compute

A PSA recommending that buyers of GPU compute include late fees on delivery as a standard contract clause in the current market, arguing it benefits everyone.

Oracle is halving its Always Free ARM compute limits, cutting Ampere A1 allowance from 4 OCPUs/24 GB to 2 OCPUs/12 GB, with automatic termination of excess instances starting August 18, 2026. The article explains the change and steps users should take to shrink or consolidate before the deadline.

SpaceX reported that its AI revenue grew to $2.6 billion, surpassing its space revenue, as the company shifts focus to AI compute with deals with Anthropic and Google. The company's IPO documents emphasize AI as its primary value driver.

Anthropic signs a reported $10B deal with AI cloud startup Volta to secure 133 MW of compute capacity over six years, with Bitdeer building a Norway data center powered by Nvidia Vera Rubin systems.

Modular announces ModCon 2026, its annual developer conference themed 'Compute Unlocked,' featuring keynotes, panels, and workshops on AI infrastructure. The event includes investor panels and hands-on Mojo programming sessions.

An analysis arguing that circular deals in AI (e.g., OpenAI-Microsoft, Nvidia-CoreWeave) signal the commodification of intelligence, drawing parallels to traditional commodity markets and warning of risks in deals that deviate from that model.

Elon Musk comments that Chinese companies could become leaders in AI if they had sufficient computing power, and predicts they will eventually have more compute resources.

Speculative post about next giant pre-training run potentially yielding superhuman AI within 12-14 months, leading to recursive self-improvement.

Ilya Sutskever's startup SSI announces a strategic partnership with NVIDIA, including a substantial investment that will enable a tenfold increase in compute over the next 12 months, signaling scaling of their AI research.

The tweet observes the accelerating pace of AI model releases, driven by more players and compute, with iterative improvements compounding to shape the next few years.

Grok 4.5 demonstrates surprising quality in Grok Build, suggesting xAI may be returning to competitiveness and potentially disrupting Anthropic's compute access from SpaceX.

Frontier AI data centers now consume more electricity than entire countries like Kuwait or Colombia, while the amount of compute available per human has reached that of a decent smartphone, highlighting the massive scale and energy demands of modern AI.

Modal aims to be a global aggregator of compute and resource manager, acting as a very large global computer.

This paper analyzes how different AI performance metrics (bounded vs unbounded) determine whether frontier AI capabilities remain concentrated among wealthy actors or diffuse to smaller models, with implications for regulation.

The tweet argues that the AI model layer is the least profitable, while compute, energy, and applications are where the money is, noting that Chinese open-weight models are eroding the margins of companies like OpenAI and Anthropic.

Modal announces integration with Claude Science, providing elastic compute for life sciences researchers with up to $100K in committed resources.

A commenter questions why China, despite securing massive compute resources, appears to still rely on distilling US AI models rather than training foundational models from scratch, wondering what the actual bottleneck is.

An observation that future technologies often start as seemingly trivial toys before evolving into foundational infrastructure for civilization.

Observations of a shift where large enterprises are increasingly seeking to secure compute and post-train their own models in-house, often on open-source GLM-5.2, highlighting the growing acceptance of open-source AI.

SpaceX acquires AI coding platform Cursor for $60 billion in an all-stock deal, aiming to leverage xAI's compute infrastructure and compete with Anthropic and OpenAI.