96 accelerators in a 167-kilowatt power envelope, with 27.6 TB of HBM capacity and 1.7 PB/s of aggregate HBM bandwidth; production deployments are planned for 2027. AMD introduced Helion as an open, OCP-based reference design with CPU-only and mixed CPU/GPU configurations. AMD says configurations are under development with partners and planned for November 2026; its product page lists up to 32,768 CPU cores and 16 MI350P GPUs, depending on configuration. These specifications, performance descriptions, schedules and expected benefits are vendor statements. They are not directly comparable benchmarks or independent proof of cost, availability or production results.
Between 28 and 29 September, three vendors announced different layers of control for AI agents. OpenAI began rolling out Dots, always-on agents with cloud computers and connected apps, to eligible Pro and Business Premium users; its Enterprise beta is off by default and requires an administrator to enable it. Google Cloud added preview-level, resource-specific IAM controls for individual Gemini Enterprise apps and data stores. NVIDIA announced an Open Agent Safety Platform: it says the broadly available OpenShell runtime can trace actions and enforce policy, while its Sentry reference design can monitor and quarantine agents outside the application layer. These are vendor product descriptions and rollout claims. They do not independently prove that a particular deployment is safe, reliable or profitable.