SemiAnalysis

Nvidia's next-generation AI rack system is delayed to 2028 due to manufacturing challenges, according to SemiAnalysis. The delay raises concerns about Nvidia's rapid annual release schedule conflicting with production limitations.

A panic over excess AI computing capacity was labeled 'erroneous' by research firm SemiAnalysis after a global sell-off hit semiconductor stocks. Investors misinterpreted Meta Platforms' potential shift to cloud services as a sign of an AI hardware glut, fearing the company would flood the market with unused GPUs.

The article discusses potential grid constraints in the US due to projected growth of behind-the-meter datacenters exceeding 40GW by 2028. It highlights concerns about infrastructure challenges from increased energy demands.

New AI technology, including Nvidia's Blackwell systems, is expected to significantly increase efficiency and reduce AI token prices. Companies like OpenAI and Silicon Data have noted declining token costs, with a spending index showing a drop from 2.06 to 1.75 in early June. Unnamed AI infrastructure firms also predict cheaper tokens due to improved models later this year.

Building data centers in space could become cost-effective by the early 2030s as launch costs decline and terrestrial challenges like energy shortages and permitting issues persist. Elon Musk argues space-based data centers may become the cheapest option before the decade ends, supported by advancements in space robotics for maintenance, such as those developed by New York startup Icarus Robotics.

Anthropic's new Mythos-based AI models intentionally limit assistance for AI research tasks, sparking criticism from developers who accuse the company of withholding information and degrading performance without user awareness. The move has reignited debate over Anthropic's delayed release of the Mythos model earlier this year.

Building data centers in space could become cost-effective by 2040, though current launch costs make orbital compute four times pricier than terrestrial. By the early 2030s, space data centers may cost 30% more than Earth-based ones. SpaceX CEO Elon Musk argues energy shortages, grid constraints, and regulatory hurdles could make space the only viable option for scaling AI data centers.