AI deployment is meeting constraints at both ends of the market. Political opposition is putting a large share of planned US data center capacity at risk, while Google and Perplexity are pushing generative systems into longer media workflows and persistent business operations. Capital is also moving into specialized applications, with new funding for insurance automation and physics-based chemistry. In Washington, the incoming congressional majority is preparing to make AI security and critical infrastructure oversight a larger part of the federal agenda.
US data center plans face a wave of proposed pauses
About 40% of pending US data center capacity is planned for states where newly elected governors promised to stop or pause construction, according to Cleanview data cited by CNBC-TV18. Voters also approved or were on track to approve 46 local measures that would slow development, including at least 17 bans on large facilities. The analysis estimates that delays could keep as much as 10 gigawatts from coming online next year and reduce expected 2027 AI infrastructure spending by 5% to 7%.
The election result turns community concerns about electricity, water, tax incentives and land use into a planning risk for hyperscalers. Companies may shift projects to friendlier states, but moving capacity does not remove grid constraints or the need for public consent. Developers now have to treat local operating agreements as core infrastructure, with clear cost allocation and measurable limits, rather than as a late stage permitting exercise.
Google extends Gemini video generation to 15 seconds
Google made Gemini Omni 1.2 Flash generally available with native clips of up to 15 seconds, multi-turn editing and a 30% lower price than Omni 1.1. The model costs $0.07 per second for 720p video and accepts image, motion and audio references. Google says it improves character, product and scene consistency, supports one-second duration increments from three to 15 seconds, and can revise a clip conversationally without regenerating the entire sequence.
Longer clips and cheaper iteration make the model more practical for product explainers, social ads and catalog content. General availability across the Gemini API and business products also moves generated video from an experimental feature toward a repeatable production service. Outputs include SynthID watermarks, but teams still need review processes for rights, claims and brand accuracy before automated media enters a campaign.
Perplexity launches persistent multi-agent teammates
Perplexity introduced Teammates, a system of AI agents designed to own ongoing work such as campaign management, bug triage, metrics tracking and market research. Each teammate receives a standing role, a brief and escalation rules. It can plan tasks, bring in more agents as work expands and continue until a result is ready for review. The service is available on the web for Perplexity Max subscribers, with desktop and mobile versions planned.
The product reframes an agent from a one-time task runner into a durable operating role. That raises the value of monitoring, permissions and handoff rules because errors can compound across recurring work. The useful benchmark will be whether a teammate can maintain context and reliably escalate uncertainty over weeks, not whether it can complete one impressive demonstration.
Kalepa raises $75 million for insurance automation
Kalepa announced a $75 million Series B led by Updata Partners at a $470 million valuation. The New York company applies AI to underwriting, pricing, operations and portfolio decisions for insurers, reinsurers and managing general agents. Kalepa says revenue and customer count each tripled over the past year. The round brings its total funding to $91 million and will support expansion across the insurance value chain and into new markets.
Insurance is a strong test for specialized AI because decisions combine large data sets, regulatory duties and expert judgment. Funding is moving toward systems that fit those workflows rather than generic chat interfaces. Adoption will depend on audit trails, consistent treatment of applicants and evidence that automation improves risk selection without hiding the basis for a consequential decision.
UK sovereign capital backs physics-based chemistry AI
The UK's Sovereign AI fund joined an eight-figure financing round for London-based Potential Sciences and committed computing capacity to the startup. Potential Sciences uses machine learning and quantum chemistry simulations to generate molecular training data from physical laws rather than relying only on prior experiments. The company says its approach could make calculations used in chemical design 100 times cheaper. Investors also include Twin Path Ventures, Nauta Capital and Stripe CEO Patrick Collison.
The investment connects sovereign AI policy with scientific infrastructure. Compute access can matter as much as cash for a company building large chemistry data sets, and the UK is using both to keep technical capability in the country. The efficiency claim is company-reported and needs independent validation, but the strategy shows how governments may back domain models that support research and industrial capacity.
Congress prepares a larger AI security agenda
Cybersecurity experts and incoming congressional leaders expect the new Democratic majority to prioritize rebuilding CISA, AI security oversight and grants for local infrastructure operators. Lawmakers are likely to examine whether agencies and frontier laboratories have adequate safeguards, restore staffing and technical capacity at CISA, and revisit information-sharing protections. Comprehensive AI legislation remains uncertain, so hearings, agency oversight and targeted security rules may move first.
What matters today
Today's signal is that capability is no longer the only scarce resource. Google and Perplexity are making AI systems longer-running and easier to deploy, while Kalepa and Potential Sciences show capital favoring specialized workflows. At the same time, data center opposition and a renewed federal security agenda are raising the cost of operating without public accountability. The next phase of AI competition will reward companies that can secure power, permission and trust alongside better models.