.png)
.png)
April 11, 2025
The Lithography Inflection Point - How Two Different Approaches to EUV Could Reshape the Semiconductor Industry
By
Kristal Investment Desk
How Jensen Huang is redefining computational progress by turning power constraints into strategic advantage in the AI era
In October 1997, Michael Dell, founder and CEO of Dell Computer, was asked what he would do if he were in Steve Jobs' position at the struggling Apple Computer. His response was unequivocal:
Dell's perspective made perfect sense within the PC industry paradigm of the time. His company had pioneered a revolutionary business model: selling standardized computers directly to customers, bypassing retailers, and assembling machines only after receiving orders. This approach minimized inventory, maximized capital efficiency, and delivered precisely what customers wanted at lower prices than competitors.
The key to Dell's success wasn't making components – it was integrating them more efficiently than anyone else. In a world of modular, interchangeable parts, Dell's horizontal integration was magnificently effective.
Steve Jobs, of course, took Apple in the opposite direction. Rather than embracing modularity, he doubled down on vertical integration – controlling everything from silicon to software to retail. At the time, this seemed like a quixotic strategy, a relic of a bygone computing era.
History vindicated Jobs' approach spectacularly. Apple's control over its entire stack – from custom processors to the operating system to the user experience – created differentiation that horizontal PC makers couldn't match. The iPhone, built on this vertically integrated foundation, transformed Apple into the most valuable company in the world.
The debate between horizontal and vertical integration continues to shape technology markets. And at NVIDIA's GTC 2025 conference last week, Jensen Huang made it clear which side of this divide he's chosen.
The conventional wisdom regarding AI acceleration has been coalescing around a particular narrative: NVIDIA may dominate training with its GPUs, but inference – running the trained models – will be vulnerable to competition from specialized, lower-cost chips like ASICs (Application-Specific Integrated Circuits).
The logic seems sound. Once a model is trained, why pay a premium for NVIDIA's expensive, general-purpose GPUs when a cheaper, specialized chip can run the same model more cost-effectively? Major cloud providers like Google, Amazon, and Microsoft have all developed their own AI chips, and venture capital has poured billions into AI chip startups.
Jensen Huang, however, used his GTC keynote to systematically dismantle this narrative. His counterargument rests on a fundamental insight: in the AI era, power – not transistors, not chip count, but actual electricity – is the ultimate constraint.
This framing brilliantly shifts the competitive landscape. It's no longer about the cost per chip – it's about the cost per watt and how much intelligence you can extract from each watt of power.
But NVIDIA's strategy goes far beyond silicon. The company unveiled NVIDIA Dynamo, which Huang described as "the operating system of an AI factory." This software layer optimizes how AI models run across potentially hundreds or thousands of GPUs, dynamically allocating resources based on workload needs.
This is particularly critical for the latest advancement in AI: reasoning models. Unlike previous generations that generate responses in a single pass, reasoning models break problems down step by step. They might go through phases of "thinking" (prefill) and "answering" (decode) that have radically different computational patterns.
Huang explained: "I can dis-aggregate the prefill from the decode and I could decide I want to use more GPUs for prefill, less for decode, because I'm thinking a lot. It's agentic, I'm reading a lot of information, I'm doing deep research."
This comprehensive approach creates what economists call "complements" – products that become more valuable when used together. Each layer of the stack makes the others more valuable, creating a flywheel effect that's difficult for competitors to match.
Perhaps the most compelling part of Huang's presentation centered around what he called the "Pareto Frontier" – the fundamental tradeoff between system throughput and individual response time in AI inference.
The Y-axis represents tokens-per-second for the entire system (throughput), while the X-axis represents tokens-per-second for individual users (responsiveness). These goals are in direct opposition – maximizing one typically comes at the expense of the other.
What makes this particularly relevant to the ASIC debate is that specialized chips must pick a fixed point on this curve. They might optimize for throughput (good for batch processing) or responsiveness (good for interactive applications), but they can't easily do both.
NVIDIA's GPUs, however, can operate at any point on this curve. And with Dynamo software, they can dynamically shift resources to push the entire curve outward – achieving better performance at all points.
"The Pareto Frontier," Huang explained while displaying a rainbow-colored graph of performance metrics. "And each one of them, because of the color shows you it's a different configuration, which is the reason why this image says very, very clearly, you want a programmable architecture that is as homogeneously fungible, as fungible as possible, because the workload changes so dramatically across the entire frontier."
This is a devastating argument against specialized AI chips. The very specialization that makes them efficient also makes them inflexible – and in a rapidly evolving field like AI, inflexibility is a serious liability.
What we're witnessing is a classic pattern in technology markets: the pendulum swing between integration and modularization.
In the early days of a technology, vertical integration typically dominates because the components aren't good enough on their own. As the technology matures and standards emerge, the industry often modularizes, with specialized firms focusing on individual components that can be assembled by integrators.
The PC industry followed this pattern, evolving from vertically integrated companies like IBM to a modular ecosystem with Intel, Microsoft, Dell, and countless component manufacturers all focusing on their specific layers.
The conventional wisdom has been that AI would follow a similar trajectory – from NVIDIA's vertically integrated approach to a more modular ecosystem with specialized chips for different workloads.
But NVIDIA is making a compelling case that AI computation might be different. The combination of rapidly evolving models (like the shift to reasoning), the complex interplay between different computational phases, and the fundamental power constraints of data centers may favor vertical integration for much longer than expected.
The most significant implication, though, may be for the development of AI itself. By driving down the cost of inference so dramatically – 87% with Blackwell, potentially 99.97% with Rubin – NVIDIA is enabling entirely new categories of AI applications.
Reasoning models that generate thousands of tokens while solving problems step by step would be prohibitively expensive at Hopper's efficiency levels. But at Blackwell or Rubin efficiency, they become economically viable for a wide range of applications.
When Michael Dell dismissed Apple's integrated approach in 1997, he wasn't wrong about the PC industry as it existed then. Horizontal integration and modularity were indeed the winning strategies for that era.
But Steve Jobs understood that vertical integration creates opportunities for innovation that modular approaches cannot match. By controlling the entire stack, Apple could create experiences that were impossible when relying on components designed for the broadest possible market.
Jensen Huang appears to be following a similar playbook. By controlling everything from silicon to systems to software, NVIDIA can optimize the entire AI stack in ways that would be impossible in a more modular ecosystem.
If the the GTC is any indication, the "Chief Revenue Destroyer" has no intention of ceding ground to more specialized competitors. By advancing on multiple fronts simultaneously – from silicon to software – NVIDIA is creating an integrated AI platform that may prove as enduring as Apple's approach to consumer devices.
The lesson, perhaps, is that in technology markets, integration versus modularization isn't a one-way evolution – it's a pendulum that swings based on the unique characteristics of each market and era. And right now, that pendulum is swinging decisively toward NVIDIA's vertically integrated vision of AI.
Disclaimer: The views in the post are for for informational purposes only and should not be considered as investment advice. Please contact your RM or Kristal.AI for investment advise.
By
Kristal Investment Desk
April 10, 2025
Liked it?
Share it with your friends & colleagues!
April 11, 2025
By
Kristal Investment Desk
A fully digital onboarding process that can be completed within 15 minutes.
No more voluminous paperwork and queuing!
.svg)
We are licensed in Singapore, Hong Kong and India. Kristal Advisors (SG) Pte. Ltd. presently operates under the CMS license by the Monetary Authority of Singapore (MAS). Kristal Advisors (HK) Ltd is licensed and regulated by the Securities and Futures Commission (SFC) to carry out Type 4 and Type 9 regulated activities and is not involved in the discretionary management of any collective investment scheme. Kristal Advisors Private Ltd. presently operates as a Registered Investment Advisor under the jurisdiction of the Securities and Exchange Board of India (SEBI).
Read more about our privacy policy - Here
Copyright © 2023 Kristal Advisors (SG) Pte. Ltd. UEN: 201711235E
.png)
.png)
We are licensed in Singapore, Hong Kong and India. Kristal Advisors (SG) Pte. Ltd. presently operates under the CMS license by the Monetary Authority of Singapore (MAS). Kristal Advisors (HK) Ltd is licensed and regulated by the Securities and Futures Commission (SFC) to carry out Type 4 and Type 9 regulated activities and is not involved in the discretionary management of any collective investment scheme. Kristal Advisors Private Ltd. presently operates as a Registered Investment Advisor under the jurisdiction of the Securities and Exchange Board of India (SEBI).
Read more about our privacy policy - Here
Copyright © 2023 Kristal Advisors (SG) Pte. Ltd. UEN: 201711235E
WE ARE LICENSED IN SINGAPORE, HONG KONG, ABU DHABI, AND INDIA. KRISTAL ADVISORS (SG) PTE. LTD. PRESENTLY OPERATES UNDER THE CMS LICENSE BY THE MONETARY AUTHORITY OF SINGAPORE (MAS). KRISTAL ADVISORS (HK) LTD IS LICENSED AND REGULATED BY THE SECURITIES AND FUTURES COMMISSION (SFC) TO CARRY OUT TYPE 4 AND TYPE 9 REGULATED ACTIVITIES AND IS NOT INVOLVED IN THE DISCRETIONARY MANAGEMENT OF ANY COLLECTIVE INVESTMENT SCHEME. KRISTAL ADVISORS PRIVATE LTD. PRESENTLY OPERATES AS A REGISTERED INVESTMENT ADVISOR UNDER THE JURISDICTION OF THE SECURITIES AND EXCHANGE BOARD OF INDIA (SEBI). KRISTAL.AI MIDDLE EAST LIMITED IS LICENSED AND REGULATED BY THE FINANCIAL SERVICES REGULATORY AUTHORITY (FSRA) FOR CAT 3 A LICENSE.
Read more about our privacy policy - Here
Copyright © 2023 Kristal Advisors (SG) Pte. Ltd. UEN: 201711235E
USD 10,000
Minimum investment amount
ETF
Available Investment Products
Financial guide
Use while investing on Kristal
We encourage our India investors to use a financial guide. Kristal does not charge any additional fees for investing through them.
In case you already have a guide, we will try to bring them onboard. In case not, we can recommend one of our qualified partners to advise you through the journey.
I understand the financial products and would want to proceed with investing without a financial guide
ProceedIt is a structured product issued at par. In working it is similar to a Reverse Convertible but includes the barrier feature to protect downside to some extent. The underlying can be a basket of shares where the worst performing share may be delivered on expiry at the strike price.
Learn more about BRCs
It’s a structured product which is similar to ELONs, except the underlying can be a basket of stocks. This means that in addition to normal ELON factors, there are additional knock-out, knock-in rules associated with them.
Learn more about ELNs
An equity linked note which is issued
at par. The payment is made on initiation, by the client, in the form of shares that he/she might already hold and wants to unlock more returns especially during the periods when the stock is not giving any dividends.
Learn more about RELNs
.png){kind=icon}
