American A.I. Losing Market Share To China

Here is a detailed breakdown of the content from the video titled "American A.I. Losing Market Share To China" by Wall Street Millennial:

1. Overview of Market Shifts (Consumer vs. Enterprise)

Consumer Market Dominance:

Western platforms remain dominant for everyday end-users [00:39].

ChatGPT leads with over 1 billi

on monthly active users (MAUs) [00:46].

Google Gemini follows with over 600 million MAUs [00:53].

After a brief spike to ~200 million MAUs in January 2025, DeepSeek's consumer usage steadily declined [00:58].

Enterprise & Startup Shift:

Enterprises and startups are heavily adopting open-source Chinese models due to

their significantly lower cost per token [01:08].

According to data from Open Router, token usage skyrocketed from 2.4 trillion/week to 47 trillion/week, with the top 6 most processed models all being Chinese open-weight models [09:42].

2. Financial & Business Case Study: Z.AI (GLM Family)

IPO & Valuation:

Z.AI list

ed on the Hong Kong Stock Exchange in January 2026 at HKD 116 per share [01:43].

By early July, the stock surged over 10x to ~$1,800/share (~$100B USD market cap / 800B HKD) [01:58].

Financial Performance (2025):

Revenue: $107M USD (more than doubled year-over-year) [02:51].

Gross Profit: $44M USD [02:56].

Operating

Loss: >$500M USD (driven by $470M USD spent on R&D for model training) [02:59].

Revenue Breakdown:

Cloud Segment (26% of revenue): Subscriptions and API token sales; operates on a low gross margin (19%) as part of a deliberate "growth over profit" strategy [03:11].

On-Premise Deployment (74% of revenue): One-time s

ales to custom fine-tune models for local corporate enterprise servers, yielding higher gross margins (49%) [03:22].

3. Model Capabilities & Pricing Comparison (GLM 5.2 vs. Claude)

GLM 5.2 Benchmarks: Released in June 2026, performing strongly on long-horizon coding tasks [04:42].

Cost Comparison per Million Tokens:

+------------------+------------------+-------------------+

| Model | Input Tokens ($) | Output Tokens ($) |

+------------------+------------------+-------------------+

| Claude Opus 4.7 | $5.00 | $25.00 |

| GLM 5.2 | $1.40 | $4.40 |

+---------------

---+------------------+-------------------+

Real-World Test Case: A prompt to generate a 3D first-person shooter game resulted in GLM 5.2 producing a functional game with recoil and bullet hole mechanics, outperforming Claude Opus 4.8 at roughly 1/5th the cost [06:05].

4. Enterprise Adoption Examples & Privacy

Airbn

b Adoption: Airbnb selected Alibaba’s Qwen model over ChatGPT to power its property listing tools because Qwen is cheaper, fine-tunable on proprietary data, and runs securely on their terms [08:24].

Privacy & Security: Data security concerns are minimized when using open-weight Chinese models, as enterprises can hos

t them locally or through third-party cloud providers, preventing external data monitoring [10:00].

5. Model Distillation & Strategic Challenges

How Chinese Labs Scale Efficiently:

Total funding raised by Z.AI is under $2 billion, compared to >$100 billion raised by OpenAI and Anthropic [10:24].

Chinese labs rely on

distillation—using outputs from top models (ChatGPT, Claude) as training data to build competitive models at a fraction of the cost [10:49].

Elon Musk's xAI admitted to using similar distillation techniques with ChatGPT to train Grok [11:22].

The Commoditization Dilemma:

Purely AI-generated content cannot be copyri

ghted under current legal precedents, making distillation legal in practice [11:38].

Western labs face high capital expenditures without strong enterprise pricing power, as open-source models rapidly catch up and undercut token costs [11:56].

American A.I. Losing Market Share To China

Wall Street Millennial · 63K vi