
Jane Street's 13 Billion Dollar Crusoe Deal Was Already Loan Collateral
Jane Street committed $13 billion to Crusoe for AI compute, a contract reportedly pledged as loan collateral before it was even announced.

Jane Street committed $13 billion to Crusoe for AI compute, a contract reportedly pledged as loan collateral before it was even announced.

ChatGPT, Claude, and Grok all suffered outages within the same five-hour window on 3 September 2026. Each vendor named a separate cause, but all three lean on Microsoft Azure - here is what that means for DORA and NIS2 concentration-risk registers.

Cloudflare's new Vulnerability Discovery and Remediation service pairs Managed Defense network context with OpenAI's Daybreak and GPT-5.6 Cyber models to hunt vulnerabilities inside customer codebases. It proposes patches and evidence, not automatic fixes, but the source-code access itself is the real due-diligence question for EU operators.

Microsoft's Project Zenith lets Windows run AI models over 30 billion parameters locally, but only on hardware with 64GB or more of memory.

Anthropic's $35 billion cloud deal with Nvidia-backed Lambda puts Nvidia on three sides of one transaction. For EU and UK operators, that makes independent verification of AI capacity claims harder, not easier.

Cornell researchers cut the temperature needed to grow tantalum qubit chips from over 400C to 200C by swapping the sputtering gas from argon to krypton, bringing qubit fabrication within the thermal limits of a standard semiconductor foundry for the first time.

SK Hynix broke ground on a $4 billion Indiana HBM plant, but wafer fabrication, the real bottleneck, stays in South Korea through at least 2029.

Data center construction hit 50 billion dollars in April 2026. Building-trade unions are partnering with tech firms while nurses and teachers back a moratorium.

Nvidia paused parts of its AI Compute Partnership financing program weeks after launch. Its own SEC quarterly filing puts commitments at $36 billion.

Nvidia's new NVHBM memory debuts on AWS's Trainium4, the chip Amazon marketed as the escape route from Nvidia dependency. What that reversal means for EU and UK buyers of cloud compute.

Nvidia's Q2 FY2027 call put a number on the memory shortage: margin guidance fell to 71-72 percent while FY2028 growth guidance rose to about 70 percent.

Memory prices have exploded because two separate bottlenecks are compounding at once: DRAM fabs chasing HBM margins, and TSMC's CoWoS packaging capacity. Here is how both work.
Page 2 / 11
One considered note on infrastructure, governance, and measurement, most mornings. No theory.