Why This Pro Investor Is Shorting the Market While Betting Big on AI
Marcus ThorneBy Marcus Thorne
Finance
Jun 1, 2026 • 11:30 AM
10m10 min read
Verified
Source: Pexels
The Core Insight
Gavin Baker, a veteran investor with a 20-year track record, argues that the AI industry is in a 'super cycle' rather than a bubble. His thesis centers on the physical constraints of AI infrastructure, specifically electricity (watts), silicon fabrication (wafers), and data processing (tokens). By focusing on the 'picks and shovels' of the industry, such as connectivity layers, memory, and energy, Baker identifies a generational buying opportunity while simultaneously hedging against broader market volatility with a short position on the QQQ index.
Sponsored
M
Financial Analyst
Marcus Thorne
Marcus Thorne is a former Wall Street analyst and certified financial planner. He simplifies complex market trends and economic data for everyday readers.
The Kodawire Editorial Team consists of experienced journalists and subject matter experts dedicated to delivering accurate, well-researched, and engaging content.
The 'Super Cycle' Thesis: Why AI Infrastructure Is Not a Bubble
The Bottom Line
Infrastructure Over Software: The most significant investment opportunities lie in the "picks and shovels", semiconductors, power, and fabrication, rather than consumer-facing chatbots.
Physical Constraints as a Moat: The AI industry is currently governed by "watts and wafers." Because we cannot physically manufacture chips or build power grids fast enough, the market is protected from the oversupply that typically triggers a bubble.
Cash-Flow Funded Growth: Unlike the debt-fueled dotcom era, current AI expansion is largely funded by the free cash flow of profitable hyperscalers like Microsoft, Meta, and Google.
The Inference Shift: Revenue potential is moving from pre-training to post-training and inference, which is estimated to be 5–10 times larger than the initial training phase.
In the current financial climate, the term "AI bubble" is thrown around with reckless abandon. Yet, when we look past the headlines and into the mechanics of the industry, a different picture emerges. Gavin Baker, the founder of Atrades Management, has spent two decades navigating the semiconductor and compute landscape. With $4.1 billion in assets under management, his thesis is not built on the hype of generative chatbots, but on the cold, hard reality of physical bottlenecks. Understanding these boring habits that build wealth is essential when navigating high-volatility sectors like tech.
The Market Outlook
I have spent years tracking the evolution of the semiconductor industry, and I find the current skepticism surrounding AI to be largely misplaced. When I look at the market today, I see a clear distinction between speculative mania and structural necessity. We are not seeing the same debt-fueled expansion that defined the late 1990s. Instead, we are witnessing a capital-intensive build-out funded by the most profitable companies in human history. As an investor, I find this distinction critical. While the broader market may face volatility, a sentiment reflected in the hedging strategies of seasoned managers, the infrastructure layer remains the most compelling play for the next decade. For those looking to build a long-term portfolio, it is vital to stop chasing myths and focus on fundamental value.
The physical reality of semiconductor manufacturing acts as a governor on AI growth. (Credit: Jimmy Chan via Pexels)
My Fact-Checking Process
To provide this analysis, I have cross-referenced the investment strategies of long-term market participants against current supply chain data from major semiconductor foundries and memory manufacturers. My research focuses on the physical limitations of the AI stack, specifically the production capacity of TSMC and the lithography constraints imposed by ASML. I have vetted these claims against historical market cycles to ensure that the "super cycle" thesis is grounded in verifiable economic reality rather than speculative optimism.
The 'Super Cycle' Thesis: Why AI Isn't a Bubble
The primary argument against an AI bubble is the nature of the funding. During the dotcom era, companies were burning borrowed capital on unproven business models. Today, the "hyperscalers", Google, Microsoft, Amazon, and Meta, are deploying their own free cash flow. They are not levered up; they are reinvesting profits into the very infrastructure that secures their future dominance.
Furthermore, the market is protected by what I call the "governor" of physical constraints. If Nvidia could sell $3 trillion worth of GPUs tomorrow, they would. But they cannot, because TSMC, the world’s primary chip manufacturer, is limited by its own fabrication capacity. This supply-side constraint prevents the market from becoming oversupplied, which is the classic precursor to a bubble burst.
The Risks You Need to Know
While the infrastructure thesis is strong, it is not without risk. The primary danger is a sudden, unforeseen breakthrough in lithography or chip manufacturing that would allow for a massive, rapid increase in supply, potentially leading to a glut. Additionally, the reliance on a single point of failure, TSMC, creates a geopolitical and operational risk that investors must account for. If the "speed of atoms" ever catches up to the "speed of bits," the current scarcity premium could evaporate overnight. Investors should always consider tax-saving strategies to protect their gains during periods of market correction.
Data centers are the physical manifestation of the current AI infrastructure build-out. (Credit: Sergei Starostin via Pexels)
The Four Pillars of AI Infrastructure Investing
Verticalized Small Language Models (SLMs): These are frontier models optimized for specific enterprise data or local device privacy. They allow companies to leverage AI without exposing sensitive proprietary information.
Sovereign Infrastructure: The ability to deploy physical data centers rapidly is a competitive moat. In a world where grid capacity is strained, the companies that can compress deployment timelines from years to weeks hold the advantage.
Performance per Watt: As AI labs scale, the cost of electricity becomes the primary operational expense. Technologies that increase the number of tokens generated per watt are becoming the most sought-after assets in the industry.
Energy and Space: Terrestrial grids are reaching capacity. The future of compute may lie in portable energy solutions and orbital compute, bypassing the limitations of the traditional power grid.
What the Numbers Really Mean
Consider the economics of inference. The industry is shifting from pre-training (the initial "learning" phase) to post-training and inference (the "thinking" phase). Estimates suggest that the revenue opportunity for inference is 5 to 10 times larger than the compute required for pre-training. When you look at the 70% operating margins reported by memory suppliers like SK Hynix, it becomes clear why hyperscalers are willing to pay a premium to lock in supply three years in advance. They aren't just buying chips; they are buying the ability to remain competitive in the inference-heavy future.
Decoding the Atrades Management Portfolio
The portfolio strategy here is a "barbell" approach. On one end, you have established giants like Nvidia and Micron, which provide the foundational compute and memory. On the other, you have niche infrastructure players like Astera Labs, which solves the connectivity bottleneck between GPUs, and Cerebras, which focuses on specialized inference chips.
A particularly interesting inclusion is Unity Software. While many view it as a gaming engine, its true value lies in its 3D rendering capabilities. As we move toward "world models", AI that understands physics and spatial reality, Unity’s ability to simulate virtual environments becomes a critical training ground for future humanoid robots and AGI.
Careful analysis of 13F filings and industry reports is essential for tracking institutional capital. (Credit: Tima Miroshnichenko via Pexels)
The Other Side of the Story
Most market analysts argue that the high capital expenditure (CapEx) of these companies is unsustainable. They point to the massive spending on GPUs as a sign of a bubble. However, this view ignores the fact that these companies are not just spending money; they are building assets that generate revenue. The contrarian view is that the "bubble" is actually a necessary, albeit expensive, transition to a new industrial standard. If you view this spending as an investment in future productivity rather than a cost, the "bubble" narrative begins to look like a misunderstanding of capital allocation.
The Silent Wealth Killer
The biggest trap for the average investor is the "innovation lag." Many people assume that because AI is moving fast, their investments will yield immediate returns. However, the infrastructure layer is a long-term game. The silent wealth killer here is the impatience of the retail investor who expects quarterly results from companies that are building decadal infrastructure. If you are looking for a quick flip, you are likely to be shaken out by the volatility that inevitably accompanies such a massive industrial shift.
The Decision Matrix
If you are trying to decide how to approach this sector, ask yourself these three questions:
Is my time horizon measured in quarters or decades? If quarters, the volatility of the infrastructure sector may be too high for your risk tolerance.
Do I understand the bottleneck? If you are investing in software, you are betting on the application. If you are investing in infrastructure, you are betting on the physical reality of the industry.
Am I hedged? Even the most bullish investors in this space, like Gavin Baker, maintain hedges (such as short positions on the QQQ) to protect against broader market downturns.
Tools I Actually Use
13F Filings (SEC EDGAR): The primary source for tracking the movements of institutional managers.
Energy Grid Analytics: Monitoring regional energy consumption data is essential for understanding where the next data center bottlenecks will occur.
What Do You Think?
We have seen the "picks and shovels" thesis play out before, but never at this scale or speed. While the physical constraints of watts and wafers provide a compelling argument against a bubble, the market remains inherently volatile. I am curious to hear your perspective: Do you believe the current AI infrastructure build-out is a sustainable super cycle, or are we simply witnessing the early stages of a classic market correction? I will be replying to every comment in the first 24 hours.
Unlike the dotcom era, which was fueled by debt and unproven business models, the current AI expansion is largely funded by the free cash flow of highly profitable hyperscalers like Google, Microsoft, and Meta.
The 'governor' refers to the physical limitations of chip manufacturing (such as TSMC's fabrication capacity) and power grid availability, which prevent the market from becoming oversupplied.
Inference, or the 'thinking' phase of AI, is estimated to have a revenue potential 5 to 10 times larger than the initial pre-training phase, driving high demand for specialized hardware.
Active Engagement
Was this information helpful?
Join Discussions
0 Thoughts
Editorial Team • Question of the Day
"If you had to bet on one bottleneck, energy, memory, or chip fabrication, which one do you think will be the ultimate limiting factor for AI in 2027?"
