Infrastructure & Hardware B

Simon Willison released condense-json 1.0, a small Python library that shrinks JSON by replacing repeated strings with a short replacements map (e.g. mapping a key to a recurring phrase). Now a year and a half old, it graduates to a stable 1.0 with sensible, non-disruptive fixes.

Meta raised its 2026 capex forecast for AI data centers to $130-145 billion, underscoring the rising cost of scaling generative-AI infrastructure. Shares fell as free cash flow declined, highlighting how heavy AI spending weighs on near-term returns.

Simon Willison discussed OpenAI's 'ten advances in mathematics and theoretical computer science,' noting that just days earlier Anthropic had reported similar discoveries. The post reflects on a growing trend of frontier AI models contributing to open problems in mathematics.

A sponsored piece argues that rising densities and a growing variety of equipment and services are driving complexity in AI data centers. It contends the market needs fully integrated solutions to manage this complexity rather than piecemeal components.

Simon Willison highlighted DeepSeek-V4-Flash-0731, the latest release in DeepSeek's V4 family, described as having substantially enhanced capabilities. The fast, lightweight-oriented model underscores the continued momentum of open-weight AI model development.

NVIDIA describes co-designing model attention with hardware to speed up interactive long-context inference. As agentic and long-context workloads grow, attention takes a larger share of inference time, and the approach targets that bottleneck for faster serving.

Simon Willison introduced smevals, a small evaluation suite for testing models, prompts, and harnesses. Built in collaboration with Jesse Vincent's Prime Radiant applied AI research lab, the framework aims to help answer questions about the capabilities of different AI models.

AWS reported its fastest growth since 2021 as cloud sales boomed, according to DatacenterDynamics. Amazon is ramping up data center construction, with annual capital expenditure expected to reach $220bn, driven in part by rising AI memory costs.

NVIDIA released Video Codec SDK 13.1, adding zero-copy transcoding, AV1 B-frame support, and frame-accurate seeking. The update targets accelerating demand for high-quality video across industries, from immersive streaming to media pipelines.

DatacenterDynamics argues that emerging 'next wave' data center markets are central to Europe's push for data sovereignty. Amid policy and investment moves to keep data within the region, the piece analyzes how these growing markets shape the sovereignty debate.

Elon Musk confirmed a fourth xAI-linked data center in Memphis and said the company has begun removing gas turbines flagged as 'illegal,' per DatacenterDynamics. The moves underscore the environmental and permitting tensions accompanying rapid AI compute buildout.

Amazon and Duke Energy were accused of evading the Clean Air Act at a data center under development in Hamlet, North Carolina, DatacenterDynamics reported. The complaint centers on separate applications to deploy a combined 649 diesel generators at the site.

CenterPoint Energy raised its investment plan by $1.2bn as its data center load pipeline expands. The utility expects to energize up to 8GW of data center demand in the Greater Houston area by 2029, reflecting surging power needs from AI computing.

GCM expanded its range of high-performance heat sinks aimed at liquid-cooled data centers. The lineup targets the growing cooling demands of increasingly power-dense AI servers, supporting more efficient thermal management in modern facilities.