Topic
Explore the latest artificial intelligence news and research.
OpenAI accelerated enterprise revenue run-rate and API spending growth in Q3, but Anthropic still leads on annualized revenue. The two companies' figures use different metrics and can't be compared directly.
As AI chip TDPs exceed 1 kW and liquid-cooling adoption jumps from 6% to 53% in three years, thermal management has replaced raw performance as the top constraint in semiconductor design. From imec's STCO research to Samsung's zHBM concept and KAIST's AI design agents, here's a look at the front lines of the battle against heat.
DeepSeek's experimental vision model caps each image at 384 tokens and is available through three API formats. Self-reported benchmarks show it near Opus 4.8 on some metrics, but reproducibility and real-world accuracy remain unverified.
Anthropic will keep the 30-day retention period for its Mythos-tier models but move enterprise data storage to customer-controlled clouds, with Anthropic itself no longer holding the data, starting in fall 2026.
Reports indicate NVIDIA agreed to pay $6 billion for a license to Poolside's AI model-building infrastructure and extended job offers to 109 employees. The deal appears to prioritize the machinery behind iterative model development over any single finished model.
SK hynix and collaborators have published a peer-reviewed technical roadmap for co-packaged optics. The plan to extend optical connections eventually to shared memory pools would push AI infrastructure bandwidth design beyond HBM.
An examination of Google and Marvell's TPU-related semiconductor agreement based on SEC filings shows that most of the roughly $12.2 billion stock purchase right is tied to $120 billion in cumulative target revenue—and reveals differing levels of certainty across Broadcom, MediaTek, and AMD.
A single Reddit report claims Codex Plus's weekly limit dropped from roughly $160 to $80 in API-equivalent value. A look at NerfTrack's estimation method, comparison conditions, and OpenAI's official documentation shows why the claim can't yet be confirmed.
Waymo has publicly detailed the onboard computer of its sixth-generation Driver, showing how a custom 5nm ASIC handles sensor fusion alongside CPUs and GPUs to cut latency and keep driving through hardware failures.
OpenAI is testing with early customers a Zero Data Retention (ZDR) compliant method that examines related conversations while keeping raw content on customer-controlled infrastructure and returning only limited safety signals. The company plans a phased rollout and a technical white paper in September.
