custom-silicon
(6)
Apple Sells You a Smaller World, and You Keep Buying It
It started in a garage and now runs on custom silicon, a services machine, and a billion-plus devices that all talk to each other. Here is how the walled garden actually works - and why leaving it feels like moving countries.
Broadcom
Broadcom Inc. (NASDAQ: AVGO) is one of the world's largest semiconductor and infrastructure-software companies, headquartered in Palo Alto, California. Led by President and CEO Hock Tan since 2006, it designs custom AI accelerators, Ethernet switching and routing silicon, wireless and broadband chips, and optical components, while also running a large enterprise-software business anchored by VMware, Symantec, CA and Brocade. Its products connect and power much of the modern internet, from smartphones and set-top boxes to the hyperscale data centers training today's largest AI models.
Marvell Technology
Marvell Technology is a Santa Clara-based fabless semiconductor company that designs the chips underpinning modern data infrastructure. Founded in 1995, it has pivoted under CEO Matt Murphy from consumer and mobile silicon into custom AI accelerators (XPUs), electro-optics, high-speed Ethernet switching, DSPs and storage controllers. Its parts move data inside and between the hyperscale data centers run by companies like Amazon, Microsoft and Alphabet, making it one of the key merchant and custom silicon suppliers powering the AI buildout.
Fabric Cryptography
Fabric Cryptography is a San Francisco hardware startup building the Verifiable Processing Unit (VPU) - a fully programmable custom-silicon chip with an instruction set designed specifically for advanced cryptography. Founded in 2021 by Michael Gao, Tina Ju, and Sagar Reddy, Fabric aims to do for cryptography what Nvidia did for AI: dramatically accelerate workloads like zero-knowledge proofs, fully homomorphic encryption, and multi-party computation compared with CPUs, GPUs, and fixed-function accelerators. The company has raised $39M total, including a $33M Series A in August 2024 co-led by Blockchain Capital and 1kx.
Tensordyne
Tensordyne is a Sunnyvale- and Munich-based AI hardware company building inference systems that use a hardware-native logarithmic number system to cut the cost and power of running large AI models. Formerly the computer-vision startup Recogni, it rebranded in September 2025 and in June 2026 unveiled Napier (TDN), a 3nm, air-cooled inference platform it claims delivers roughly 13x the throughput and 17x the efficiency of Nvidia's GB300 NVL72 rack. The company positions itself as a direct challenger to Nvidia in the market for profitable, high-throughput generative-AI inference.
Ricursive Intelligence
Ricursive Intelligence is a Palo Alto frontier AI lab using artificial intelligence to compress the slow, expensive process of semiconductor design. Founded in late 2025 by Anna Goldie and Azalia Mirhoseini, the co-creators of Google's AlphaChip, the company is building an end-to-end AI model that designs, verifies and closes silicon, aiming to create a recursive loop where AI designs better chips and those chips train better AI. It raised $335M across seed and Series A rounds, reaching a $4B valuation roughly two months after launch.