Filtered to AI Hardware & Semiconductors · show all articles
Forward earnings multiple, backlog-to-revenue ratio, growth rate — the actual numbers behind Nvidia's valuation, laid out plainly, without a promotional or a doom-laden frame either way.
Groq built its entire business on the argument that GPUs are the wrong architecture for AI inference. Nvidia's reported move into Groq's technology is either the smartest hedge in the industry or a quiet admission the argument was right.
Individual accelerators keep getting faster. The copper connecting thousands of them together increasingly cannot keep up — which is why Nvidia is betting heavily on replacing that copper with light.
Big Tech's nuclear announcements make great headlines. The actual timeline for any of those reactors delivering power to an AI datacenter is longer than most coverage lets on.
Intel's Loihi, IBM's NorthPole, and BrainChip's Akida all post genuinely impressive efficiency numbers against classical GPUs. None of them is a shipping, general-purpose commercial product yet — and that gap is the actual story.
Gordon Moore observed a trend in 1965. The industry spent the next sixty years treating that observation as a target to hit rather than a fact to discover — which is the real reason it held for so long.
SK hynix and Samsung are Korean. The only remaining US-headquartered company producing DRAM and HBM at real volume is Micron — a supply- security story as much as a technology one.
Two of the largest cloud providers in the world run custom AI silicon that Marvell helped design. The company's own name rarely appears next to either chip's.
Co-packaged optics and denser accelerators raised local heat density past what air can remove. Liquid cooling isn't a premium upgrade anymore — it's a prerequisite for the technology this cohort's other briefings describe.
Nvidia and Marvell are both betting on optics to connect chips together. Lightmatter's bet goes one step further: using light for the connections inside a chip package as well.