🎙️ NEW EPISODE: Intel's Earnings Disaster: Bad Luck, Bad Planning, or Both? Intel had a strong quarter. Its outlook told a different story. @vikramskr and @austinsemis unpack Intel’s Q3 2026 results and the gap between demand for its chips and its ability to deliver them. They cover the AI memory squeeze hitting PCs, the substrate and chiplet bottlenecks limiting server CPUs, and the fallout from the Fab 28 accident. Then they debate what Intel could have done differently and whether its long-term foundry story still holds. Presented by @CrusoeAI: Fine-tune and serve open models without managing a single GPU. When we sat down with Crusoe’s product team, Austin asked: what will you do with my data? Their answer: They don’t touch or see it. Your training data is encrypted so even Crusoe’s own admins can’t see it, and the fine-tuned model is yours. Deploy on Crusoe in one click, or download and run it anywhere. Upload a dataset, pick an open model, and pay per token rather than per GPU hour. Run a few versions, compare them, and download the one that works best. Accounts are free, and get $5 in credit when you sign up. Try Serverless Fine-Tuning: This episode is supported by @meter. You have the GPUs and the power. Now who builds the network, and who do you call when it breaks? Meter designs, deploys, and runs frontend, storage, and management networks on its own switches and its own network OS. One call reaches the team that designed the hardware and wrote the software. Chapters: 0:00 Intel's 35% Stock Drop 1:22 The Pre-Earnings Bull Case 4:03 Reason 1: Weak PC Market 12:25 The Flawed Q2 Pivot Plan 17:17 The IDM Model's Failure 20:23 Reason 2: Substrate Bottleneck 23:18 Chiplet Design's Fragility 25:18 Reason 3: The Fab 28 Explosion 40:46 The GPU Intelligence Gap 45:51 The Foundry Thesis Question 50:22 Packaging as a Hope Story 51:19 Four Lessons from the Meltdown Get more of Austin and Vik daily, free! Sign up: Connect with Vik and Austin: Vik's Paid Substack: Austin's Paid Substack: @austinsemis @vikramskr

Infineon Commits €1.1 Billion to Villach 300mm GaN Factory Vik: A monolithic bidirectional switch collapses two devices into one, so fewer stages in the solid state transformer. But it only cleared testing - series qualification is next, and 650V is the only part actually in the field today.

Maia 300's scale-up fabric allows prefill/decode disaggregation without crossing a scale-out network. The KV cache is passed all on the same fabric, from prefill to decode instances.

Maia 300 networking insights: 1. The NIC is on the die and built only for scale-up: low latency, high bandwidth between accelerators. 2. Tier 1 is copper inside the rack. Neighboring Maias talk the most (tensor parallelism), so about half of each chip's bandwidth stays in the server and half goes to the in-rack switch. 96 accelerators per rack. 3. Tier 2 is optics between racks, and it's still scale-up. Up to 128 racks, 12,288 accelerators, one domain, one communication stack. 4. Standard Ethernet switches and cables for the scale economics, with Microsoft's own transport on top. The NIC and transport are where Microsoft does its own engineering. 5. Why go past one rack? Bigger models, more aggressive sharding, and flexible job placement. The scheduler can put a job anywhere in 12,288 chips. 6. Fast data movement. Prefill and decode run as separate pools, and the KV cache moves between them over the scale-up fabric, never the slower scale-out network. For agents with long context, that's the expensive data movement.

🎙️ NEW EPISODE: @Microsoft Maia 300 with Saurabh Dighe Why Maia 300 makes memory bandwidth and capacity a first-order design principle. - Moving data to feed the compute engine can cost two orders of magnitude more energy than the compute itself - Maia 200 had six HBM3E stacks at 7 TB/s. Maia 300 has eight HBM4 stacks at 18 TB/s - Compute grows about 35% while rack-level memory bandwidth grows 5x - 7x token throughput per rack over Maia 200 at 2.5x perf per watt - Prefill and decode are disaggregated, and the KV cache moves over the scale-up fabric instead of the slower scale-out network - Memory interface chiplets "extend the shoreline" so the TSMC N2P compute die stays one efficient die This episode is presented by @CrusoeAI. Crusoe's Serverless Fine-Tuning trains an open model like Llama, DeepSeek, or GLM on your own data, and you pay per million tokens, not per GPU hour. About $2 buys roughly 5M training tokens on smaller models. Get $5 in credit to start: Chapters: 0:00 Intro: Saurabh Dighe 1:39 Microsoft's Silicon Philosophy 5:52 Workload Shift: Agentic AI 9:32 The Data Movement Problem 14:20 Crusoe Ad Read 17:47 Maia 300 Specs 24:07 Chiplets & HBM4 29:12 Prefill/Decode Disaggregation 30:33 Two-Tiered Network 35:45 NPO vs. CPO 36:46 Rack-Scale Power & Cooling 41:35 Agentic Kernel Optimization 49:08 What's Next: Maia 400 Get more of Austin and Vik daily, free! Sign up: Connect with Vik and Austin: Vik's Paid Substack: Austin's Paid Substack: @austinsemis @vikramskr

1.6T LPO from Linktel No DSP, and you can see pre-FEC BER is about 1e-9. Temps stay cool, at 35C due to no DSP The deployment is LPO is expected to be limited to specific use cases. LPO makes it difficult for the link to work with different vendors on either end. The link needs to be trained on a per port basis, and parts of the CTLE and amp are adjusted to close the links. This extra work is what LPO difficult and limits its widespread use

Optimize all the things, even down to interface IP!! When 1-2 applications drive all your revenue, it pays to optimize the full stack, from rack to chip. Standards-based IP gives way to application-optimized IP.

$CRDO had on display their microLED cable and the wound up cable had 10m length. A lot of interest is around their microVCSELs which they adopted for longer reach. Definitely high interest booth here at #OCP2026#

How does the Synopsys-OpenAI deal actually work? OpenAI brings: • Specialized model • Compute • Forward deployed engineers Synopsys brings: • Tools • Expertise • Skills to make GPT Synopsys an expert user of their own software

🎙️ NEW EPISODE: Synopsys Will Take a Royalty on Amazon's Chips and Sell GPT-Synopsys: Ravi Subramanian Ravi joins Austin to discuss why the company is moving to a royalty-based model for its custom "application optimized IP", starting with a billion-dollar deal with Amazon. Ravi also unpacks the OpenAI partnership and AI "agent engineers", and explains why physics-based simulation remains the ground truth. Insights: - Because 1-2 applications drive most of a hyperscaler's revenue, companies like Amazon now want custom "application optimized IP" instead of standards-based IP - The custom model has three parts. There is a license fee, a "customization fee" to prioritize scarce engineers, and for the first time a royalty on the final chip - The royalty model mirrors the Arm-Amazon partnership from a decade ago and fits 12-month design cycles. It also expands Synopsys's role to system-level bring-up - Customers can "hire" an AI agent engineer with a specific skillset, which fits existing budgets and team structures - GPT-Synopsys with OpenAI bundles a specialized LLM, the compute, and forward-deployed engineers - A CFET needs 200x more device equations than a planar FET, so physics-based simulators stay the ground truth This episode is presented by @CrusoeAI. Try Crusoe's managed inference for open models with $5 in free credits: Chapters: 0:00 Intro: The Amazon & OpenAI Deals 2:39 Application Optimized IP 6:04 The New Royalty Business Model 12:14 How Standard IP Differentiates 18:20 The Arm-Amazon Analogy 22:00 Synopsys's Role in Test Chips 26:06 The Three Stages of Chip Design 31:27 The 'Agent Engineer' AI Model 36:20 The GPT-Synopsys Partnership 39:27 In AI We Believe, In Physics We Trust Get more of Austin and Vik daily, free! Sign up: Connect with Vik and Austin: Vik's Paid Substack: Austin's Paid Substack: @austinsemis @vikramskr @Synopsys

Google has the phones, the TPUs, the models, and the people for on-device agents. So why is Meta Muse in the lead and most likely to do some edge orchestration? We argue Google has everything except the courage to ship a winning AI agent. What's holding them back?

Why would an AI company want to run models on your phone instead of their cloud? For model labs, the business is cloud inference. For Meta, the business is ads. They have a direct incentive to offload compute costs to your device. Who wins the battle for the edge?

Is Lumentum's laser moat disappearing? Co-packaged optics (CPO) shifts modulation onto silicon photonics chips (like Nvidia's), replacing complex EMLs with simpler, high-power (300-400mW) continuous-wave lasers. Essentially, a flashlight. Does the real moat still exist?

In this episode, you will notice that the video looks ... different. We hired a designer to come up with a beautiful looking theme that is inspired around campfire colors, outdoors looking at the night sky. We take aesthetic very seriously, because it provides a certain look to the end product. And we wanted it to have a premium feel. Our editor then took all the brand design, and implemented it here, and will continue to do so every episode from now on. We have a small team of dedicated, and hardworking folks who tirelessly help us with this podcast. We could never do this alone, and it always helps to take a step back and acknowledge those who make it possible. 🙏🏽

🎙️ NEW EPISODE: TSMC Will Buy High NA After All: ASML's New 6x12-Inch Mask Austin and Vik break down ASML's High NA EUV, whose anamorphic optics cut the exposure field in half, the new 6x12-inch masks that restore the full field size, and what the change means for the mask supply chain. - High NA uses anamorphic optics with 4x by 8x demagnification - A 12-inch mask restores the full 26mm x 33mm field size - Samsung targets High NA for DRAM in 2028, ahead of TSMC. TSMC is interested now though! - Intel is a million wafers deep on its High NA learning curve - Nearly 30% of ASML's revenue is recurring services income Chapters: 0:00 Episode Opening 2:23 The Rise of AI Agents 6:00 Instinct: The Autonomous Agent 10:50 Productizing AI for Mass Adoption 14:03 From AI Agents to ASML 24:29 The Photomask Stitching Problem 31:49 High NA's Anamorphic Optics Tradeoff 36:39 The Cost of a Halved Reticle 37:26 The 6x12-Inch Mask Solution 39:08 A Full Supply Chain Problem 41:17 TSMC, Samsung & Intel Timelines 48:24 Intel's EUV History Lesson Get more of Austin and Vik daily, free! Sign up: Connect with Vik and Austin: Vik's Paid Substack: Austin's Paid Substack: @austinsemis @vikramskr