Advertise on podcast: NOVA — Industry & Products | AGORA Intelligence

Real launches and quarterly numbers: what actually shipped, what it changes for the buyer, what stays an announcement.

Google is giving K-12 schools separate controls for Gemini, Gemini notebook and Search, all off by default, while Search is still set to become an AI chatbot. New York's moratorium, which the city calls the broadest in the country, covers 0.4% of the students Google turned Gemini on for in August. That suggests the buyer's leverage lies in the clause written in at renewal.

The $15 billion in data center revenue Qualcomm is promising for 2029 is about 1.4 times the 2026 revenue reported by The Verge and almost six times the annualized pace of its Q1 2027 guidance. For boards, that figure is leverage in compute contract renewals today, before it is supply.

Pliant is opening B2B payments to Claude, ChatGPT and customers' own agents with an $80 million Series C: the most likely reading is that the agent becomes replaceable and value shifts to whoever holds the license and the liability. In the US, where part of the round is headed, that layer depends on regulated partners, so boards need to ask who signs in each country.

In NVIDIA's post on cities, Nemotron recurs in five of eight cases while prices and contract terms are missing from all of them. Service delivery stays with integrators such as ITMAX and 1MillionBot, so municipalities risk their strongest lock-in in the long-term service contract and shared infrastructure. All figures are claims by NVIDIA and its partners.

In Seoul, Siemens and Naver put payback on display in place of the pilot: 10,000 machines paid back in under three months according to Busch, and more than 200 robots in service for over four years according to Choi. Industrial AI lock-in thus shifts to the data connector, the barrier that 41% of respondents to the audience poll named as the first.

Nvidia is in talks to guarantee the construction debt for Nscale's campus for Anthropic, which would take it beyond the leases and chips that have made up most of its support so far. Together, the contract Anthropic signed and the one under discussion are worth about 2.6 times the top-end valuation Nscale is seeking in its IPO. In August, Nscale had about one-twentieth of its planned chips running, and the guarantee would put that delivery risk on Nvidia's balance sheet.

Microsoft is opening up its Copilot harness on Foundry. The harness orchestrates more than 20 frontier models across five named surfaces, which makes at least 100 combinations run by a single layer, and turns the model into an interchangeable part. If customers hand over model selection, the Redmond router sets the cost of each task, and the October 28 post says nothing about service status or pricing.

At SuiteWorld 26, Oracle NetSuite said its more than 45 SuiteSkills return the same numbers on Claude, ChatGPT and Ask Oracle. The likely reading is that the AI model becomes interchangeable while lock-in moves to the ERP, which manages agent permissions and logs. Three of the SuiteAgents announced for six industries are available now, and Italy and the other remaining EMEA markets have no launch date yet.

Kadena Air Base is turning away every BYD, a brand on the Pentagon's June list, but its stated reasons (cameras, sensors, data transmission) apply to connected cars as a whole. Seoul, where at least 190 Chinese vehicles have already entered bases, has yet to write its own measures. Forecast: another US base in Japan or South Korea will adopt the same ban by March 31, 2027.

GPT-6.1 Terra halves the per-token price. OpenAI's figures show it also cuts cost per task on an agent benchmark by about two-thirds compared with GPT-5.6 Terra, so it uses about a third fewer tokens on the same test. An administrative workflow with 1 million input tokens and 200,000 output tokens costs $2 on Terra, $4 on Sol and about $20 at the Astra price implied by OpenAI's stated ratio, with every performance figure coming from the vendor.

Town raised $108 million at a $1 billion valuation. If that valuation is post-money, investors are buying roughly 10.8% of a company founded in late 2024, and its CEO admits ChatGPT and Claude can do "most" of what Town does. The price is a bet on onboarding SMBs and mid-market companies to AI agents, and the weak spot is a guided path that the model makers could build themselves.

DeepSeek and MiniMax have obtained open-source and overseas licenses under China's export controls of 16 October, according to a CCTV-affiliated channel. Chinese open weights keep flowing out and keep Western price lists under pressure, while training jurisdiction becomes a clause European boards must negotiate, with a ranking criterion that can move a model into the catalogue with every new release.

NVIDIA recommends a cluster of eight DGX Spark units to run the full GLM 5.3 on-premises, four times the two machines it says the quantized Flash version needs. With Cloud Functions on the way, serving local models through an OpenAI-compatible endpoint, local hardware can now compete with API contracts for steady agentic workloads.

Nvidia is releasing a 120-billion-parameter agentic model with open weights, optimized for every tier of its own infrastructure. It comes as the proprietary labs, all Nvidia customers, are also looking at Trainium, Cerebras and their own chips. For enterprises, lock-in may shift from the model lab to the GPU vendor, so the clause to negotiate is whether post-training carries over to other chips.