▶ 1:50 · AI Hardware & Chip Architecture · Hyperscalers building custom accelerators to cut Nvidia dependence, interconnect emerges as critical bottleneck

episode briefing

Can Big Tech’s AI Chips Rival Nvidia?

2026-08-12 · 9 company · 5 thematic

sentiment

4 bull0 bear5 neu

speakers

patrick soheili

Co-founded Eliyan; previously worked in semiconductor industry; provides deep technical perspective on memory bottleneck, hyperscaler custom silicon, and interconnect landscape.

patrick

Guest on The Information podcast discussing AI chip competition, hyperscaler custom silicon, and rack-scale infrastructure; appears to be an executive at a company providing interconnect technology to hyperscalers and chip makers.

ramin farjadrad

Co-founded Eliyan, a chiplet interconnect startup that raised $145M at $1B valuation; focuses on die-to-die and optical rack-to-rack connectivity for AI systems.

ramin

Guest on The Information podcast discussing memory supply chain dynamics and semiconductor ecosystem; appears to be an executive at a company with strategic investments from memory suppliers, foundries, switch vendors, and major chip makers.

now playing · AI Hardware & Chip Architecture

Memory & Storagetailwindscore 8/10patrick soheili

Structural HBM bottleneck: three suppliers, years to expand, insatiable AI demand

Only Samsung, SK Hynix, and Micron can produce high-bandwidth memory at advanced nodes; all are at full capacity with no near-term relief because building new fabs takes years and billions,…

AI Hardware & Chip Architecturetailwindscore 9/10patrick soheili

Hyperscalers and merchant vendors converge on rack-scale integration to challenge Nvidia's system moat

Google (TPU v8-v10), AWS, Microsoft, Meta, Alibaba, Tencent, and Baidu are maturing custom accelerators, while AMD, Intel, and Arm are moving into rack-level systems to optimize the full da…

AI Hardware & Chip Architecturetailwindscore 9/10patrick

Hyperscalers building custom accelerators to cut Nvidia dependence, interconnect emerges as critical bottleneck

Major cloud providers (Google, AWS, Microsoft, Meta, Chinese giants) are developing proprietary AI chips to reduce costs, but interconnect technology remains a rare bottleneck that limits u…

Semiconductorstailwindscore 8/10ramin farjadrad

Interconnect emerges as the critical differentiator for custom AI accelerator utilization

Custom compute chips alone are insufficient; the bottleneck is interconnect efficiency between accelerators, and the technology to solve it is rare and concentrated in a few players (e.g.,…

Data Center Infrastructuremixedscore 7/10patrick

Rack-scale integration becomes three-way battleground between Nvidia, hyperscalers, and merchant semiconductor vendors

The industry is converging on rack-scale systems: Nvidia with proprietary racks, hyperscalers with custom silicon, and AMD/Intel/Arm offering merchant rack solutions; all aim to optimize th…

←