Filtered to AI Hardware & Semiconductors · show all articles
Every accelerator covered elsewhere in this cohort, however specialized, still computes digitally. Rain AI is building analog, brain-inspired silicon instead — the single highest-risk, highest-upside bet in this cohort's entire startup track.
IonQ, Rigetti, and PsiQuantum build genuinely different hardware for a genuinely different computing paradigm. This briefing covers what each is actually building, and why quantum and classical AI silicon are complements, not rivals.
Qualcomm built an empire on mobile chips it doesn't fully control the ceiling of. Moving into AI datacenter inference is a bet on a market with no such ceiling — and a much more crowded field of competitors already established there.
This cohort has covered wafers, EUV tools, and HBM stacks as the inputs that limit chip supply. Electricity access belongs on that same list now — not as a downstream side effect, but as a hard constraint on deployment itself.
Most accelerator startups covered in this cohort hedge toward flexibility. Positron went the other way — a chip that does inference and nothing else, betting that specialization beats generality even against Nvidia.
Optical interconnects, already shipping, move data between chips faster and cooler. Optical compute, still emerging, tries to do the actual arithmetic with light. Conflating the two overstates how far this technology has actually come.
A 30% conduction-loss reduction, an 80% jump in chips per wafer from a 150mm-to-200mm transition, and a slowdown in SiC demand hitting at the same time — onsemi's 2026 is a case study in a right bet, badly timed.
Software-defined vehicles are reshaping how much chip content each car needs — a structural tailwind NXP is positioned for, even as the broader automotive-chip cycle takes another year to fully turn.
Blackwell was a GPU with a name. Rubin is a whole platform — six distinct chips engineered to work as one system — and that shift says as much about where AI hardware is headed as any single spec.
Nvidia's market capitalization gets treated as a headline in itself. The revenue, backlog, and data-center figures underneath it are more informative than the number everyone quotes.