Latest
In July data from companies using Ramp's token management product, Fable 5 accounted for 6% of tokens purchased and 11.4% of spend on Anthropic. As AI spending grows, we examine the pricing and cost-effectiveness of the top-tier model.
As CPO mass production slips to 2028 or later, NPO—which achieves DSP-free power efficiency with field-replaceable optical engines—is emerging as a practical choice to meet optical interconnect demand in the late 2020s, backed by product launches from Broadcom, Alibaba, and Tencent, and the standardization of the Open CPX MSA.
Google shipped Gemini 3.7 Flash just 23 days after 3.6's debut. It boosts code evaluation scores and is priced at half the 3.6 launch rate through year-end, but the price will double in January 2027.
SK hynix has introduced AI agents into back-end production equipment, while Samsung cut verification time for a customer-specific SoC to two days. Alongside the speedup, human re-verification to prevent malfunctions and on-site AI infrastructure are becoming conditions for mass-production deployment.
Gregory Kurtzer, founder of Rocky Linux, has released a new project called OpenWALDO aimed at increasing transparency in AI training data. Rather than simply accumulating volumes of data, the project manages data through a trackable BOM format that records origin, licensing, and processing history, with the goal of building a foundation that can withstand legal audits.
Kyoto Prefectural Police arrested a man alleged to be the "first uploader" on Nyaa. An analysis tool that did not directly monitor the swarm captured information about the suspect, leading police to identify the individual.
CoreWeave signed a long-term contract through 2029 for use of NVIDIA's previous-generation A100 GPU. While this shows demand persisting beyond cutting-edge models, the specific contract terms and the equipment's activation timeline remain undisclosed, and it does not guarantee profitability across the entire older-generation fleet.
While the DRAM industry's quarterly revenue hits $97 billion, the spot market remains gridlocked as buyers and sellers fail to agree on price. Weekly data reveals where AI server demand-driven price surges collide with the purchasing limits of the consumer market.
NVIDIA's NeMo Switchyard curbs calls to expensive LLMs during managed evaluation, cutting costs by 74%. Given the accuracy drop and pre-alpha status, we examine the conditions under which enterprise adoption actually pays off.
Navitas filed a GaN patent infringement suit against Renesas on August 10, 2026, but behind it lies a three-way tangle of interests: Renesas can hold up to 39.9% of Wolfspeed, the very company that sued Navitas. This piece uses the Infineon v. Innoscience precedent to unpack the practical implications.
