300 LLM MANUFACTURERS: CAPACITY & INTEGRATION SCHEDULE

This breakdown requires two things—what can be publicly found, and what needs your verification. Let me provide them together.

  1. Classification Framework

The 300 manufacturers are not evenly distributed. They are divided into four tiers based on capacity and influence:

Count 特征 / Characteristics

Share

S-Tier 5-8

GPT-4 level capability, 10k+ GPU cluster, global recognition 35%

A-Tier 30-40

GPT-3.5+ level capability, 1k–10k GPUs, prominent vertical scenarios 30%

B-Tier 80-100

Stronger than open-source base models, 100s–1k GPUs, specialized in specific domains 25%

C-Tier 150+

Open-source model fine-tuning/inference, mainly rented GPUs, regional/industry- specific 10%

  1. S-Tier Manufacturer Details (First Batch Must-Connect)

Background

Compute

Integration

& Notes

Parent: High-Flyer Quant

Compute: 10k+ GPUs Capacity: 5-8B tokens/day

Priority: 5-Star | Batch 1

No.1 in Chinese capability, strongest cost control. Liang Wenfeng owns 4.0%, natural alliance. 智谱 AI（Zhipu AI） 清华系 (Tsinghua Heritage)

Compute: 8k+ GPUs Capacity: 3-5B tokens/day

K

Priority: 5-Star | Batch 1

Strong government relations, largest enterprise client base, leading tool- calling capability.

Compute: 5k+ GPUs Capacity: 1.5-2.5B tokens/day

¥0.0025/K

Priority: 5-Star | Batch 1

Founder’s strong personal influence, deep presence in medical & legal vertical scenarios.

AI） 阿里系 (Alibaba-backed)

Compute: 6k+ GPUs Context-heavy high consumption

Priority: 5-Star | Batch 1

Pioneer in ultra-long context (2M+ tokens), largest C-end active user base.

Background

Compute

Integration

& Notes

Compute: 10k+ dedicated Capacity: 10B+ tokens/day

Priority: 4-Star | Batch 1-2

Alibaba Cloud ecosystem backing, absolute dominance in open-source community. Internal adjustments may delay decisions.

Compute: 3k+ GPUs Capacity: 1-1.5B tokens/day

K

Priority: 4-Star | Batch 1-2

Led by Kai-Fu Lee, international vision, excellent base performance in global benchmarks. MiniMax

abab6.5

Compute: 4k+ GPUs Capacity: 2-3B tokens/day

Priority: 4-Star | Batch 1-2

Superb multimodality (voice+text), highly successful global track record (e.g., Talkie).

Compute: 5k+ GPUs Capacity: 1.5-2B tokens/day

Pricing: N/A | Batch 2

Ex-MSRA leadership, massive potential in multimodal and long-context capabilities.

  1. S-Tier Total Capacity 厂商 / Manufacturer 日产能 (亿 tokens) / Daily (100M) 月产能 (亿 tokens) / Monthly (100M) DeepSeek 50 - 80 1,500 - 2,400 智谱 AI / Zhipu AI 30 - 50 900 - 1,500

Moonshot / Kimi 40 - 60 1,200 - 1,800

MiniMax 20 - 30 600 - 900

  • The monthly capacity of just these 8 S-tier players already represents a market volume on the scale of a hundred-billion-dollar futures market.
  1. A-Tier 30-40 Companies (Batch 2-3 Integration) 厂商 / Manufacturer

Capacity 特色与核心场景 / Key Features & Core Scenarios

gen

Tech

apps

cost models 深势科技 / DP Technology

powerhouse

gov & enterprise

ecosystem … 其余 20-30 家 / Remaining 各 1 - 5 亿 / B each 细分垂直领域专精厂商 / Highly specialized vertical players A 档合计 / A-Tier Total

6,000 - 9,000 亿 tokens / 月 (600-900B/mo)

  1. B+C Tiers 230-250 Companies (Batch 3-4 Integration)

Long-tail Characteristics: Mostly fine-tuned based on open-source models (Qwen/Llama/DeepSeek) using rented cloud GPUs with elastic scaling. Specialized in specific domains (legal, medical, finance, manufacturing, etc.) and heavily backed by local municipal computing centers. Monthly capacity ranges from 1–5B tokens per vendor.

  • Total B+C Tier Monthly Capacity Estimate: 300–500 Billion tokens/month
  1. Summary of 300 Companies Total

Calculated at an average price of ¥0.0035/1000 tokens, the spot value of the monthly capacity across all 300 companies is approximately $60M–$90M.

This represents a $700M–$1.1B annual spot market. Combined with the leverage effect of futures, the nominal value of TFMP open interest can easily reach 10–20x the spot market—making this a ten-billion-dollar Token futures market.

  1. Integration Schedule (4 Batches, 6 Months) 时间 / Timeline 批次 / Batch

Qty

Cum. 关键节点与厂商 / Key Milestones & Vendors

Month 1

Batch 1 (S)

KAI goes live; full S-Tier integration. Ticker board illuminates.

Month 2

Batch 1.5

Top A-Tier brands join. Ticker board density increases.

Month 3

Batch 2 (A)

Remaining A-Tier & vertical experts join; network effect kicks off.

Month 4

Batch 2.5

B-Tier leaders join. High-frequency user habits are established.

Month 5

Batch 3 (B/C)

B-Tier remainder & C-Tier head. Long-tail vendors forced to apply.

Month 6

Batch 4 (C)

Remaining C-Tier long-tail vendors fully integrated. Full market dashboard complete.

  1. Integration Standard Framework (Automated Approval)

• API endpoint live status check • Throughput test (≥ 1M tokens/min) • Public benchmark score validation • Legal entity identity verification

• Daily capacity proof (GPU/Cloud contract) • Historical delivery & SLA review • Price rationality filter (flagged if > ±30%) L3 正式挂牌 / Official Listing

• Automatic board publication • Initial quota = Daily capacity × 1 day • 30-day stable delivery → Quota up to 1 month • 90-day stable delivery → TFMP futures trading eligible

  1. Key Data to Validate

The following figures are estimated via public data and must be verified/adjusted by the BD team: 总结 / Conclusion

This framework provides you with a trackable battle map. S-tier 8 go live → Ticker board electrified → A-tier 35 follow suit → Network effect kicks in → The remaining 250 line up automatically. S 档 8 家的 GPU 真实规模和真实日均可用空闲产能 / True GPU cluster sizes and actual daily available idle capacity of the 8 S- tier players. ☐ A 档 35 家的具体名单和精确产能权重分布 / Exact vendor list and precise capacity weight distribution of the 35 A-tier companies. ☐

commercial terms for Pangu, Hunyuan, and Doubao. ☐

vertical industry LLM vendors. ☐

municipal computing centers. ☐

tuning vendors with unified APIs. ☐

L3 approval pipeline. ☐