PixVerse Raises $439M at Over $2B Valuation as Video Generation AI Competition Heats Up
PixVerse, a video generation AI startup, has completed a $439 million funding round, pushing its valuation above $2 billion (roughly 310 billion yen). The new capital will go toward expanding the company's "world model" technology and acquiring customers across regions.
With Runway, Pika, and OpenAI's Sora battling it out in the video generation space, the key question is how far PixVerse can differentiate itself with its newfound war chest. A world model is a next-generation AI approach that integrates the understanding and prediction of physical environments, and it is expected to enable applications that go beyond mere video generation.
Nous Research in Talks for Funding at $1.5B Valuation as Hermes Agents Draw Attention
Nous Research, known for its open-source "Hermes" series of AI models, is reportedly in talks for a funding round at a $1.5 billion (roughly 233 billion yen) valuation. The round is expected to raise at least $75 million, with Robot Ventures leading and notable investors such as USV participating.
Nous Research has earned strong trust within the open-source community, and the Hermes series is drawing attention as models optimized for agentic use cases. Its value as an open alternative to commercial closed models appears to be resonating with investors.
FP4 Attention Kernels for NVIDIA B300 Achieve 1.69x Speedup Over FA4
A new set of FP4 (4-bit floating-point) attention kernels optimized for NVIDIA's next-generation B300 GPU has been announced, delivering up to 1.69x speedups compared to the existing FlashAttention-4 (FA4).
This is a result that translates directly into lower inference costs. FP4 quantization is a technique that dramatically reduces memory usage and compute while preserving accuracy, giving it significant importance for running large-scale models in production. As kernel optimization tailored to the B300 architecture advances, the practicality of local LLMs is expected to improve further as well.
Accenture and OpenAI on the Breakthrough for AI Transformation at Japanese Companies
ITmedia has published an in-depth interview about the Accenture–OpenAI partnership. The theme is how to break through a common challenge: many Japanese companies stall AI at the proof-of-concept or partial-efficiency stage and never move on to company-wide transformation.
Accenture believes that combining OpenAI's technology with business process reform spanning departments can achieve organizational transformation that goes beyond simple efficiency gains. Whether AI can serve as a catalyst for cross-functional change within the siloed organizational structures typical of Japanese companies looks set to become a key factor shaping future competitiveness.
Uber Makes a Full-Fledged Entry into Autonomous Driving Data Business with "AV Labs"
Uber's product lead Sachin Kansal spoke to TechCrunch about the company's autonomy strategy. Through its newly established data operations division, "AV Labs," the company is said to be building the data infrastructure essential for improving autonomous driving accuracy.
On its relationship with Waymo, Kansal described it as "getting complicated," while making clear that Uber will not own autonomous vehicles itself and will maintain its partnership model. He also touched on how AI shows up in the rider and driver experience on its ride-hailing platform — a sign that Uber is increasingly strengthening its identity as an AI company.
---