💾 Chip craze
1. Key Themes
Theme 1: The AI Chip Rush Is Hitting a Manufacturing Ceiling
Nearly every major AI player is racing to design custom silicon to reduce Nvidia dependence — but the real bottleneck isn't design, it's production. The same narrow set of manufacturers (TSMC, Samsung, ASML) must serve all of them simultaneously.
"Building a custom AI chip can reduce dependence on Nvidia. It doesn't reduce dependence on the handful of companies capable of manufacturing the world's most advanced semiconductors."
"TSMC handles the majority of cutting-edge chip manufacturing, and while it is investing tens of billions of dollars to expand capacity, executives have repeatedly said demand outpaces supply."
"If you're just starting to design a chip right now, you won't see silicon for three years." — Stacy Rasgon, Bernstein
Theme 2: AI Model Releases Are Now a Government-Negotiated Process
The Trump administration has established a de facto case-by-case approval regime for frontier AI model releases — a new political and regulatory layer that every frontier lab must now navigate.
"The government and the world's most advanced AI companies are negotiating how people get access to powerful technologies case-by-case, in real time."
"AI firms and the government are operating before more concrete standards for releasing such models — called for in President Trump's latest AI executive order — have been finalized."
Theme 3: The AI Battleground Is Shifting from Models to Deployment
As frontier models converge in capability, the competitive advantage is moving downstream — to who can actually embed AI into enterprise workflows. OpenAI is acquiring its way into implementation.
"As frontier models become increasingly comparable, it's harder to win on model performance alone. Enterprises need to know how to use the tools."
"The next phase of the AI race may be defined by who can get businesses to use their AI tools rather than model releases."
Theme 4: The "Palantir Playbook" Is Being Adopted Across AI
Forward-deployed engineers (FDEs) — engineers embedded directly with enterprise customers — are becoming the go-to enterprise GTM strategy for AI companies.
"That strategy mirrors Palantir's long-standing approach of embedding engineers directly with customers to build software around their operations. (Northslope's founders come from Palantir.)"
"FDEs have the benefit of speaking both tech and business languages, helping bridge the gaps between teams and employees who want to use AI models for certain tasks but may struggle to prompt solutions themselves."
2. Contrarian Perspectives
Perspective 1: Custom Chips Don't Actually Solve the Nvidia Problem
The consensus narrative is that building custom chips = Nvidia independence. The article challenges this: you still compete for the same foundry capacity, packaging, high-bandwidth memory, and lithography equipment. And even companies with custom chips continue buying Nvidia GPUs.
"Companies trying to reduce their reliance on Nvidia are still competing for many of the same scarce resources: leading-edge foundry capacity, advanced packaging, high-bandwidth memory and lithography equipment."
"Even companies developing custom chips continue buying Nvidia GPUs for many workloads because Nvidia's hardware, networking and software ecosystem remain difficult to replicate."
Perspective 2: The Rising Tide Still Lifts Nvidia
The common assumption is that custom chip success harms Nvidia. In reality, surging AI demand means Nvidia's revenues can grow even as competitors take share — and the article suggests this dynamic is well understood by insiders.
"Nvidia suggested it would do $1 trillion in cumulative revenue from 2025–2027. If a competitor gets to tap into a small percentage of that pie, 'it could still be tens of billions of dollars.'"
"A rising tide of demand can also lift all boats: Most of the companies working on their own chips still use Nvidia GPUs too."
Perspective 3: ASML Is the Most Overlooked Monopoly in AI Infrastructure
Everyone focuses on TSMC and Nvidia, but ASML — a Dutch company — holds a singular, unreplicable chokepoint in the entire AI chip supply chain.
"Those factories also depend on lithography machines from Dutch company ASML, which is the only company that makes the most advanced tools needed to manufacture AI chips."
3. Companies Identified
Nvidia Description: Dominant AI GPU and ecosystem provider Why mentioned: The incumbent every major AI company is trying to reduce dependence on; still the benchmark for total cost of ownership Quote: "I want something in my pocket when I'm sitting across the table from Jensen negotiating." — Stacy Rasgon, Bernstein
TSMC Description: World's leading semiconductor foundry Why mentioned: Primary manufacturer for cutting-edge AI chips; demand already outpaces its expanding capacity Quote: "TSMC handles the majority of cutting-edge chip manufacturing, and while it is investing tens of billions of dollars to expand capacity, executives have repeatedly said demand outpaces supply."
ASML Description: Dutch manufacturer of lithography equipment Why mentioned: Sole producer of the most advanced chip-making tools; a hidden monopoly underpinning the entire AI chip supply chain Quote: "ASML...is the only company that makes the most advanced tools needed to manufacture AI chips."
OpenAI Description: Leading frontier AI lab Why mentioned: Two major developments: (1) GPT-5.6 cleared for broad release by Trump administration; (2) Deployment Company acquiring Northslope to expand enterprise AI implementation Quote: "The OpenAI Deployment Company agreed to acquire Northslope, an applied AI firm...marking its second acquisition focused on enterprise AI use since launch."
Northslope Description: Applied AI firm with Palantir-pedigreed founders Why mentioned: Being acquired by OpenAI's Deployment Company; brings forward-deployed engineer model to OpenAI's enterprise strategy Quote: "Northslope's founders come from Palantir."
Anthropic Description: AI safety-focused frontier lab Why mentioned: Multiple storylines: custom chip talks with Samsung; Mythos/Fable models restricted by Commerce Department; building an AI services company for mid-market enterprises Quote: "Anthropic is building an AI services company to help mid-sized businesses use Claude."
Broadcom Description: Semiconductor and infrastructure technology company Why mentioned: Partner to both Apple ($30B chip deal) and OpenAI (first custom inference chip) Quote: "Apple said today that it expects to spend over $30 billion in its expanded partnership with Broadcom that will lead to more than 15 billion chips being made in the U.S."
DeepSeek Description: Chinese AI lab Why mentioned: Latest entrant reportedly developing its own AI chips, expanding the global competition for scarce manufacturing resources Quote: "China's DeepSeek is the latest company reportedly working on its own AI chips, per Reuters."
Palantir Description: Enterprise software company known for deep customer embedding Why mentioned: Named as the strategic model OpenAI is replicating with its forward-deployed engineer approach Quote: "That strategy mirrors Palantir's long-standing approach of embedding engineers directly with customers to build software around their operations."
Samsung Description: South Korean conglomerate and semiconductor manufacturer Why mentioned: In talks with Anthropic on custom chip manufacturing; one of few leading-edge foundry alternatives to TSMC Quote: "Anthropic is said to be in talks with Samsung on its own chip, too."
Meta Description: Social media and AI company Why mentioned: Has a custom chip effort; also debuted Muse Image, a photo-generating model using public Instagram profiles Quote: "Meta debuted Muse Image, a photo-generating model that will power a range of new Meta AI features, including the ability to incorporate images of anyone who has a public Instagram profile."
Microsoft Description: Enterprise software and cloud company Why mentioned: Has a custom chip effort; also actively replacing OpenAI and Anthropic AI tools in its own products with its own models Quote: "Microsoft is continuing to replace AI tools from competing labs OpenAI and Anthropic with their own."
Intel Description: U.S. semiconductor manufacturer Why mentioned: Attempting to become a foundry competitor to TSMC but currently lagging in manufacturing capability Quote: "Intel is also aiming to be a so-called foundry, but its manufacturing processes have lagged in recent years."
Tomoro Description: AI deployment firm Why mentioned: OpenAI Deployment Company's first acquisition, preceding the Northslope deal Quote: "Northslope is the Deployment Company's second acquisition, following AI deployment firm Tomoro."
4. People Identified
Stacy Rasgon Description: Senior analyst at Bernstein Research, semiconductor specialist Why mentioned: Provided the two most quotable analytical insights in the piece — on why AI companies want leverage over Nvidia and the long lead times for custom chip development Quotes:
- "I want something in my pocket when I'm sitting across the table from Jensen negotiating."
- "If you're just starting to design a chip right now, you won't see silicon for three years."
Jensen Huang Description: CEO of Nvidia Why mentioned: Referenced as the counterparty that chip buyers want leverage against in supply negotiations Quote: "I want something in my pocket when I'm sitting across the table from Jensen negotiating." (Rasgon, referring to Huang)
5. Operating Insights
Insight 1: Enterprise AI GTM Is Now an Engineering Services Business
AI companies are discovering that selling models is insufficient — enterprises need embedded implementation support. The winning GTM playbook is to send engineers directly into customer organizations, not just sales teams.
"FDEs have the benefit of speaking both tech and business languages, helping bridge the gaps between teams and employees who want to use AI models for certain tasks but may struggle to prompt solutions themselves."
Implication for operators: If you're selling AI tools to enterprises, consider whether your customer success function needs to operate more like a professional services or consulting team.
Insight 2: Regulatory Approval Is Now Part of Every Frontier AI Launch Roadmap
AI labs can no longer assume a clean, global launch of powerful new models. Government review — including technical expert deployment to D.C. — is now baked into the release process.
"Testing was done by the Center for AI Standards and Innovation within the Department of Commerce, with OpenAI sending technical experts who have remained in D.C. to address potential questions."
Implication for operators: Build government relations and regulatory review timelines into product launch planning for any frontier or near-frontier AI capabilities.
Insight 3: Supply Chain Positioning Is the New AI Moat
For chip-focused investors and operators, the leverage point is not chip design (which AI is commoditizing) but securing allocation from constrained manufacturers — foundry capacity, HBM, packaging.
"Custom chips can also be tailored to specific workloads, potentially making them more efficient and less expensive to operate."
"Companies trying to reduce their reliance on Nvidia are still competing for many of the same scarce resources: leading-edge foundry capacity, advanced packaging, high-bandwidth memory and lithography equipment."
6. Overlooked Insights
Insight 1: The OpenAI Deployment Company Has $4 Billion Earmarked for Acquisitions — and Has Only Used a Fraction
The Deployment Company launched with $4 billion specifically to fund M&A. With only two acquisitions completed (Tomoro and Northslope), the acquisition runway remains substantial — signaling more deals are likely coming as OpenAI consolidates the enterprise deployment market.
"OpenAI's deployment arm, which is majority-owned and controlled by OpenAI, kicked off with $4 billion under its belt to fund acquisitions, but the terms of this deal weren't disclosed."
Insight 2: The Consumer Backlash Against AI Customer Service May Be a Real, Monetizable Signal
Buried in the newsletter's closing item: TikTok users are going viral celebrating human customer service representatives, with some forming personal relationships with reps who helped them. This consumer sentiment — largely ignored in the AI-everywhere narrative — may represent a durable differentiation opportunity for companies that preserve or restore human touchpoints.
"It's an example of the desire for connection and community that bucks the broader AI everywhere all at once vibe."