4 min remaining
0%
AI & Machine Learning

"Temu-fikasi" AI: Kebenaran di Balik GLM-5 dan Kematian Pasar Menengah

Temukan bagaimana GLM-5 mengubah pasar AI, menciptakan perpecahan antara model premium dan komoditas, serta menandakan akhir bagi pemain pasar menengah.

4 min read
Progress tracked
4 menit baca·

TL;DR: The release of Zhipu AI’s GLM-5 is not a technological breakthrough; it is a manufacturing breakthrough. Just as Chinese EVs commoditized electric cars, Chinese AI models are commoditizing intelligence via "Knowledge Distillation." We are entering a brutal phase of "Involution" (Neijuan). The market is splitting: You are either the expensive Genius (OpenAI/Anthropic) or the dirt-cheap Commodity (GLM/DeepSeek). Everyone in the middle will die.

James here, CEO of Mercury Technology Solutions. Tokyo - February 16, 2026

A screenshot of GLM-5 benchmarks has been circulating on Threads this weekend. People are freaking out. They are saying China has achieved "Compute Sovereignty" using Huawei Ascend chips and is challenging Claude Opus performance for pennies on the dollar.

Is this a miracle? No. It is industrial warfare. To understand what is happening, you need to understand the concept of "Distillation" and why the AI market is about to look exactly like the EV market.

1. The Secret Sauce: Knowledge Distillation

How does GLM-5 offer Claude-level logic for $0.80 per million tokens? It’s not magic. It’s Distillation.

Think of OpenAI’s o1 or Anthropic’s Claude 4.5 as the "Genius Professor." They spent billions of dollars reading every book in the library to learn how to think. GLM-5 (and DeepSeek) are the "Smart Students." Instead of reading the library, the Student just asks the Professor millions of questions and memorizes the answers and the logic.

  • The Teacher (OpenAI): Pays Billions in R&D and "Trial and Error."
  • The Student (GLM): Pays minimal training costs by mimicking the Teacher's optimal path.

The Result: GLM-5 effectively uses a "MoE" (Mixture of Experts) architecture—744B total parameters, but only 40B active per token. It is lean, mean, and built on the back of someone else's R&D.

2. Innovation vs. "Biomimetic Optimization"

Is this stealing? OpenAI thinks so (they banned DeepSeek for TOS violations). But in the open-source world, Data Distillation is an open secret.

This proves that China’s advantage isn't in fundamental algorithm breakthroughs (those still happen in Silicon Valley). Their advantage is Extreme Engineering Optimization. They are paving the road that Silicon Valley cleared. They don't need Nvidia’s H100s to discover the path; they just need Huawei’s Ascend 910s to follow it efficiently.

3. The "Involution" (Neijuan): The Middle Class Dies

This is the "EV Moment" for AI. Just as BYD forced Tesla to slash prices, Chinese models are forcing a "Race to the Bottom." This creates a market phenomenon called Involution: Intense competition that destroys profit margins to capture market share.

The Impact:

  • The Leaders (OpenAI, Anthropic): Will survive by being the "Premium Brand" with the absolute smartest (but expensive) models.
  • The Commodities (GLM, DeepSeek): Will survive by being "Good Enough" and virtually free.
  • The Middle: The startups trying to build "Medium Models" with "Medium Prices"? Dead. They cannot compete on IQ with the Americans or on Price with the Chinese.

4. The Investment Shift: Goodbye OpenAI Monopoly

I used to be 100% bullish on the OpenAI ecosystem. I am no longer so certain. The "Moat" of pure intelligence is evaporating faster than we thought. The market is currently washing out "OpenAI Concept Stocks."

However, there is one company that is perfectly positioned. They are not betting on just one horse.

  • They have their own proprietary "Brain" (Gemini).
  • They are the largest investor in the "Alternative Brain" (Anthropic).
  • They own the "Android" OS that these agents will live on.

Google (Alphabet). While everyone fights over the model, Google owns the Platform, the Backup Model, and the Chips (TPU).

Conclusion: Pick Your Lane

The era of "AI as a Tech Miracle" is over. The era of "AI as a Utility Bill" has begun. You now have a choice: Do you pay a premium for the Original Brain (Silicon Valley)? Or do you build your business on the Distilled Commodity (Shenzhen)?

For mission-critical logic, I still pay the Americans. For everything else? The economics of $0.80/million are hard to ignore.

(Which company do you think is the real hedge against OpenAI? Drop a comment below.)

Mercury Technology Solutions: Accelerate Digitality.

Frequently Asked Questions

What is GLM-5 and how does it differ from other AI models?

GLM-5 is a Chinese AI model that utilizes a manufacturing breakthrough known as Knowledge Distillation to offer competitive performance at a significantly lower cost. Unlike traditional models like OpenAI's GPT, which rely heavily on extensive research and development, GLM-5 mimics the learning process of leading models, making it a 'Smart Student' that can deliver high-quality results for a fraction of the price.

What does 'Involution' mean in the context of the AI market?

Involution refers to a phase of intense competition that drives down profit margins and forces companies to either become premium brands or low-cost commodities. As seen in the electric vehicle market, this phenomenon is leading to the decline of mid-market players who struggle to compete on both intelligence and price.

How is the AI market evolving with the introduction of cheaper models like GLM-5?

The introduction of cost-effective models like GLM-5 is reshaping the AI landscape by creating a clear divide between premium models and low-cost alternatives. Companies must now choose to invest in highly intelligent models from established players like OpenAI or opt for cheaper, 'good enough' solutions from competitors, leading to a rapid transformation in business strategies.

What are the implications for startups in the AI space?

Startups aiming to develop mid-tier AI models are facing significant challenges as they cannot compete effectively against either the high-end models from companies like OpenAI or the low-cost models from Chinese firms. This market shift suggests a difficult path ahead for these startups, as they risk being eliminated due to their inability to carve out a sustainable niche.

Why might companies choose to invest in GLM-5 or similar models?

Companies may opt for GLM-5 due to its extremely low cost of $0.80 per million tokens, making it an attractive option for applications where high-level intelligence is not critical. This economic advantage allows businesses to leverage AI technology for a variety of use cases without the significant financial burden associated with premium models.