Fidenza Macro's Geo Chen Turns Bearish on AI: A 4-Pillar Thesis for Why the Bull Market Has Peaked

In June 2026, Geo Chen of Fidenza Macro did something few macro analysts had the conviction to do: he liquidated his entire AI semiconductor and infrastructure portfolio. Since then, he has published a detailed bearish thesis arguing that the AI-driven equity bull market has peaked and that stagflationary conditions will bring more turbulence ahead. His framework rests on four interconnected pillars that together make a compelling, data-backed case for a regime change .

1. Korean Retail Leveraged Speculation: The "Low-Quality Capital" Blowup

Chen identifies Korean retail investors as the defining low-quality capital of this cycle's overshoot phase. They piled into 2x and 3x leveraged ETFs on SK Hynix, Samsung, and the KOSPI with what he describes as "reckless abandon," creating hundreds of billions in artificial buying power . The hedging mechanics of these products forced market makers to accumulate massive negative gamma exposure — buying on up days and dumping on down days — which amplified volatility and triggered cascading margin calls .

Citi estimates the leveraged product unwind has erased $38.7 billion, with approximately 1.2 million accounts liquidated — roughly 1 in 30 Korean adults . The Korean KOSPI has since plummeted nearly 30% from its June peak, forcing the market operator to suspend trading multiple times after losses exceeded 10% . Chen argues this impaired capital is unlikely to return, and that parabolic bull markets fueled by this kind of leverage almost always end with prolonged bear markets .

2. Chinese AI Model Competition: Collapsing the Closed-Source Moat

Chinese open-source models are directly undermining the business logic of frontier AI labs, Chen argues. The competitive landscape has shifted dramatically:

• Moonshot AI's Kimi K3 scored 1,679 on Code Arena, surpassing Claude Fable 5 (1,631) .
• Alibaba released Qwen 3.8, an open-weight model with 2.4 trillion parameters .
• Zai's GLM-5 and Sakana AI's Fugu multi-agent system demonstrate that smaller, nimbler competitors using novel efficiency methods — such as DeepSeek Sparse Attention and Recursive Self-Improvement — are closing the quality gap at a fraction of the cost .

Chen's core insight is stark: if 80–90% of tokens shift to cheaper or open-source models, the revenue pool at the top of the quality spectrum narrows sharply, breaking the capex flywheel that has supported AI infrastructure valuations . The business model that underpins the entire AI investment narrative — that frontier models will command premium pricing and justify massive infrastructure spend — is being disrupted from below.

3. Compute Supply Glut: From Bottleneck to Surplus

Perhaps the most structurally damaging pillar, Chen argues the narrative that compute is structurally supply-constrained is reversing. He flags several converging developments:

• Meta is building a cloud business to rent out its compute capacity, increasing overall supply .
• AI labs are rapidly lowering inference costs, and token spending indexes from Silicon Data have been declining since peaking in June, indicating hardware efficiency improvements are outpacing demand growth .
• OpenAI is delaying its IPO and struggling to raise funds, a development that will trickle down to the semiconductor supply chain .
• Google raised its 2026 AI capex forecast to $205 billion, but the stock fell 7% on the news — the market no longer rewards capex expansion .
• Data center credit spreads are widening, signaling rising credit risk in the compute infrastructure sector .

Chen makes a critical observation: the very act of investing to resolve bottlenecks destroys the bottleneck narrative that made these stocks attractive . The more hyperscalers build, the less scarce compute becomes, and the lower the returns on that capital.

4. Fed Policy Backdrop and Stagflation

Two macro forces compound the AI-specific risks:

• The Iran war has already cost the US $37.5 billion after the ceasefire collapsed in July 2026. Brent crude rallied from $70 to $85, and Chen warns of a prolonged "on again, off again" conflict that periodically spikes oil prices . Combined with four inflation shocks, this creates what he calls "the strongest and most persistent stagflationary pressure" in years .
• The Fed's credibility is eroding. The 30-year US Treasury yield broke above 5.2%, hitting a two-year high. Bearish steepening in the bond market reflects the market's view that the Fed, under Christopher Waller's leadership, is losing control of inflation. The FOMC's failure to hike rates has allowed inflation expectations to become unanchored .

A firming PCE print could bring the Fed, front-end yields, and the dollar directly into the blast radius for crowded AI trades, Chen warns . The AI trade, already wobbling from its own excesses, would face a macro environment that punishes its longest-duration, most speculative components.

Chen's Bottom Line: Cash, Gold, and Rotation

Chen explicitly compares the current moment to March 2000, March 2008, January 2020, and December 2022 — all inflection points before violent market turmoil . He is now sitting on cash, accumulating gold into weakness, and sees capital rotating out of AI into healthcare, financials, software, and biotech . His base case is a multi-month topping process, with further turbulence ahead as the unwind of speculative excess continues.

The thesis is not merely bearish — it is a structural argument that the very dynamics that drove AI stocks to parabolic highs are now actively reversing. For investors still holding AI-related assets, Chen's publication provides a sobering, source-backed framework for questioning whether the bull market has further to run or has already hit its peak.