A reply to an investor’s query to our Letter To Investors

By: TC Thio

At our firm, we deeply appreciate the continuous engagement, sharp insights, and rigorous questions we receive from our investors. Our recent Letter to Investors sparked several excellent discussions regarding the structural realities reshaping the global technology and economic landscapes. We believe that addressing these complex dynamics openly provides valuable context for all of our readers, as separating transient market sentiment from structural reality is a full-time discipline.

The following inquiries strike at the very heart of these shifting macro forces, offering a look into how we underwrite risk and position our capital for the long term. Below, we outline our strategic framework for navigating this environment.

Question 1: The Two-Model Framework & Portfolio Intent

When looking at the global AI landscape, it appears increasingly split into a two-model framework. On one hand, the US ecosystem seems focused on concentrated rent capture at scarce bottlenecks, such as frontier models and advanced compute. On the other hand, China appears to prioritize state-coordinated productivity diffusion across traditional industries.
Given this dynamic, I am concerned about the Chinese state’s multi-objective approach, which constantly balances financial returns with strategic autonomy and technological diffusion. Could this structural priority lead to severe firm-level margin compression, similar to what we witnessed in the Tesla-BYD price war?
Following that logic, I would like to better understand the true intent behind your portfolio construction. Can you confirm if your China holdings—specifically Cambricon, Horizon AI, SMIC, and Alibaba, alongside Ubtech and Fudan in incubation—are intended as something more than just a hedge against US tech dominance? Are these positioned as targeted exposure to firms that you believe possess durable cost-scale advantages in application engineering?

We completely agree with your observation regarding the two-model framework, and your reframing of our portfolio perfectly aligns with our intent. We firmly believe that a dominant, alternative tech stack will emerge globally alongside the US stack, and our portfolio is structured to capture that reality.

The US ecosystem provides a superior landscape for foundational research—going from "0 to 1." Profits there are indeed derived from concentrating rent capture at scarce bottlenecks. However, building physical infrastructure in the US to support these advancements is increasingly fraught with delays, such as local zoning disputes (NIMBYism) affecting downstream packaging facilities like Amkor in Arizona.

China, conversely, excels at going from "1 to 10." Its sheer volume of STEM engineering talent, backed by the world's deepest manufacturing supply chain, provides an unparalleled bedrock for rapid prototyping. The Chinese state has made a clear choice: focus on foundational infrastructure and technological diffusion. State coordination provides governance oversight, ensuring that founders reinvest in innovation rather than draining cash through dividends. This is a critical factor—and one that heavily influenced Charlie Munger’s decision to invest in BYD years ago.

We absolutely agree there is a risk of a "race to the bottom" in pricing if AI products and services become commoditized—a dynamic we are already seeing play out in Western frontier models as well. Because of this margin compression risk, we have actively avoided recent AI IPOs trading at astronomical price-to-sales multiples of 500x. To express our view on the China AI application layer, we choose hyperscalers generating massive free cash flow (FCF), like Tencent and Alibaba. They possess substantially more proprietary transactional data than any other platform, possess full-stack capabilities, and have the cash flow necessary to fund their own aggressive investments without relying on external capital markets.

Furthermore, companies demonstrating substantial revenue and FCF growth will receive greater capital allocations from us over time. Note that Horizon AI, Ubtech, and Fudan currently remain in incubation.

Question 2: Open Models, Pricing Power, and Infrastructure Moats

Looking at the rapid capabilities improvement and drastically declining deployment costs within open-model ecosystems—particularly coming out of groups like DeepSeek and Qwen—I have reservations about the long-term durability of your pricing power thesis for frontier models.
Specifically, if open-source alternatives continue to commoditize intelligence, wouldn't the true economic moats shift toward enterprise governance, data security, and system integration? If so, it seems logical that the lion's share of economic surplus will accrue to infrastructure and cybersecurity providers—such as CrowdStrike, Datadog, and Palo Alto Networks—rather than the model labs themselves.
Given this potential shift in where value is captured, how does this impact your current tactical position in Zoom, which you’ve previously framed as a proxy for Anthropic?

Open models have distinct advantages, as do closed models. Because China faces structural constraints in computing power and memory, its AI community had to collaborate to optimize algorithms. Techniques like Mixture of Experts (MoE), test-time compute, and RAG were rapidly adopted to reduce reliance on massive training databases.

Due to structurally lower costs in power, data center construction, and engineering wages, the cost of producing an AI token in China is substantially lower than in the West. This stark cost differential brings a very real risk of price deflation.

Despite this, Fortune 500 and 1000 enterprises are highly unlikely to adopt a system that is even slightly less accurate than the absolute best available. Therefore, the model makers that can consistently demonstrate superior efficiency and accuracy will maintain pricing power and win the enterprise race. Input and output token costs for enterprises will remain the standard pricing protocol. Quality output drives direct business value - a 1-page input prompt generating a 16-page analysis of a bank earnings report in 30 minutes holds immense value for capital allocators. Consequently, the premier model labs will likely capture the highest margin revenue, reinforcing our belief that the divergence between input costs and output value is a structural fixture of the market.

Regarding infrastructure software: companies like CrowdStrike (CRWD), Datadog (DDOG), and Palo Alto Networks (PANW) are integral to future cyber defense, especially as autonomous AI agents are deployed across corporate systems. Strict governance and data-loss protection protocols are required to prevent an AI agent from falling for a phishing attack. Frontier model developers like Anthropic can generate attack sequences, but they lack the foundational historical databases that dedicated cybersecurity firms possess. Hyperscalers recognize this - Microsoft,

for instance, now holds a 20% market share in cybersecurity and boasts capabilities on par with CrowdStrike. Pure-play security providers will not completely lose out, but surplus will be shared.

As for Zoom: our position is strictly tactical, and we will likely exit prior to an Anthropic IPO. We originally took this position based on Zoom's superior free cash flow margins of over 30% and the likelihood that it would become an acquisition target.

Question 3: Navigating Price Wars and Defensible Share
Looking closely at your China portfolio, how are you navigating the structural trade-off between firm-level margin compression and system-level technological diffusion?
More specifically, what is the core thesis driving your preference for firms positioned in foundries, robotics, embedded AI, and industrial automation, over those operating in sectors that are more exposed to intense, state-encouraged price wars? How do you ensure your capital is protected in an environment where competition is so heavily subsidized?

We do not see price wars as state-coordinated; instead, intense competition is the natural state of a high-growth market. By providing the foundational platform and infrastructure, the state lowers the barrier to entry for startups, particularly in sectors prioritized in the national five-year plan. This ease of entry fuels aggressive, early-stage competition, creating a high-stakes environment where firms must innovate rapidly to survive.

Because of this, we avoid asset-light tech companies or easily scalable manufacturing firms in China - unless they are elevated to a state-level imperative (like Cambricon and possibly Zhipu), where the government actively curates collaboration among the "best of breed."

Instead, we favor what Josh Brown calls Heavy Asset Low Obsolescence (HALO) companies. These businesses - like SMIC and advanced robotics manufacturers - are incredibly difficult to replicate. They require massive capital, elite technology leadership, and complex supply chains. The semiconductor foundry business is the ultimate HALO industry. It is a punishingly difficult science to master. Recent US AI export controls served as a wake-up call for Beijing, triggering unprecedented cross-industry collaboration (such as Huawei partnering with SMIC) that has led to genuine technological breakthroughs.

We apply these exact macro considerations to the robotics and automation stack. China currently accounts for over 50% of the world's annual industrial robot deployments. Because true autonomous robotics relies heavily on edge computing (processing data on the device to save battery weight) and 6G networks, China’s existing ecosystem - right down to the miniature precision actuators Elon Musk is sourcing - makes it the inevitable factory for the world's robotic future.

Question 4. Drawdowns and Asymmetric Entry Points
Looking back at the Q1 drawdown—which appears to have been partly driven by underperformance in your Chinese hardware allocations like SMIC, Fudan Micro, and Ubtech—how do you practically manage a multi-year structural bet through these periods of volatility?
Specifically, what are the fundamental or macroeconomic signals that would prompt you to add to, rather than trim, these positions during a downturn? Furthermore, if global supply chain pressures persist, where exactly within this ecosystem do you see the most compelling asymmetric entry points?

The recent drawdown, while more acute for our Chinese technology names, was driven by market perceptions of geopolitical and supply-chain risks rather than by a significant shift in underlying fundamentals.

Upon closer examination, fears regarding supply chain disruptions through the Straits of Hormuz (specifically for helium, petrochemical solutions, and aluminum) are overstated due to China's aggressive self-sufficiency push. China is already a massive domestic producer of Wet Semiconductor Chemicals (such as high-density polyethylene, Isopropyl alcohol, and PGMEA).

While Helium presents some risk of supply constraints, domestic suppliers are actively increasing capacity, successfully reducing their dependence on US exports, and tilting the trade balance toward a healthier level.

Because the fundamentals remain largely intact, we viewed this sentiment-driven drawdown as an asymmetric opportunity to add to our highest-conviction positions rather than exit them. In contrast, the industrial ecosystems of Japan, Korea, and Taiwan lack comparable structural resilience regarding domestic self-sufficiency. Notwithstanding these vulnerabilities, the broader market remains curiously sanguine, largely discounting the multifaceted geopolitical frictions and supply chain intricacies inherent in the region.

Concluding Thoughts: Intellectual Chokepoints vs. Supply Chain Chokepoints

Ultimately, our strategy rests on distinguishing between intellectual chokepoints and supply chain chokepoints.

A supply chain chokepoint is tactical, uncertain, and vulnerable to substitution or massive capacity expansion. An intellectual chokepoint - held by a truly innovative company with visionary leadership and superior execution - is highly defensible and compounds over time.

We invest strategically only in innovative companies with intellectual chokepoints. While we hold tactical trading positions (like Zoom) for relatively short periods, our core book is built for structural durability. We also rigorously avoid technology companies with unusually high foreign ownership or founders actively attempting to move assets out of the country. By maintaining this discipline, we believe we are well-positioned to compound capital through the complexities of the global AI divide.

Disclaimer:

This article expresses the author's views as of the date indicated on the publication and such views are subject to change without notice. Gratus has no duty or obligation to update the information contained herein. Certain statements contained herein constitute forward-looking statements, which are subject to known and unknown risks, uncertainties, and assumptions. Actual events or results may differ materially from those reflected or contemplated in such forward-looking statements. Further, Gratus makes no representation, and past investment performance should not be assumed to indicate future results. Moreover, wherever there is a potential for profits, there is also the possibility of loss. This publication is made available for educational purposes only and should not be used for any other purpose. The information contained herein does not constitute and should not be construed as an offering of advisory services, a recommendation to buy or sell any securities or related financial instruments in any jurisdiction. Any reference to specific companies or securities is for illustrative purposes only and may not represent actual holdings of any Gratus’ portfolio. Certain information contained herein concerning economic trends and performances is based on or derived from information provided by independent third-party sources. Gratus Investment Management Private Limited (“Gratus”) believes that the sources from which such information has been obtained are reliable. However, it cannot guarantee the accuracy of such information and has not independently verified the accuracy or completeness of such information or the assumption on which such information is based. This publication, including the information contained herein, may not be copied, reproduced, republished or posted in whole or in part in any form without the prior written consent of Gratus. Accordingly, neither we nor any of our associates, directors, connected parties and/or employees accept any liability whatsoever for any loss, whether direct or indirect, that may arise from the use of information or opinions provided.