- Kyle Green/Bloomberg NewsArtificial intelligence requires faster and faster chips, but it is getting more expensive to keep making them. Most chip packaging, the final step in semiconductor fabrication, is outsourced to companies specializing in assembly and testing, with cost as a key concern. One particular use case is in memory chips, especially with the explosion of demand from AI. Nvidia’s H200 AI chip, for example, integrates six HBMs with its own graphic processing unit. It might be harder to keep making chips smaller, but more-innovative ways to package them could still open many opportunities.

Source: Wall Street Journal June 16, 2024 12:46 UTC