AI daily

2026-09-15

NVIDIA Advances AI Factory Efficiency as Google Debuts Gemini 3.8 Audio Models and Buildots Raises $130M for Construction AI

AI DAILY BRIEFING

The day's developments highlight a dual focus on optimizing AI infrastructure and expanding AI applications. NVIDIA, with partners like Emerald AI, Amazon's Annapurna Labs, d-Matrix, Lambda, and Pinterest, is driving efficiency in AI factories, shifting metrics to validated agentic tokens per megawatt. Google's new Gemini 3.8 Live and Extended Thinking models aim to enhance real-time voice and video interactions across 97 languages. In the construction sector, Buildots' $130M funding round, led by O.G. Venture Partners, supports AI adoption in a $16 trillion industry. Meanwhile, China's AI boom is pushing data centers to western regions with cheaper electricity, and startups like Poolday and Pave Finance are reporting significant adoption and funding. Tencent's open-sourcing of WeKnora provides enterprises with a private LLM knowledge platform.

NVIDIA and its partners are prioritizing AI factory efficiency, with Emerald AI's Conductor platform successfully reducing power consumption while maintaining high-priority workloads, and new collaborations aiming for up to 1.4x more tokens per megawatt.

Google's Gemini 3.8 Live and Extended Thinking models represent a significant step in real-time audio AI, supporting 97 languages and complex multi-step tasks, and are being rolled out across consumer, developer, and enterprise channels.

Buildots' $130M funding round, bringing total funding to $297M, signals growing investor confidence in applying AI to the $16 trillion construction industry, particularly for data center buildouts.

Watch next: Monitor how NVIDIA's efficiency metrics and partnerships translate into real-world deployments, and whether Google's new audio models drive adoption in live interactions. Also watch for further AI infrastructure investments in China's western hubs and the expansion of AI in construction and financial advisory sectors.

Featured

APriority81
NVIDIA AI Blog
1 sources

From Megawatts to Tokens: How NVIDIA Maximizes AI Factory Production

Emerald AI's Conductor platform, an NVIDIA partner, successfully deployed across thousands of NVIDIA GPUs to adjust data center power consumption in response to grid demand signals from Silicon Valley Power. The system reduced power from four megawatts to three while maintaining high-priority AI workloads, and has since responded to over 200 demand signals without failure. At the AI Infra Summit, NVIDIA's Ian Buck highlighted AI factory efficiency as a key infrastructure priority.

APriority81
NVIDIA AI Blog
1 sources

NVIDIA Showcases AI Factory Efficiency Advancements at AI Infra Summit

At the AI Infra Summit, NVIDIA VP Ian Buck highlighted new collaborations and platform advancements aimed at improving AI factory efficiency. Partners including Amazon's Annapurna Labs, d-Matrix, Emerald AI, Lambda, and Pinterest are integrating NVIDIA technologies such as NVHBM, NVLink Fusion, DSX MaxLPS, and Blackwell. The focus is shifting from peak performance to validated agentic tokens per megawatt, with NVIDIA's full-stack platform promising up to 1.4x more tokens per megawatt.

APriority78
Google AI (X)
2 sources

Google launches most advanced Gemini audio models: Gemini 3.8 Live and 3.8 Live Extended Thinking

Google announced two new Gemini audio models, Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, designed for more natural real-time voice and video interaction. Gemini 3.8 Live focuses on speed, scale, and cost efficiency with support for 97 languages and mid-sentence interruptions, while Extended Thinking adds parallel reasoning for complex multi-step tasks. Both are rolling out across consumer, developer, and enterprise channels, including Search Live, the Gemini API, and Gemini Enterprise.

APriority79
The AI Insider
1 sources

Buildots Secures $130M to Bring AI to the $16T Construction Industry

Buildots has raised a $130 million funding round led by O.G. Venture Partners, bringing its total funding to $297 million. The company's AI platform uses computer vision and eight years of proprietary construction data to create digital twins of projects, helping over 100 major firms like Intel and Digital Realty manage large-scale builds. The investment aims to accelerate AI adoption in the $16 trillion construction industry, particularly for data center buildouts.

APriority76
Kimmonismus (X)
1 sources

Poolday reports 700,000 agent runs and 100 million video edits

Poolday announced that its AI agent has completed 700,000 agent runs and over 100 million video edits. The agent takes a video brief, works with existing footage and brand assets, and assembles an editable, on-brand video, optionally using generative models. The company also raised $11M to automate video production and is now opening access to the agent for the first 50 users.

APriority76
The AI Insider
1 sources

Pave Finance Raises $15M Series A to Expand AI-Powered Portfolio Management Platform

Pave Finance, an AI-powered portfolio management platform for financial advisors, raised over $15 million in an oversubscribed Series A round at a $100 million pre-money valuation. The funding follows a $14 million seed round in September 2025 and will support expansion of the company's market-facing and engineering teams. Advisors using Pave now oversee more than $130 billion in assets across over 300,000 accounts, reflecting growing adoption of AI-driven portfolio personalization.

APriority77
meng shao (X)
1 sources

Tencent Weixin Team Open-Sources Enterprise LLM Knowledge Platform WeKnora

Tencent's Weixin team has open-sourced WeKnora, an enterprise-grade LLM knowledge platform that transforms scattered raw documents into three knowledge assets: a question-answering RAG knowledge base, an autonomous reasoning ReAct Agent, and a continuously evolving automatic Wiki. The framework serves as the core technology of WeChat's open dialogue platform and emphasizes private deployment, data autonomy, and enterprise features such as multi-tenancy, four-level RBAC, and audit logs.

SynthePulse InsightRead deep analysis
APriority79
量子位
1 sources

Honor launches MagicOS 11 with AI assistant YOYO capable of 100+ step tasks

Honor tonight launched MagicOS 11, upgrading its AI assistant YOYO into a system-level agent framework called YOYO Harness. YOYO can execute over 100 steps with a 90% task completion rate, 91.8% intent understanding accuracy, and access to 700 tools, up from 400. It supports cross-app long-horizon tasks, smart home and car control, conditional triggers, and proactive services across over 1,000 scenarios.

More

APriority77
Latent Space
1 sources

Can Skills Learned in Games Transfer to Real-World Work?

Good Start Labs, co-founded by CEO Alex Duffy, is turning games like Diplomacy into training material for AI models to teach strategic thinking and other transferable skills. The startup, spun out of Every with $3.6 million in funding, was inspired by a 2025 Twitch stream where OpenAI's o3 model won by planning a betrayal while Claude's Opus 4 refused to lie and lost. Duffy claims fine-tuning on Diplomacy improved model performance on customer support and industrial operations benchmarks.

APriority78
量子位
1 sources

Shengshu Technology Releases Vidu S2 Model with Real-Time Interaction and Video Editing

Shengshu Technology has released the Vidu S2 model, which includes two sub-models: Vidu S2-Avatar and Vidu S2-Editing. Vidu S2-Avatar upgrades real-time output resolution to 720p and supports receiving reference images during interaction to perform actions; Vidu S2-Editing enables real-time changing of clothing, background, characters, and style on playing videos. The model is fully open to the public on the day of release, marking a significant advancement in real-time interaction and editing for AI video generation.

APriority78
量子位
1 sources

Oracle lays off employees at 6 AM via email, accounts disabled at 4 AM

Oracle conducted a sudden layoff in Silicon Valley at 6 AM via email, with employee accounts disabled at 4 AM and Slack access cut between 5 and 5:30 AM, requiring same-day exit procedures. The layoffs affected about half of the CSS organization and the North American CSM team, while the company is actively hiring for cloud computing, data center, and AI infrastructure roles to support its growing AI orders. This move reflects Oracle's efforts to cut costs and redirect funds to data centers and AI computing power amid massive AI infrastructure investments and negative free cash flow.

APriority78
The AI Insider
1 sources

Ubtech Opens Humanoid Robot Factory in Liuzhou with 10,000-Unit Annual Capacity

Ubtech Robotics has started operating a humanoid robot factory in Liuzhou, China, with an annual capacity of 10,000 units and the ability to produce one industrial humanoid every 10 minutes. The 14,000-square-meter facility, built with Siemens Digital Industries Software, primarily manufactures the Walker S and Cruzr series and uses Cruzr robots for tasks like depalletizing, palletizing, and material transport. The factory also includes a 65-square-meter automated warehouse that can store 112 humanoids and increases space utilization by over 50%.

APriority79
InfoQ
1 sources

Arm Launches AI Portal: A Unified AI Software Entry Point for Developers and Agents

Arm officially launched the Arm AI Portal at the Arm Create Developer Conference in Shanghai. It is a unified entry point for developers and agents to discover, optimize, and deploy AI software on Arm compute platforms, covering cloud, edge, and physical AI scenarios. The platform initially supports language, speech, vision, and neural graphics workloads, with pre-optimized models including Alibaba Qwen, Google Gemma, and Ultralytics YOLO, and supports runtimes such as ExecuTorch, LiteRT, and ONNX Runtime. Arm says the goal is to lower the barrier for developers in model selection, target platform choice, and deployment optimization, with resources already available via Hugging Face and developer.arm.com, and coming soon to ModelScope.

APriority78
InfoQ AI/ML/Data Eng
1 sources

Grab's Agent Framework LLM-Kit Accelerates AI Agent Production Deployment

Grab has implemented LLM-Kit, a framework that standardizes over 500 internal agent services. The system enhances service integration, evaluation, and secret handling, reducing the time to deploy new AI agents from two weeks to one hour. It centralizes infrastructure management, allowing runtime tool discovery and flexible model integration, while maintaining operational control.

APriority79
InfoQ
1 sources

Kubernetes Promotes KYAML as a Safer, More Consistent Way to Handle Configuration Manifests

Kubernetes is encouraging developers to adopt KYAML, a stricter YAML subset designed to make Kubernetes configuration more explicit, predictable, and less prone to common YAML errors. Introduced as Alpha in v1.34 and upgraded to Beta and enabled by default in v1.35, it supports output via kubectl -o kyaml and conversion of existing manifests with yamlfmt tools. The article notes that as Helm, GitOps, and AI coding agents increasingly generate configurations, KYAML's strictness helps reduce ambiguity and makes machine-generated manifests easier to validate.

APriority77
DeepTech深科技
1 sources

Trace Intelligence raises 100M+ yuan to build custom 6nm edge inference chip

Trace Intelligence (淬思科技) has closed a new nine-figure yuan funding round led by a major fund, with Yunqi Partners participating and existing investors Monolith and Fudan Science & Technology Innovation continuing to back the company. The startup plans to define, design, and tape out a 6nm custom inference chip with customers, starting with edge voice recognition. Founder Pan Hongyang argues that as AI model architectures converge and move into distillation and deployment, custom inference chips for fragmented edge scenarios such as wearables, AIoT, and security will offer better PPA than general-purpose GPUs or NPUs.

Selected
16
Sources
11
Featured
8
More
8
Previous2026-09-14Next2026-09-16
All reports