Nvidia Corp.'s (NASDAQ:NVDA) AI chips may be powering today's artificial intelligence boom, butSeagate Technology Holdings PLC(NASDAQ:STX) thinks the next leap in performance won't come from more compute alone.
輝達公司（納斯達克代碼：NVDA）的人工智慧晶片可能正在驅動當今的人工智慧熱潮，但希捷科技控股有限公司（納斯達克代碼：STX）認為，下一階段的性能飛躍將不僅僅依賴於更多的算力。
Instead, the data storage company argues that smarter storage architecture can help AI systems get more work out of the same expensive GPUs—a shift that could lower infrastructure costs while boosting productivity.
這家數據記憶體公司認為，更智能的記憶體架構可以幫助人工智慧系統從同樣昂貴的GPU中獲得更高的工作效率——這一轉變有望在提升生產力的同時降低基礎設施成本。
AI Needs More Than Faster Chips
人工智慧需要的不只是更快的晶片
The argument stems from a recent white paper Seagate published withSK hynix Inc.(NASDAQ:SKHY), which examines how inference and agentic AI workloads are changing the way data is managed. Rather than repeatedly generating the same information, modern AI applications increasingly rely on retaining context that can be reused across interactions.
這一觀點源於希捷科技與SK hynix Inc.（納斯達克：SKHY）近期聯合發佈的一份白皮書，該檔案探討了推理和智能體人工智慧工作負載如何改變數據管理方式。與重複生成相同資訊不同，現代人工智慧應用越來越多地依賴於保留可跨互動複用的上下文。
"Our recent white paper with SK hynix illustrates the importance of tiered storage for inference and agentic AI workloads, which show a direct benefit to hard drive storage," CEODave Mosleysaid on the company's fiscal fourth-quarter earnings call.
公司首席執行長Dave Mosley在公司第四財季業績電話會議上表示：「我們最近與SK hynix Inc.聯合發佈的白皮書闡明瞭分層記憶體對推理和智能體人工智慧工作負載的重要性，這些工作負載直接受益於硬盤記憶體。」
At the center of that approach is key-value, or KV, cache, which stores previously generated context so AI models can retrieve it instead of recreating it each time. "Key-value, or KV cache, is used to retain and reuse that context efficiently," Mosley said.
該方法的核心是鍵值（KV）緩存，它記憶體先前生成的上下文，以便人工智慧模型可以檢索使用，而無需每次都重新生成。Mosley表示：「鍵值（KV）緩存用於高效地保留和重用該上下文。」
Read Also:QUICK SPARK: Bloom Energy CEO Invokes Henry Ford to Explain the AI Power Boom
延伸閱讀：QUICK SPARK：Bloom Energy首席執行長引用Henry Ford解釋人工智慧電力需求激增
How Storage Unlocks Nvidia GPUs
記憶體如何釋放輝達GPU的潛力
According to Seagate, that seemingly simple change has an outsized impact on AI economics. By moving context across memory, solid-state drives and hard drives instead of forcing GPUs to recompute it, AI infrastructure can make better use of its most expensive hardware.
希捷科技表示，這一看似簡單的改變對人工智慧的經濟性產生了巨大影響。通過在內存、固態硬盤和硬盤之間移動上下文，而不是強迫GPU重新計算，人工智慧基礎設施可以更高效地利用其最昂貴的硬體。
"This drives the need for increased hard drive storage and reduces GPU usage during the most compute-intensive phases of an agentic application. As a result, GPU resources are available for additional revenue-generating workloads," Mosley said.
Mosley表示：「這推動了對硬盤記憶體需求的增長，並減少了在智能體應用程序計算最密集階段的GPU使用量。因此，GPU資源可用於處理更多創收型工作負載。」
The message isn't that GPUs become less important. Rather, Seagate argues that storage is becoming a bigger contributor to AI performance as inference workloads expand and models retain more context over time. That makes storage architecture an increasingly important part of the AI stack alongside compute and memory.
這並不是說GPU變得不那麼重要了。相反，希捷科技認為，隨著推理工作負載的擴大以及模型隨著時間推移保留更多上下文，記憶體對AI性能的貢獻正變得越來越大。這使得記憶體架構與計算和內存一起，日益成為AI技術棧中愈發重要的一部分。
The Next Winner In AI Infrastructure
AI基礎設施領域的下一個贏家
The comments also reinforce Seagate's broader investment thesis that AI is creating structural demand for high-capacity storage. Management said cloud data centers now account for roughly 90% of the company's exabyte shipments, while customers continue extending long-term supply commitments into 2029 and beyond as AI infrastructure scales.
這些評論也進一步強化了希捷科技更廣泛的投資邏輯，即AI正在為高容量記憶體創造結構性需求。管理層表示，雲數據中心目前約佔公司艾字節出貨量的90%，而隨著AI基礎設施的擴展，客戶仍在將長期供應協議延長至2029年及以後。
For investors, the takeaway is that the next phase of the AI race may not be won solely by building bigger GPU clusters.
對投資者而言，關鍵在於AI競賽的下一階段可能不會僅僅依靠構建更大的GPU叢集來取勝。
As companies look to squeeze more value out of every Nvidia accelerator they buy, the biggest upgrade could come from the storage systems working quietly behind the scenes.
隨著企業希望從每一塊輝達加速器中搾取更多價值，最大的升級潛力可能來自在幕後默默運行的記憶體系統。(Benzinga)