Nvidia Orders 300K H20 AI Chips for China After US Export Ban Reversal

Nvidia has quietly placed an order for 300,000 H20 AI accelerators with Taiwan's TSMC, sources told Reuters on July 29, marking a sharp strategic pivot after Washington reversed an April ban on these export-compliant chips. The H20, developed as a lower-performance alternative to the top-tier H100 series for China, had been drawing down existing stockpiles of 600,000–700,000 units.

Despite White House assurances, Nvidia still needs formal approval from the U.S. Commerce Department before shipments can proceed, leaving the blockbuster deal in regulatory limbo. This move underscores the high-stakes interplay between U.S. commercial interests and national security in the tech rivalry with China.

India has collected 6,915 public comments on its draft Digital Personal Data Protection Rules 2025 as it moves to operationalize the 2023 Digital Personal Data Protection Act (DPDPA). The government says the rules aim to ensure a safe, trusted and accountable cyberspace, covering consent, user rights and parental controls, and will be backed by awareness campaigns and cybersecurity training initiatives.

Meanwhile, a bipartisan group of four Democrats and one Republican in a state legislature has drafted a bill to regulate A.I. in healthcare, mandating that patients be informed when A.I. is used, that a human retains decision-making authority and that bias be minimized across insurers, hospitals and clinicians. The proposal is pending formal introduction as lawmakers finalize the budget.

AI-powered cybersecurity solutions are on the rise. On July 29, London-based Intruder unveiled GregAI, an AI security analyst fully integrated with its attack surface management platform and available in beta to Cloud, Pro and Enterprise customers. Meanwhile, Cyware has expanded its Intelligence Suite with a private sandbox service, sector-specific threat feeds and dark-web domain sightings to streamline cyber-threat intelligence operations. In Seattle, startup Dropzone AI closed a $37 million Series B round to develop autonomous SOC agents that integrate with SIEM, endpoint and cloud security tools and can cut investigation times from hours to minutes. Together, these moves underscore growing demand for AI to boost efficiency across vulnerability management, threat intelligence and security operations centers.

Linux 6.17 merges major performance updates across data integrity and locking subsystems:

• CRC32C Boost: The VPCLMULQDQ instruction on Intel and AMD AVX-512 CPUs accelerates long-message CRC32C checks.


• 10× qdisc Change: A one-line switch to synchronize_rcu_expedited() in lockdep_unregister_key() cuts tc qdisc replace time from 13.195 s to 1.789 s, reducing network egress downtime.


• Crypto Refinements: FS-VERITY and AppArmor now use SHA-2 library functions instead of the older crypto_shash API, simplifying code and trimming overhead.

Microsoft Threat Intelligence disclosed Sploitlight (CVE-2025-31199), a critical macOS flaw that abuses the Spotlight importer plugin mechanism to bypass Apple's Transparency, Consent, and Control (TCC) protections. Attackers can drop an unsigned .mdimporter to silently exfiltrate files from Downloads, Desktop or Pictures and steal sensitive Apple Intelligence cache such as photo metadata and geolocation without user prompts.

Apple issued a patch in macOS Sequoia on March 31, 2025, and users should update immediately. Microsoft has bolstered Defender for Endpoint to flag anomalous .mdimporter installations and Spotlight indexing of protected directories.