Exchange Ranking
Top Exchanges
24h Volume Ranking
Popularity Ranking
Exchange BTC Balance
Proof of Reserves
Decentralized Exchanges
Funding Rate
Funding Heatmap
Liquidation Data
Max Pain
Long/Short Ratio
Whale L/S Ratio
Binance/Okex/Huobi L/S
Bitfinex Margin L/S
ETF Tracker
Solana ETF
XRP ETF
Hong Kong ETF
Bitcoin Treasuries
Crypto Reversal
Ethereum Reserves
HyperLiquid Wallet Analysis
Hyperliquid Whale Watch
Large Transactions
On-chain Movement
Bitcoin ROI
Stablecoin Market Cap
Options Analysis
News
Articles
Economic Calendar
Features
Wallet
Contract Calculator
Security
Collections
Watchlist
Following
BTC 英伟达主导AI计算，但科技巨头正加速自研专用芯片
尽管英伟达在AI计算领域占据主导地位，但谷歌、Meta、亚马逊等超大规模云服务商正日益投入资源，构建专为自身AI工作负载设计的处理器。其核心差异在于“专业化”。
英伟达的GPU旨在处理广泛的加速计算任务；而定制化的AI加速器则通过剔除企业不需要的功能模块，将更多的硅片面积、内存带宽和电力资源集中于特定且狭窄的计算场景，从而实现效率最大化。
谷歌TPU架构：专业化的典范
谷歌的TPU（张量处理单元）架构清晰地展示了这一逻辑。其中，TPU 8t专注于大规模模型训练，而TPU 8i则针对推理任务进行了优化，相比谷歌上一代推理产品，其在单位成本下的性能表现更优。
英伟达的优势：灵活性与生态系统
英伟达的领先优势不仅体现在芯片的原始性能上，更在于其卓越的灵活性。其GPU能够胜任前沿模型的训练、推理运行、科学计算，甚至应对数据中心建设时尚未出现的新兴工作负载。
This flexibility is crucial because AI software iterates quickly. Cloud service providers with thousands of Nvidia systems can flexibly allocate GPU resources as needs change, rather than being locked into a single narrow workload. In addition, Nvidia also has a CUDA ecosystem, mature network products and a complete developer community. These software-level advantages make it easier to deploy GPUs in different models and customer groups.
This also explains why companies that are committed to developing their own chips still purchase Nvidia hardware. Even Google's infrastructure uses TPUs in conjunction with Nvidia GPUs, rather than viewing the two as complete replacements.
Effect of scale: Economic advantages of customized chips
When the same workload is run at a million or billion times, the economic ledger changes. Inference is the most typical example. If Google knew exactly how requests were handled in the Gemini model, it could optimize its hardware design around these calculations, avoiding paying extra fees for rarely used flexibility.
This approach not only reduces power consumption and the cost of a single query, but also improves hardware utilization. The following are key comparisons of different options:
Meta followed a similar strategy to develop MTIA chips, while Amazon developed the Trainium and Inferentia series of chips for cloud infrastructure. This is why both Broadcom and Nvidia can benefit from growing AI spending: Nvidia provides flexible accelerators, while Broadcom helps hyperscale cloud service providers design chips optimized for specific workloads.
Hyperscale cloud service providers still need support from external chip experts
Designing custom accelerators does not mean that all components need to be manufactured in-house. Cloud vendors usually cooperate with professional companies such as Broadcom and Marvell, involving physical design, networking, interfaces and production.
例如，美满电子与谷歌的合作关系已扩展至涵盖定制推理加速器、网络控制器以及与谷歌TPU生态系统相连的近内存计算等领域。这也进一步解释了为何AI基础设施投资的范围远超GPU制造商本身。
定制芯片难以完全取代英伟达
更可能出现的结果是“混合架构”的普及。一家公司可能使用英伟达GPU进行大型模型训练，利用自有加速器处理海量推理任务，并采用独立的定制芯片负责网络传输或内存移动。
定制芯片在可预测的工作负载中通过降低成本和提升效率来竞争；而英伟达则凭借其在几乎所有领域的适用性保持竞争力。随着AI基础设施支出的增长，这两种路径将同步扩张，而非一方彻底淘汰另一方。
