TPUs Via Cloud Next, Intel Earnings, Foundry Scarcity

Why It Matters

Google’s new TPUs and Intel’s earnings underscore a shift toward in‑house AI silicon, reshaping competitive dynamics and creating upside for firms that can overcome memory and foundry constraints.

Key Takeaways

• Google unveiled TPU v8 training and inference chips with HBM3 memory.
• TPU pods now double compute density, introducing “Boardfly” low‑latency interconnect.
• Intel beat forecasts, signaling a CPU resurgence amid limited foundry capacity.
• TSMC’s capacity constraints keep AMD reliant on Intel’s fab improvements.
• Memory wall remains critical; Google explores SRAM‑based inference solutions.

Summary

The episode covered three intertwined stories: Google’s Cloud Next reveal of next‑generation TPU hardware, Intel’s surprisingly strong quarterly earnings, and the broader semiconductor supply crunch that limits AMD and other rivals.

Google introduced the TPU‑8T training accelerator and the larger TPU‑8i inference chip, both built on HBM3 and SRAM‑M, with pod counts jumping from ~9,400 to 9,600 cores and a new “Boardfly” inter‑die network to cut latency. Intel reported revenue and EPS ahead of consensus, driven by a surge in CPU sales as data‑center demand rebounds and its own fabs gain capacity while TSMC remains constrained.

The hosts highlighted Google’s analogy to its early web‑indexing spend, arguing that today’s training spend will pay off once inference at scale ramps, and noted the shift to Axon x86 CPUs as head nodes for TPU clusters. Intel’s stock rallied on the earnings beat, and analysts cited the firm’s ability to add capacity as a rare upside in a market dominated by foundry bottlenecks.

For investors, the announcements signal a widening gap between companies that control silicon (Google, Intel) and those dependent on external fabs (AMD, Nvidia). Memory‑wall solutions and low‑latency packaging become differentiators, while Intel’s execution could translate into multi‑digit earnings growth as the CPU resurgence gathers momentum.

Original Description

In this episode of The Circuit, Ben Bajarin and Jay Goldberg dive deep into an action-packed week for the semiconductor industry. Ben shares his firsthand insights from Google Next, detailing the launch of the new TPU v5p and v5i (referenced as 8T and 8I) and Google’s strategic shift toward disaggregated training and inference silicon. The duo then pivots to Intel’s surprisingly strong earnings, discussing whether the "CPU resurgence" and foundry improvements signal an end to the company’s existential crisis. Finally, they analyze the "drama" from the TSMC Symposium regarding High-NA EUV adoption and debate the long-term durability of the current semiconductor bull cycle.

00:01 – Introduction

00:10 – Google Next: The TPU Family Expands

05:03 – The Memory Wall and Custom ASICs

07:44 – Google’s Historical Perspective on CapEx

11:11 – Gemini and the Enterprise UX Gap

15:24 – Intel Earnings: Better Than Feared

18:31 – The Foundry Thesis and Capacity Advantage

25:51 – Intel’s Roadmap: 18A and 14A

33:23 – Intel Foundry Financials

40:27 – TSMC Symposium and High-NA EUV Drama

44:23 – WFE Forecasts and Market Sentiment

47:06 – Is the Semi Cycle Durable?