The Broken Monopoly: Why AI's Future is Open, Fierce, and Free

Not long ago, industry insiders treated it as an inevitable law of gravity: as the cost to train cutting-edge AI models skyrocketed into the hundreds of millions, the market would eventually consolidate. By 2026 or 2027, the logic went, only a handful of mega-corporations would survive to control the world’s AI infrastructure. Yet, as we navigate through this critical era of tech development, gravity seems to have broken.

Instead of shrinking into a closed monopoly, the AI ecosystem is exploding with new, heavily funded players who are doing the exact opposite of hoarding their technology—they are giving it away.

The driving force behind this paradox is a fundamental shift in how AI companies make money. Take Thinking Machines, a US-based startup that emerged in early 2025. They recently released Inkling, a massive multimodal model with nearly a trillion parameters. Giving away this technological marvel isn't charity; it's a brilliant customer acquisition engine. Their commercial fine-tuning service is already generating hundreds of millions in annual revenue. In today's landscape, the model is the bait, and the customized computation is the hook.

This open-model gold rush is fiercely global, with Chinese tech giants and startups pushing boundaries at an astonishing pace. Tencent recently open-sourced its Hy3 model—which successfully tackled a 50-year-old math problem—moving it to a highly permissive Apache 2.0 license. Meanwhile, Meituan’s massive 1.6-trillion-parameter LongCat-2.0 proved that frontier models can now be trained entirely on domestic Chinese accelerators, bypassing traditional hardware bottlenecks and proving the resilience of the local supply chain.

The competition is translating directly into wins for consumers. Just one day after OpenAI slashed the prices of their smallest models by 80%, Chinese AI powerhouse DeepSeek dropped an update to their highly efficient Flash model, keeping the price war raging.

However, "open" doesn't always mean "free for all." Companies are getting incredibly strategic with their licenses. Moonshot AI’s new Kimi K3, for instance, was released under a model that requires commercial partners to sign revenue-sharing agreements if they want to offer the model to their own customers. It’s a shift from selling packaged software to taxing the broader digital ecosystem.

We are entering a decisive era where the fundamental currency is the "token"—the basic unit of AI processing. The strategy across the globe is clear: seed the market with open models to drive an insatiable demand for computation. For the public, this is excellent news. It means the AI revolution won't be locked behind a few corporate walled gardens, but woven into the fabric of everyday digital life through a diverse, fiercely competitive, and increasingly accessible marketplace.

Key Points

• Contrary to predictions of market consolidation, the AI industry is experiencing a massive boom in open-source models.
• Companies are giving away trillion-parameter models for free to drive lucrative revenue in customized fine-tuning and computing services.
• Chinese AI labs are achieving major milestones, including training massive models entirely on domestic hardware.
• Fierce competition between global players is driving down costs for users while introducing new, strategic revenue-sharing licenses.

Why It Matters

The proliferation of open-weight models ensures that the foundational layer of the next digital revolution remains accessible, preventing a future where a few tech giants control all AI capabilities.

Sources:

• Latest open artifacts (#23): Laguna S2.1, Inkling, & Kimi K3 show the utility of open models on the Pareto frontier — Interconnects (Nathan Lambert)