📋 摘要
AI 基礎設施投資面臨嚴重的折舊問題，隨著新一代芯片技術快速迭代，現有硬體將在短期內失去競爭力並需要更換。加密貨幣計算市場有望成為解決方案，透過為非前沿計算需求提供穩定的需求端支撐，改善上游資本支出回收效率。 新興的流動性計算市場將建立在加密軌道上，硬體可隨時插拔並重新分配，計算資源將流向提供最高回報的協議。這些協議的核心競爭優勢在於開發能夠延長舊硬體使用壽命的軟體技術。 Bittensor 已展現此模式的實際應用，訓練、推理和 GPU 網路競爭同一計算資源池，礦工將計算力導向收益最高的協議。這代表 L1 工作量證明挖礦時代的演進，預測 PoW 機制將強勢回歸。隨著 AI 資本支出規模超越想像，延長芯片可用壽命的技術將成為關鍵解決方案。
📄 完整內容
Don’t miss what’s happening
Sami Kassab
@Old_Samster
An investment thesis for 2026:
Problem - The scale of AI capex being deployed into chips will create massive depreciation problems. As new generations come online, large portions of today’s hardware will no longer be competitive and will need to be replaced.
Thesis - Crypto compute markets expand and stabilize the demand side for non-frontier compute which will improve capex recovery for everyone upstream.
We're going to see a liquid compute market emerge on crypto rails where hardware has the ability to plug in, unplug, and reallocate at any moment. Compute will flow to the protocols offering the highest return.
The core competitive dimension for these protocols will be developing software that extends the useful life of older hardware better than anyone else.
What this looks like in practice is already starting to form on Bittensor - training , inference, and GPU networks compete for the a pool of GPUs, with miners rotating compute to whichever protocol is generating the highest returns.
I see this as the next evolution of the L1 PoW mining era, and predict PoW is going to make a huge come back
Will Squires
AI Capex is beyond comprehension. The problem when you realise half of this is depreciation on chips that will need to be replaced in 5 years as they become obsolete.
Technologies that extend the usable life of chips, like software that enables training loads over heterogenous
1:38 AM · Dec 29, 2025
Read 10 replies
Create account
By signing up, you agree to the Terms of Service and Privacy Policy, including Cookie Use.
Something went wrong. Try reloading.
Terms of Service
Privacy Policy
Cookie Policy
Accessibility
© 2025 X Corp.
Sami Kassab
@Old_Samster
An investment thesis for 2026:
Problem - The scale of AI capex being deployed into chips will create massive depreciation problems. As new generations come online, large portions of today’s hardware will no longer be competitive and will need to be replaced.
Thesis - Crypto compute markets expand and stabilize the demand side for non-frontier compute which will improve capex recovery for everyone upstream.
We're going to see a liquid compute market emerge on crypto rails where hardware has the ability to plug in, unplug, and reallocate at any moment. Compute will flow to the protocols offering the highest return.
The core competitive dimension for these protocols will be developing software that extends the useful life of older hardware better than anyone else.
What this looks like in practice is already starting to form on Bittensor - training , inference, and GPU networks compete for the a pool of GPUs, with miners rotating compute to whichever protocol is generating the highest returns.
I see this as the next evolution of the L1 PoW mining era, and predict PoW is going to make a huge come back
Will Squires
AI Capex is beyond comprehension. The problem when you realise half of this is depreciation on chips that will need to be replaced in 5 years as they become obsolete.
Technologies that extend the usable life of chips, like software that enables training loads over heterogenous
1:38 AM · Dec 29, 2025
Read 10 replies
Create account
By signing up, you agree to the Terms of Service and Privacy Policy, including Cookie Use.
Something went wrong. Try reloading.
Terms of Service
Privacy Policy
Cookie Policy
Accessibility
© 2025 X Corp.
文檔 ID：
568366d6-2124-417d-9cf4-2c7bb695614d
向量 ID：
doc_568366d6-2124-417d-9cf4-2c7bb695614d
建立時間：
2025-12-30 20:45:33.375308
更新時間：
2025-12-30 20:45:33.375308