Semi Doped

Huawei's Tau Scaling Law: Is the "EUV Killer" Real?

Lithography Masterclass

Cerebras IPO

Cerebras’s wafer-scale engine represents a paradigm shift in semiconductor design, replacing traditional diced GPUs with a single, massive silicon wafer that integrates nearly a million cores. This architecture leverages high-bandwidth on-wafer SRAM to deliver superior low-latency inference performance, though it neces...

Gimlet's Cross-Vendor Inference Cloud

Heterogeneous silicon and software orchestration are essential for scaling AI inference as agentic workloads become increasingly complex. Gimlet Labs addresses this by disaggregating inference tasks and routing them to the most efficient hardware, such as GPUs for high-performance compute and SRAM-based accelerators fo...

Power as the Next Physics Wall for AI

Data center power delivery represents the next major physics constraint as AI accelerator racks scale toward one-megawatt capacities. Current 48-volt architectures face extreme efficiency losses due to high current requirements and resistive heating, necessitating a shift toward 800-volt systems similar to those used i...

CapEx is just Memory Tax Now, Deepseek V4 NAND impact

Rising capital expenditure among major hyperscalers is increasingly driven by surging memory and storage costs rather than purely compute expansion. As AI models scale, high-performance HBM and NAND flash have transitioned from commodities to critical, non-fungible infrastructure components. Samsung and SanDisk are cap...

Masterclass on Google's TPU v8 Networking

Meta VP Matt Steiner on Ads Infra, GPUs, MTIA, and LLM-Written Kernels

Meta’s advertising business relies on a massive, vertically integrated infrastructure that balances retrieval and ranking models to process over three billion daily active users. Retrieval systems like Andromeda and ranking models like Lattice and the generative ads recommendation model (GEM) function within strict sub...

Dust Photonics, XPO, Nuvacore

Is Intel Finally Back with a $300B market cap? OpenClaw can Dream?

Reiner Pope (MatX): Designing AI Chips From First Principles for LLMs

MatX optimizes LLM performance by developing specialized chips that prioritize matrix multiplication and a hybrid SRAM-HBM memory architecture. Unlike traditional GPU-centric models, this design addresses the prohibitive costs of large-scale inference by eliminating weight-loading bottlenecks and enabling significantly...

$300M for 70K Viewers | Intel x Elon, OpenAI x TBPN, Citrini's Strait of Hormuz Stunt

NVIDIA's Marvell Strategy, Is Memory Different This Time?, Intel's Ireland Fab

ARM AGI CPU has entered the chat, TurboQuant thrashes memory stocks

MicroLEDs Ain’t Dead, Micron Snags Vera Rubin

Quick Takes: Nvidia Keynote at GTC

The Semi Doped podcast features Austin Lyons and Vik Shaker dissecting NVIDIA's GTC keynote, focusing on the shift towards agentic AI and its implications for compute needs. They explore Jensen Huang's framing of AI's evolution, from training to inference and now agentic AI, requiring exponentially more tokens. The dis...

Meta's Inference Accelerator & Applied Optoelectronics (AAOI)

The Great Optics-Copper Crossroads

The podcast explores the evolving landscape of optics and copper in data center infrastructure, particularly regarding scaling. It highlights Nvidia's $4 billion investment in optics companies like Lumentum and Coherent, driven by capacity needs, geopolitical considerations, and the pursuit of competitive pricing. The ...

Optical Supply Chain: What would you buy?

The Semi Doped podcast explores investment opportunities in the optics supply chain, particularly focusing on companies that have seen substantial growth in the past year. Vik and Austin play a game where they evaluate different companies, discussing their business models, competitive advantages (moats), and potential ...

Optical Networking Supercycle - ALL the Tech You NEED to know

The podcast explores the optics and networking super cycle, focusing on connectivity bottlenecks in data centers and the increasing demand for faster interconnects between GPUs. It draws parallels between the optics market and the HBM market, noting the limited number of suppliers and the resulting undersupply and pric...