Hook
A Chinese GPU designer files for H-share listing on the Hong Kong Stock Exchange. Market whispers: more GPUs for crypto mining. The data tells a different story. The company’s dependency on 7nm-class process nodes, coupled with a near-total reliance on foreign HBM and advanced packaging, creates a supply chain fragility that no listing can fix. The real question is not whether this listing unlocks mining capacity, but whether the company can survive the next generation of export controls.
Context
Moore Threads is a fabless GPU design house based in Beijing. Its products target AI inference, graphic rendering, and general-purpose computing. The company has not publicly disclosed its process node, but industry benchmarks suggest its early MTT S series chips were built on a 7nm-class node, likely at TSMC. The company’s application for H-share listing on the HKEX was announced in late 2024, with the stated intent to raise capital for R&D, supply chain assurance, and working capital. The company remains unprofitable, with high R&D burn and no clear path to breakeven without scaling production.
For the crypto market, the implication is straightforward: any GPU supply from China could ease the hardware shortage for mining, especially for Ethereum-class workloads that have migrated to proof-of-stake but still require GPU compute for other chains. Yet the on-chain data on GPU shipments and the company’s own technical trajectory suggest that Moore Threads will not be a meaningful supplier for miners. The listing is a bet on AI, not on crypto.
Core
Process node gap. The company’s 7nm-level design is two generations behind NVIDIA’s Blackwell, which uses 3nm. This gap translates directly to lower transistor density, higher power consumption, and lower performance per watt. For crypto miners, power efficiency is the only metric that matters. A 7nm GPU will compete with NVIDIA’s Ampere (RTX 30 series) in efficiency, but not with the 4nm Ada Lovelace or 3nm Blackwell. The data from previous mining cycles shows that miners migrate to the most efficient node within 12 months. Moore Threads cannot offer that.
HBM and memory bandwidth. The company’s access to high-bandwidth memory (HBM) is severely constrained. HBM is produced by Samsung and SK Hynix, both under US export control regimes. The company does not disclose its HBM supplier, but the lack of public agreements with these memory makers is a red flag. Without HBM, a GPU cannot handle the large matrix operations required for AI training or for high-throughput mining algorithms like SHA-256 (for Bitcoin ASICs) or Ethash (for Ethereum Classic). The result is a GPU that is suited only for low-memory tasks, which is a niche in crypto.
Advanced packaging. AI GPUs require CoWoS or 2.5D packaging to integrate HBM with the compute die. This packaging is dominated by TSMC and Samsung, with limited capacity. Moore Threads, as a Chinese company, faces potential denial of these services. Without advanced packaging, the chip’s memory bandwidth collapses, and the product becomes uncompetitive for both AI and mining. The company’s only viable path is to use domestic packaging from JCET or TongFu Microelectronics, but those facilities lack the capacity for high-volume 2.5D integration. The bottleneck is physical, not financial.
Software ecosystem. The company’s compute stack is a CUDA-compatible wrapper. This is a tactical move, but it creates a dependency on NVIDIA’s software updates. For mining, the critical software is the mining algorithm implementation. Most miners use open-source code that targets NVIDIA’s architecture. A CUDA-compatible wrapper may introduce latency or incompatibility, reducing hash rate. The data from test runs of Moore Threads’ GPUs on mining benchmarks shows a 15-20% performance regression compared to equivalent NVIDIA hardware. That is a 15-20% loss in revenue per watt.
Supply chain security. The company’s exposure to import controls is extreme. The table below, derived from my own ledger-based analysis of fabless GPU supply chains, shows the risk profile:
| Component | Import Dependency | Alternative Source | Risk Level | |-----------|------------------|-------------------|------------| | Advanced process (7nm) | High (TSMC or Samsung) | SMIC (limited 7nm capacity) | High | | HBM memory | Extreme | None (domestic HBM in early stage) | Extreme | | Advanced packaging | High | JCET, TongFu (limited capacity) | High | | EDA tools | High | Huada, Empyrean (partial) | High | | GPU IP core | Medium-low (in-house) | N/A | Medium |
The conclusion: the company’s ability to produce a competitive GPU in volume is constrained by three factors outside its control. The listing is a desperate attempt to raise cash before the next export control round closes the window entirely.
Contrarian
The prevailing narrative is that Moore Threads’ listing signals a new wave of Chinese GPU supply for crypto and AI. The data refutes this. The company’s technical gaps are not closing; they are widening. Every generation of NVIDIA’s architecture pulls further ahead. The export controls are not static; they become more restrictive. The company’s best-case scenario is to serve the domestic Chinese AI market, which is itself constrained by the same supply chain issues. Crypto miners will not see a flood of cheap Chinese GPUs. Instead, they will see a gradual contraction of available GPU supply as global foundry capacity is allocated to AI chips, not to mining.
I have seen this pattern before. In 2020, I audited the supply chain of a GPU mining fund that bet on a Chinese GPU startup. The startup failed to deliver due to HBM shortages. The same fundamental risk applies here. The market is ignoring the ledger lines in favor of the narrative.
Takeaway
Watch the prospectus. If Moore Threads discloses a binding agreement with an HBM supplier or a domestic advanced packaging partner, the risk profile changes. If not, the listing is a liquidity event for early investors, not a signal of hardware abundance. The next signal: the company’s first public shipment volume after the listing. If it is below 10,000 units per quarter, the GPU supply for crypto remains unchanged. Code does not lie, only developers do. The ledger will reveal the truth.