Semi Doped

A deep dive into NVIDIA's $2B tie-up with Marvell and what it might mean for AI interconnects and multi-vendor racks. A lively debate on whether NVLink could become the dominant scale-up standard or if multiple protocols will persist. A look at the sharp DRAM and NAND price surge and how AI, new LPDDR buyers, and long-term contracts are reshaping memory markets. Analysis of Intel repurchasing its Ireland fab and the strategic signals behind the buyback.

They dig into a high‑stakes patent clash in the semiconductor world and what it could mean for companies. They unpack TurboQuant and how aggressive KV cache compression reshapes memory needs. They explore ARM moving into CPUs tailored for agentic AI and the practical challenges of cores, NUMA and memory supply. Short timelines and system designs get close attention.

They debate Jensen Huang's $250K-per-engineer AI token claim and whether companies will buy tokens or on-prem hardware. Micron's blockbuster earnings and a Vera Rubin design-in for HBM4 drive a deep dive into pin speeds, base-die tradeoffs, and big new fab bets. The conversation ends on optical interconnects, where microLEDs, VCSELs, and a new OCI-MSA standard vie as short-reach solutions.

They break down Nvidia's GTC keynote highlights and the shift in AI into training, inference, and agentic eras. They cover new hardware moves like Groq's LPX and Vera Rubin systems, plus rising CPU demand and novel interconnects. They dig into scale-up architectures, AI-native storage concepts, and the tradeoffs between building agents and buying SaaS.

A technical deep dive into Meta's MTIA custom silicon, why chiplets let them iterate every six months, and how inference is shifting toward gen AI. A lively look at agentic AI changing engineering roles and upskilling junior talent. An investigation of Applied Optoelectronics: vertical laser manufacturing, Amazon's $4B warrant, capacity builds, and whether history might repeat for laser suppliers.

They debate the optics vs copper showdown after a surprise $4B optics buy and bold claims that 400G per lane can run over copper. They unpack supply chain, U.S. fab moves, and laser technologies versus micro‑LED active cables. They also explore reliability, telemetry-driven link fixes, and a strategic theory about why big players are pushing copper messaging.

They trace the AI optical supply chain from substrates to finished modules. Key debates cover indium phosphide bottlenecks and export risks. Silicon photonics foundry strategies and wafer-size competition get heavy focus. Laser suppliers, vertical integration plays, and precision optical assembly dynamics are discussed. Fiber demand for AI data centers and timing uncertainties close out the conversation.

They explore why silicon photonics is surging and how optics beats copper for speed and distance. They outline scale-across, scale-out, and a new scale-up market for GPU connectivity. They dig into laser supply bottlenecks and indium phosphide limits. They highlight co-packaged optics, high-power lasers, and the rising role of optical circuit switches.

Runaway memory demand from AI is driving dizzying price spikes and squeezing lower-end device production. Massive hyperscaler CapEx is reshaping who buys DRAM, NAND, and high-bandwidth memory. Big cloud players are balancing GPUs, custom accelerators, and ad-driven AI strategies while betting on infrastructure and optics for the next wave.

Wayne Nelms, CTO of Ornn and former quantitative trader, is building a financial exchange to trade GPU compute and memory as standardized commodities. He discusses pricing dynamics, hedging with futures, memory trading, depreciation and residual-value products. Short, forward-looking conversation about how financial markets could reshape GPU utilization, data-center finance, and hardware lifecycle risk.