Is it feasible for the Chairman of Hynix to implement the memory as a service model? How many listed companies in the US stock market have similar models? Last night, SK Hynix ADR went public on the US stock market. During an interview, Choi Tae won mentioned that Hynix is researching the "memory-as-a-service" (MaaS) model. The core of this model is that customers may no longer purchase memory chips at once in the future, but instead rent memory resources from SK Hynix based on usage, capacity, or lifespan. This model aims to cope with the tight memory supply, high procurement costs, and the reality that customers want to avoid huge initial hardware investment. In fact, there are already mature service-oriented cases in the storage industry. Everpure (formerly Pure Storage, stock code: P) and NetApp (stock code: NTAP), two publicly traded companies, have already packaged their storage systems into a subscription service model. 1. Everpure and NetApp's Storage Services Business Models These two business models belong to the enterprise level Storage as a Service (STaaS), which has been running for many years and widely used in global enterprise data centers and hybrid cloud environments. Their core is to integrate storage hardware, software, and management services into subscription products, allowing customers to use storage resources as operational expenses (OpEx) rather than one-time capital expenditures (CapEx) to purchase equipment. 1）Everpure Launch true Storage as a Service (STaaS) subscription models such as Evergreen//One and Evergreen/Flex. Customers pay based on usage, similar to the flexibility of public clouds, while retaining on prem control. This also includes data services, hybrid cloud support, and continuous innovation subscriptions (non-destructive upgrades). 2）NetApp Provide pay as you go storage as a service (STaaS) through NetApp Keystone, supporting a unified platform for block/file/object. There are native services such as Azure NetApp Files and Amazon FSx for NetApp ONTAP in the cloud. It also includes professional services, hosting services, and subscription models. The commonality between these two companies' models lies in transforming complex storage infrastructure into predictable subscription services, where customers do not need to manage underlying hardware details, can scale on demand, and enjoy maintenance and upgrade support provided by vendors. They have demonstrated commercial feasibility in the corporate environment and formed a stable source of recurring revenue. 2. The MaaS proposed by Hynix shares similarities in philosophy with Everpure and NetApp's STaaS, Both are service-oriented transformations from "one-time hardware sales" to "subscription/pay as you go", but there are essential differences: 1) Different levels: Everpure/NetApp sells "storage systems/solutions" (hardware array+software+management), which are upper level applications of memory; Hynix MaaS targets the 'underlying memory chip/capacity' itself. Closer to the concept of "silicon as a service". 2) Implementation difficulty: STaaS is mature (hardware controllable, software defined storage mature); MaaS is extremely challenging for chip manufacturers (memory needs to be tightly coupled with computing, especially HBM integrated on GPUs, with complex splitting/pooling). 3) Technical barriers: Storage services rely on vendors' own arrays and software; Memory services require memory pooling technology (emerging technologies such as CXL), virtualization/software defined memory, and large-scale infrastructure support. 4) Customer target and value chain: The former directly serves enterprise IT/data centers; The latter may need to be provided indirectly through cloud giants (AWS, Azure, etc.) or data center operators, or in collaboration with ecosystems such as Nvidia. 3. Short term (1-3 years) is very difficult and almost unfeasible; There is a certain feasibility in the medium to long term (more than 5 years), but it will not completely replace the purchasing mode, but rather exist as a supplementary mode. As mentioned earlier, memory is not a standalone "storage device", but a tightly integrated component of servers/GPUs. HBM is directly soldered onto the GPU or packaged through advanced technology. To 'rent memory', we need to solve: how to allocate physical/virtual capacity? How to ensure performance/isolation? How to charge and SLA? At present, memory pooling technology is not yet mature and difficult to simplify like renting storage arrays. Moreover, existing customers typically purchase complete machines from server OEMs (Dell, HPE) or cloud service providers. Hiring memory directly by Hynix will break the existing chain and require deep cooperation with Nvidia and cloud giants (who may prefer to do similar services themselves). However, memory pooling (CXL technology), compute storage separation, and software defined infrastructure are all being further iterated, and in the future, "on-demand allocation of memory capacity" may also be achieved. Simply put, mature storage services have proven the commercial value of subscription models, but innovation in component level services (where memory is the component) requires more technological breakthroughs and ecological collaboration. This article is sponsored by @ bitget_zh, titled 'Bitget Buying US Stocks: Instant Entry, Smooth Trading'