SetupAI
繁
SourceStalecollected in 37m

B300 Trading Turns AI Compute Into Gambling

Read original on 虎嗅
#gpu-supply#compute-procurement#market-risk#data-center

B300 prices reportedly tripled in five months—here’s why AI compute procurement is becoming a high-risk bet.

30-Second TL;DR

What Changed

The reported B300 price climbed nearly threefold, from over RMB 4.9 million in March to more than RMB 13.5 million five months later.

Why It Matters

For AI infrastructure buyers, rapidly rising accelerator prices and unreliable counterparties can delay deployments, inflate total compute costs, and create substantial fraud and delivery risk. Enterprises may need stronger procurement controls and should avoid treating scarce GPU supply as a reason to bypass standard due diligence.

What To Do Next

Require an independent on-site B300 inspection and stress test, with escrow-based staged payments and a signed delivery contract, before committing any GPU purchase.

Who should care:Enterprise & Security Teams

Key Points

  • •The reported B300 price climbed nearly threefold, from over RMB 4.9 million in March to more than RMB 13.5 million five months later.
  • •Buyers often cannot inspect or stress-test equipment before payment, while sellers may demand deposits, bank guarantees, or proof of financial strength.
  • •Even signed contracts and agreed prices can be undermined by sellers requesting additional money after market prices rise.
  • •Intermediaries face a double bind: buyers demand trustworthy supply while sellers refuse to expose inventory or accept balanced contractual penalties.
  • •The breakdown of lock-up agreements and delivery commitments is increasing transaction risk for enterprises building AI compute capacity.

Deep Insight

AI-generated analysis for this event — not the original article.

Enhanced Key Takeaways

  • •The B300 accelerator is widely understood in the Chinese market as a derivative or localized version of NVIDIA's Blackwell architecture, specifically tailored to comply with US export controls while maintaining high-performance compute capabilities.
  • •The extreme price volatility is exacerbated by 'gray market' logistics, where chips are often routed through multiple jurisdictions (such as Southeast Asia or the Middle East) to bypass trade restrictions, adding layers of cost and risk.
  • •Major Chinese cloud service providers and AI startups are increasingly shifting toward domestic alternatives like Huawei's Ascend series to mitigate the supply chain instability and legal risks associated with the B300 black market.
  • •Financial institutions in China have begun tightening credit terms for AI infrastructure projects, citing the 'gambling' nature of hardware procurement as a significant risk to the solvency of AI-focused enterprises.
  • •The lack of official warranty support from the original manufacturer for these diverted units means that buyers are effectively purchasing 'as-is' hardware, leading to high operational costs when units fail during large-scale cluster training.

Competitor Analysis

Performance
B300 (Gray Market)
High (Blackwell-class)
Huawei Ascend 910C
High (Optimized for LLM)
NVIDIA H20 (Compliant)
Moderate (Export-limited)
Availability
B300 (Gray Market)
Extremely Low/Volatile
Huawei Ascend 910C
Moderate (Domestic)
NVIDIA H20 (Compliant)
High
Pricing
B300 (Gray Market)
Speculative (RMB 13.5M+)
Huawei Ascend 910C
Stable (Contractual)
NVIDIA H20 (Compliant)
Market Standard
Support
B300 (Gray Market)
None (Third-party)
Huawei Ascend 910C
Full (Official)
NVIDIA H20 (Compliant)
Limited (Regional)

Technical Deep Dive

  • Architecture: Based on the Blackwell GPU microarchitecture, utilizing a multi-die chiplet design to achieve high-bandwidth memory (HBM3e) integration.
  • Interconnect: Features high-speed NVLink connectivity, though often limited by the lack of official support infrastructure in gray market deployments.
  • Power Consumption: High TDP (Thermal Design Power) requirements necessitate specialized cooling infrastructure, which is often missing in ad-hoc data center setups.
  • Software Stack: Requires specific CUDA compatibility layers that are difficult to maintain without official NVIDIA enterprise support, leading to frequent software-hardware integration failures.

Future ImplicationsAI analysis grounded in cited sources

Chinese AI startups relying on gray market B300s will face a 30-40% higher TCO (Total Cost of Ownership) compared to domestic alternatives.
The combination of inflated procurement costs, lack of warranty, and high failure rates significantly outweighs the performance benefits of the hardware.
Regulatory scrutiny on AI hardware procurement in China will intensify by Q4 2026.
The 'gambling' nature of these transactions is attracting the attention of financial regulators concerned about systemic risk in the tech sector.

Timeline

2026-03
Initial market entry of B300 units in China with pricing around RMB 4.9 million.
2026-05
Supply chain constraints tighten, causing the first major price surge in the secondary market.
2026-07
Reports emerge of widespread contract defaults and 'bidding wars' for available B300 inventory.
2026-08
Market prices for B300 units surpass RMB 13.5 million, leading to the current state of market disorder.
Read original article on 虎嗅

Weekly AI Recap

Read this week's curated digest of top AI events →

PostLinkedIn

AI-curated news aggregator. All content rights belong to original publishers.
Original source: 虎嗅 ↗

This is a summary, not the original. Read the source, or get the weekly briefing.

The weekly digest

One email a week. Unsubscribe anytime.