Semis, infra, optics/networking, physical AI, and more from @austinsemis. If you like Stratechery and SemiAnalysis, you're in the right place
United States
- Tweets 255
- Following 4
- Followers 4,859
- Likes 652
Pinned Tweet
Advanced Packaging: Intel's EMIB vs TSMC's CoWoS
chipstrat.com/p/advanced-pac…
5
52
253
95,424
Chipstrat retweeted
🎙️ NEW EPISODE: MicroVCSELs vs MicroLEDs, a MicroExplainer
Austin and Vik host an explainer on the two competing microemitter technologies for AI interconnects, microLEDs and microVCSELs. They break down Credo's surprising pivot at OCP from its microLED acquisition to a microVCSEL product, and the physics that give microVCSELs better speed and reach.
- The new battleground is the 5-50 meter "reach gap," too long for copper limited to ~7 meters and too short for expensive, supply-constrained InP optics
- MicroLEDs are limited to ~3-5 Gbps over 5 meters because their broad light spectrum causes chromatic dispersion that smears the signal
- Hitting 1 Tbps with microLEDs takes ~300 slow lanes, and coupling hundreds of fibers adds a lot of manufacturing complexity
- MicroVCSELs can hit 50 Gbps+ over 50 meters. Stimulated emission gives a narrow beam with less dispersion, for 10x the speed and reach
- MicroVCSELs use a mature Gallium Arsenide (GaAs) process on 8-inch wafers, shared with cell phones, for a cost and scale advantage over InP
- MicroLEDs' niche is heat. Gallium Nitride (GaN) is highly temperature-insensitive, a candidate for short-reach optics inside a hot server tray
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: semidoped.com/crusoe-x
Chapters:
0:00 The Credo Pivot
3:49 The Talk of OCP
8:13 The 5-50 Meter Reach Gap
12:52 How MicroLEDs Work
21:10 MicroLEDs' Parallelism Problem
22:55 How MicroVCSELs Work
26:57 MicroVCSEL Performance Specs
31:51 The Physics of VCSELs
33:35 The Market Chooses MicroVCSELs
35:45 MicroLEDs' Niche: Heat
38:16 Teasing Future Interviews
Mentioned in this episode:
- PicoJool’s Al Yuen: The Case for GaAs VCSELs in Scale-Up Interconnects (Semi Doped episode): daily.semidoped.com/p/picojo…
- A MicroLED Field Trip to Avicena (Vik's Newsletter, paid): viksnewsletter.com/p/a-micro…
- MicroLEDs vs. Lasers: The Linewidth Tradeoff (Vik's Newsletter, free): viksnewsletter.com/p/microle…
- Understanding Energy Bands in Semiconductors (Vik's Newsletter, free): viksnewsletter.com/p/energy-…
Get more of Austin and Vik daily, free!
Sign up: daily.semidoped.com/
Connect with Vik and Austin:
Vik's Paid Substack: viksnewsletter.com
Austin's Paid Substack: chipstrat.com
@austinsemis @vikramskr
6
14
2,285
Dell's Tailwinds.
Why I drive my agents run from a Dell XPS, and fine-tuned models for Enterprise AI
newsletter.chipstrat.com/p/d…
1
561
Chipstrat retweeted
🎙️ 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: semidoped.com/crusoe
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: daily.semidoped.com/
Connect with Vik and Austin:
Vik's Paid Substack: viksnewsletter.com
Austin's Paid Substack: chipstrat.com
@austinsemis @vikramskr @Synopsys
3
7
38
14,250
Chipstrat retweeted
Fun chat! Check it out!
Today's guest is @austinsemis, the writer behind @chipstrat.
Austin started his career as an engineer at Intel before moving into startups and independent analysis. Today, he writes about semiconductors, optics, networking, and physical AI, bringing an engineer's perspective to what's happening across the compute stack.
We're thrilled to have him join us for a conversation on the state of the chip industry. Give it a listen!
2
7
1,358
Is EDA dead? My takeaways from @Synopsys Investor Day and the GPT-Synopsys deal with @OpenAI.
chipstrat.com/p/is-eda-dead
1
4
19
4,181
Chipstrat retweeted
Aw yeah! @semidoped is now presented by @CrusoeAI 🚀
Love the pod? Support us! Sign up for Crusoe's managed inference at semidoped.com/crusoe and get $5 in free credits. You can even fine-tune your own model with just a few clicks!
🎙️ NEW EPISODE: Every Muse User Gets 2 CPUs and 8GB in Meta's Cloud. Your Phone Already Has That!
Meta gives every Muse user a cloud VM. That gets expensive at a billion users; meanwhile, your phone is sitting idle.
- Model labs earn on cloud inference. Meta's ads pay for it.
- AI PCs flopped with no use case. Agents give the edge one.
- Qualcomm bought Modular so Mojo can run agents on any chip.
- Your computer is for your AI to use, not just for you.
This episode is presented by @CrusoeAI. Try Crusoe's managed inference for open models with $5 in free credits: semidoped.com/crusoe
Chapters:
0:00 Who Makes Local AI Happen?
1:25 Making the Intro with AI
4:43 Qualcomm's Agent Vision
6:54 Why AIPCs Failed
8:45 The Phone as a Portal
13:52 Meta's Incentive for Edge AI
24:05 The End of Apps & UIs?
32:31 The Ideal Personal Agent
34:25 The Edge as a Context Layer
38:17 Qualcomm's Modular Acquisition
43:16 The Ecosystem of You
48:01 Google's Missed Opportunity
Get more of Austin and Vik daily, free!
Sign up: daily.semidoped.com/
Connect with Vik and Austin:
Vik's Paid Substack: viksnewsletter.com
Austin's Paid Substack: chipstrat.com
@austinsemis @vikramskr
2
2
25
5,050
What does Jev from @typesafeai imply for the AI infra buildout? After all, there's no decode needed...
chipstrat.com/p/ai-with-no-d…
1
2
21
13,036
Chipstrat retweeted
🎙️ NEW EPISODE: Why calls to "pace" AI are a red herring for GPU demand.
AI safety work doesn't slow the hardware buildout.
The real governor on AI is data center construction pushback.
Today's sponsor is G2i — expert training data, evals, and RL environments for AI labs: fandf.co/4wZvKHS
Chapters:
0:00 Introduction: Pacing AI
1:36 Recapping Dario Amodei's Essay
7:42 Hot Take: Anthropic vs. OpenAI
11:34 The John Deere Playbook
15:22 The Problem with Interpretability
17:53 Pulling the Plug
19:03 The Real Bottleneck: Data Centers
22:03 The Geopolitical Blind Spot
23:41 Counter-narratives: Meta & Open Source
27:26 The Investment Thesis: Pacing Increases Compute
32:29 The Financial System Analogy
36:05 AI Safety as the New Cybersecurity
Get more of Austin and Vik daily, free!
Sign up: daily.semidoped.com/
Connect with Vik and Austin:
Vik's Paid Substack: viksnewsletter.com
Austin's Paid Substack: chipstrat.com
@austinsemis @vikramskr
1
2
12
7,967
Product innovation from @meta, @bot, and Instinct show a path for normie adoption of agentic AI.
And that's great for the AI infrastructure supply chain.
chipstrat.com/p/a-billion-ag…
2
7
3,357
Why do most semiconductor markets have ~3 main companies?
chipstrat.com/p/oligopoly-eq…
2
24
11,075
Chipstrat retweeted
AI diffusion needs open competitive frontier models
America needs agentic AI to diffuse.
Diffusion needs competitive frontier open models.
Nvidia + Poolside is a step in that direction.
chipstrat.com/p/open-weights…
$NVDA
1
2
9
3,606
America needs agentic AI to diffuse.
Diffusion needs competitive frontier open models.
Nvidia + Poolside is a step in that direction.
chipstrat.com/p/open-weights…
$NVDA
2
3
15
17,706
Here's my "NPO State of the Union" report.
We heard a lot of talk about NPO on earnings calls, less about CPO, so I dug in to see what I could find.
chipstrat.com/p/npo-state-of…
9
6,161
As enterprise AI usage grows, so does token usage.
AMD’s @MadhuR_PDX explains how intelligent model routing can match requests to the right AI resource to help enterprises optimize growing token costs without compromising output quality.
Watch his interview: bit.ly/4bGsXfe
48
7
73
32,507
A new metric for the agentic AI era: threads per megawatt.
AMD’s @MadhuR_PDX explains how AMD EPYC processors are designed to maximize concurrent workloads within existing data center power limits.
Learn more: bit.ly/4bGsXfe
6
16
118
73,285
There’s no one-size-fits-all SKU for agentic AI.
“The customer environments are diverse, the workloads are diverse, the constraints are diverse.”
AMD’s @MadhuR_PDX explains why end-to-end agentic workflows will continue to require different compute configurations: bit.ly/4bGsXfe
6
19
80
28,589
Chipstrat retweeted
Very interesting chat with @TensordyneInc
- Reusing HPE Juniper scale-up network
- Sub-reticle chips
- log math (adds instead of multiplies)
- air-cooled
- partnership with Broadcom
And fun history from R K
🎙️ NEW EPISODE: Tensordyne co-founder and CPO R K Anand joins Austin to break down an AI inference rack built on a router-based scale-up fabric and logarithmic math.
- HPE Juniper's 7th-gen router fabric provides the scale-up network at 1-2 microseconds of latency
- Log math turns expensive multiplications into cheap adds
- The compute die stays below the reticle limit, leaving room for a big on-chip SRAM cache
- 72 chips in a 13U chassis at 30 kW, a quarter of the space and power of an NVL72
- Air cooling opens brownfield enterprise and telco data centers
- One system runs both pre-fill and decode
@TensordyneInc @_rk_anand_
Chapters:
0:00 Introducing Tensordyne
5:32 The Juniper vs. Cisco Playbook
11:29 Origin Story: Automotive Power Constraints
15:37 The Secret Sauce of Log Math
18:02 Pivoting to the Data Center
22:08 Leveraging a Router Backplane for AI
27:22 Why Router Fabrics Suit MoE Models
34:12 The Three Phases of Inference Hardware
37:40 How One Chip Handles Pre-fill & Decode
40:34 The 'Too Good to Be True' System Specs
43:31 Go-to-Market: The Air-Cooled Advantage
48:21 De-risking with Strategic Partnerships
52:37 Solving the Software Problem with AI
Get more of Austin and Vik daily, free!
Sign up: daily.semidoped.com/
Connect with Vik and Austin:
Vik's Paid Substack: viksnewsletter.com
Austin's Paid Substack: chipstrat.com
@austinsemis @vikramskr
1
5
16
4,415
Agentic AI is changing the compute requirements behind AI infrastructure.
AMD’s @MadhuR_PDX joins @AustinSemis to discuss why there’s no one-size-fits-all approach, from CPU and GPU workloads to concurrency, tokenomics and an open compute portfolio.
🎥 Watch Bit by Bit: piped.video/watch?v=L7NPjx97…
6
14
92
47,355
