Gavin Uberti, Sonic AI

Skip to content

Home Discover Ask Sonic Projects

Use with Claude or ChatGPT

Home Discover Ask Sonic Projects

Use with Claude or ChatGPT

Gavin Uberti, Sonic AI

Home/Discover/Gavin Uberti

G

Gavin Uberti

Person

18

Mentions

1

Episodes

18

Claims

Gavin Uberti, mentioned 18 times across podcast episodes and expert conversations analyzed by Sonic.

What Gavin Uberti has said

The company OmniML, where Gavin Uberti worked on kernels, was acquired by NVIDIA for hundreds of millions of dollars.

neutral·Gavin Uberti

The company Xnor.ai, where Gavin Uberti first worked on kernels, was acquired by Apple for $200 million.

neutral·Gavin Uberti

Gavin Uberti believes it is inevitable that individual data centers costing over one trillion dollars will be built.

bullish·Gavin Uberti

The company OmniML, where Gavin Uberti worked on kernels, was acquired by NVIDIA for hundreds of millions of dollars.

Expert perspectiveGavin UbertiJun 30

The company Xnor.ai, where Gavin Uberti first worked on kernels, was acquired by Apple for $200 million.

Expert perspectiveGavin UbertiJun 30

Gavin Uberti believes it is inevitable that individual data centers costing over one trillion dollars will be built.

SpeculativeGavin UbertiJun 30

TSMC was willing to work with Etched before the company had raised its $100 million Series A.

Expert perspectiveGavin UbertiJun 30

Etched's first-generation product is built on a different supply chain than NVIDIA's upcoming 'Rubens' chip, using 4nm instead of 3nm process technology and a different HBM.

Expert perspectiveGavin UbertiJun 30

Etched believes that winning AI hardware solutions must exhibit economies of scale, where serving 10 times more tokens costs less than 10 times more to deliver.

Expert perspectiveGavin UbertiJun 30

Etched believes three things are fundamentally required to scale inference: building a chip with the most flops in a given power budget, achieving the lowest latency between chips, and producing the c...

Expert perspectiveGavin UbertiJun 30

Etched's custom interconnect stack, which replaces everything above Layer 2 Ethernet, reduces chip-to-chip latency by more than 5x compared to alternatives.

Expert perspectiveGavin UbertiJun 30

The point-to-point chip latency on NVIDIA's Blackwell GPUs is approximately 4,000 nanoseconds.

Expert perspectiveGavin UbertiJun 30

Many traditional semiconductor venture funds missed investing in the wave of AI chip companies.

Expert perspectiveGavin UbertiJun 30

Etched's early investors were either believers in the market and team or highly technical groups like high-frequency trading firms that conducted deep technical diligence.

Expert perspectiveGavin UbertiJun 30

Etched's Series A funding round totaled $103 million, which was composed of soft commits that the company decided to accept.

Expert perspectiveGavin UbertiJun 30

Mark Ross, former CTO of Cypress Semiconductor which was sold for $9 billion, joined Etched as its full-time CTO after initially serving as an advisor.

Expert perspectiveGavin UbertiJun 30

To overcome a vendor delay, Etched sent a dozen of its top engineers to Bangalore, India, for six months to work on-site and run a 24-hour development cycle with the US team.

Expert perspectiveGavin UbertiJun 30

Synopsys provided Etched with emulators on favorable payment terms, effectively extending a loan to the startup.

Expert perspectiveGavin UbertiJun 30

Synopsys provided Etched with access to its emulators on favorable terms, structured as a long-term loan, before the company had raised significant capital.

Expert perspectiveGavin UbertiJun 30

For silicon chips, loading data from memory is very expensive while performing mathematical operations is very cheap, which is the opposite of how human neurons function.

Expert perspectiveGavin UbertiJun 30

Etched's second key technology is "cluster-scale memory," which creates a single memory pool from the SRAM and HBM of a full cluster.

Expert perspectiveGavin UbertiJun 30

Create a free account to see Gavin Uberti's full intelligence report — every claim, the relationship network, and AI Q&A across all sources. No card needed.

Get started free

Back to Entities Entity Detail