The AI Trilemma: Geopolitics, Bubble Economics, and Weaponized Open Source

Three forces are colliding in the AI race, and none of them are about building better intelligence. First, geopolitics has turned AI into a digital arms race. Second, financial engineering has created a circular economy where the same trillion dollars flows between the same companies. Third, open source has become a weapon—undercutting prices, diverting revenue, and challenging the very business models funding the race.

This isn’t a story about AGI timelines. It’s a story about what happens when a technology becomes too strategically important to fail, too financially engineered to scrutinize, and too globally contested to control.

Part 1: The Geopolitical Chessboard – Open Source as a Weapon

The US‑China AI competition follows a familiar cold‑war script: export controls, chip bans, and the mantra that “whoever wins AGI first wins.” But China has added a new move: weaponized open source.

DeepSeek V3.2 exemplifies this strategy. By releasing a powerful, enterprise‑ready model at a fraction of the cost of US frontier models, China isn’t just competing on performance—it’s competing on price. The goal is straightforward: divert enterprise revenue away from US AI companies, starving them of the capital needed to fund the next round of R&D.

[!idea] Visual suggestion: Open‑Source Price‑Pressure Diagram A chart comparing cost per inference of DeepSeek V3.2 versus GPT‑4o, Claude 3.5, etc., with annotations highlighting the enterprise‑market impact. Use cyan for analytical cost curves, amber for revenue‑flow arrows, and violet for geopolitical boundaries.

This isn’t charity. It’s economic warfare disguised as open‑source generosity. While US companies burn billions chasing the next parameter count, Chinese models offer a “good enough” alternative that erodes their economic moat. The result is a paradoxical landscape: the country with export‑restricted hardware is exporting unrestricted software, turning open source into a strategic lever.

Meanwhile, the US has made AI a single‑bet economy. In H1 2024, AI spending accounted for all of US GDP growth. Without it, the country would be in recession. Tariffs exempt AI; export controls protect it. The entire economic engine has become one enormous wager on a technology whose profitability remains unproven.

Part 2: The Circular Economy – 2008 with GPUs

If geopolitics provides the narrative, financial engineering provides the fuel. Welcome to the circular AI economy, where money flows in loops between the same handful of companies:

• NVIDIA → OpenAI: NVIDIA invests up to $100B in OpenAI; OpenAI buys NVIDIA chips.
• OpenAI → Oracle: $300B cloud deal gives Oracle instant AI credibility.
• NVIDIA → CoreWeave: NVIDIA buys $6.3B in cloud services from a company it already owns 7% of.
• OpenAI → AMD: OpenAI agrees to deploy billions in AMD chips; AMD gives OpenAI 10% of its company.

On paper, everyone is growing. In reality, they are passing the same money between the same players. The system looks—and trades—like a bubble, yet it enables more entrepreneurship than ever before.

The parallel to 2008 is unnerving. Then, multiple financial instruments bet on the same mortgages. Today, multiple instruments bet on the same GPU demand. Special Purpose Vehicles (SPVs) hide an estimated $24B in AI market debt—49% of it completely off balance sheets. Companies like xAI create shell entities that raise $20B, buy NVIDIA GPUs, then lease them back to the parent company. The debt disappears; the risk remains.

[!idea] Visual suggestion: Circular Money‑Flow Map A network diagram centered on NVIDIA, with arrows showing circular investments between OpenAI, Oracle, CoreWeave, AMD, etc. Use violet for partnership lines, cyan for cash flows, and amber for equity stakes. Dark background with subtle concentric rings.

Geographic arbitrage completes the loop. The US three‑tier export system (unlimited, limited, restricted) creates artificial scarcity. Companies like Nebius—headquartered in Amsterdam—buy unrestricted NVIDIA chips and rent them to restricted markets. NVIDIA profits twice: from chip sales and from equity in the gateway companies. Export controls that appear to hurt NVIDIA in fact increase demand by driving premium pricing.

Part 3: The Technical Reality – March of Nines Meets the Bubble

Beneath the geopolitical maneuvering and financial engineering lies a stubborn technical reality: AI is not ready for the burden being placed on it.

The March of Nines problem illustrates why. Moving from 90% reliability (demos work) to 99% (products work) requires exponentially more effort. Reality has infinite surface area of edge cases. This is why 95% of enterprise AI pilots fail to reach production—and why self‑driving cars, first demoed in 1986, are still not economically viable at scale.

Meanwhile, scale alone is not enough. Bigger models do not equal better intelligence. Architectural breakthroughs are needed, not just more parameters. As Yann LeCun puts it: “We’re never going to get to AGI by just training from text.” Yet the industry keeps pouring money into turbocharging a car when what it needs is an airplane.

The financial foundation is therefore built on shaky ground:

• Enterprise AI ROI averages only 6%.
• 80% of companies report no significant bottom‑line impact from AI.
• GPU rental rates have already dropped 75% in some markets.

The bubble is fundamentally based on AGI expectations. AI companies need $2 trillion/year in revenue by 2030 to justify current investments—more than Amazon, Apple, Alphabet, Microsoft, Meta, and NVIDIA combined. Current OpenAI revenue: $13B/year. The math doesn’t work.

Where Does This Leave Us?

We are caught in a trilemma:

1. Geopolitical necessity demands we race for AI supremacy.
2. Financial engineering creates the appearance of growth while hiding systemic risk.
3. Technical reality says the technology isn’t mature enough to deliver the promised returns.

The most likely outcome isn’t AGI—it’s a correction. When the bubble pops, regular people will lose (62% of Americans own stocks), while tech CEOs and the companies at the center of the circular flows will be fine. The geopolitical competition will continue, but with fewer resources and more scrutiny.

Navigating this landscape requires separating signal from noise. The signal is in unit economics, predictable use cases, and infrastructure control. The noise is in grand AGI narratives, circular investment announcements, and geopolitical posturing.

The ultimate question isn’t whether AI will transform the world—it already is. The question is whether we’ve built a system capable of surviving its own success.

References & Further Reading

• DeepSeek V3.2 and the US‑China AI Landscape – Analysis of China’s enterprise‑focused open‑source strategy. Watch on YouTube
• Inside AI's Circular Economy – Deep dive into financial engineering, SPVs, and geographic arbitrage. Watch on YouTube
• AGI Narrative and Economics – How the AGI story justifies massive losses and market concentration. Watch on YouTube
• The March of Nines – Why demos are easy but products are hard. [Slide deck reference]
• Scale Not Enough – Critique of scaling‑law‑only approaches to AGI. [Slide deck reference]
• I Know We’re in an AI Bubble Because Nobody Wants Me – Pete Warden on efficiency disconnect and signaling benefits. Read on petewarden.com

For internal reference, these topics are also covered in the Obsidian vault under 02 - Personal/Research/AI/ and the presentation “A most likely outdated perspective on AGI”.