Chips have started to be rented?

As we all know, the semiconductor industry is a capital-intensive industry, with high investment, long cycle and winner-take-all. And on average, 4-5 years will experience a round of semiconductor cycle, and every downward cycle, semiconductor companies in it are often tormented.

Recently, semiconductor giants seem to want to launch a new business model to alleviate the "dark years" of the downcycle.

Semiconductor giants open chip leasing model

Samsung has come up with a novel idea to consider launching a new business model, high-end memory chip rental services, which Samsung calls "memory as a service (MaaS)."

This source of inspiration should have been born under the long-term promotion of "XaaS (X as a Service)" by cloud computing vendors. XaaS means that the "service" itself is a lease, and the consumption model for resources is pay-as-you-go rather than fixed expenses. Simply put, users can purchase services from suppliers, such as IT infrastructure, platform or program resources, through subscriptions, and on-demand purchases can reduce costs for users.

Samsung's proposed "memory as a service (MaaS)" is similar, under which Samsung lends storage products for high-performance computing, such as next-generation memory expander DRAM modules and SSDs, to cloud service companies, including Google, and collects rent from them. Cloud service customers can acquire memory chips at pre-set contract prices, while receiving system management services throughout the life of the chip, reducing associated procurement costs.

In the current trend of storage recession, Samsung hopes that MaaS will lead to a more stable stream of sales and revenue, and according to Samsung's expectations, this new business will account for at least 10% of its entire DRAM sales revenue.

Samsung is not alone. Previously, Intel incorporated CPU in-purchase functionality (SDSI) into Linux 5.18, a software-defined silicon (SDSI) post-manufacturing mechanism that allows additional CPU features to be activated through the purchase of a license.

A month ago, Intel officially launched its On Demand program. This program is dedicated to its just-released fourth-generation Xeon Scalable processor, Sapphire Rapids, and allows customers to top up and unlock CPU features such as proprietary accelerators and security technologies that are not turned on by default.

In the On Demand plan, there are two payment models: "Consumption Mode" and "Activation Mode", "Consumption Mode" is billed on a pay-as-you-go basis, and "Activation Mode" requires the purchase of a license but buyouts perpetual activation.

There are also two types of unlockable CPU functions, one is Intel's own CPU accelerators and security features, which accelerate specific workloads and improve security. For example, Software Guard Extensions, Dynamic Load Balancer (DLB), Intel Data Flow Accelerator (DSA), Intel Memory Analytics Accelerator (IAA), Intel Memory Analysis Accelerator, and Intel QuickAssist Technology (QAT), these features are basically "activation mode", a buyout.

The other category is features and services provided by Intel's partners (H3C, HPE, Inspur, Lenovo, Supermicro, PhoenixNAP, and Variscale), including but not limited to server application optimization, intelligent computing, IaaS solutions, cloud native and edge computing, etc., which are available in both "consumption mode" and "activation mode".

In summary, Intel's approach is to upgrade the customer's machine without changing the actual hardware by giving the customer a CPU that has the function disabled but can activate the function when needed.

Why try the chip leasing model?

The important reason behind the semiconductor giant's attempt at chip leasing is actually the result of the slowdown of Moore's Law. As Moore's Law slows, scaling difficulties make it difficult for chip companies to shorten time-to-market for emerging hardware. As a result, many companies want to capture more value than just selling chips at once. The service model is a means for enterprises to make up for their shortcomings, and Intel, Samsung, NVIDIA, Qualcomm, etc. are trying to enter the service model.

So far, the software industry has achieved great success through the "software-as-a-service" model. As the internet boom accelerated, the popularity of this model was boosted, with many users paying more than a month/year while providing feature and security updates.

There are already 45 SaaS listed companies in the United States, with a total market capitalization of nearly $200 billion. The success of the US SaaS software Slack has become a textbook in the industry, with a valuation of $2.8 billion in just two years. Slack users are growing 5% weekly, with 1.1 million people online every day, of which 300,000 are subscribers. Slack's success has not only become an industry myth, but also countless up-and-comers.

The core idea behind XaaS and other cloud services is that users can reduce costs and access specific types of personal resources by purchasing services from vendors through subscriptions.

Before the advent of XaaS and cloud services, enterprises often had to purchase licensed software products and install them on-site. They have to buy hardware and connect it together to create an extended network. They must do all the security work on-site and must provide expensive server setup and other infrastructure for all business processes.

In contrast, with XaaS, users only buy what they need, paying as needed. This allows users to revolutionize the service model over time. With a multi-tenant approach, cloud services can provide a lot of flexibility. These services are supported by concepts such as resource pools and rapid elasticity, and business leaders can simply add or subtract services as needed.

Instead, the semiconductor industry has been adopting a product delivery model, where products are locked into the final product through a "build" and "release" model, without providing additional functionality at the hardware level. This makes chip companies lose more opportunities to provide value services.

Demand is created, it is precisely to continue to promote customer demand from another angle, chip companies began to try to take the service model. On the one hand, this type of model can continue customer service and develop another charging model; On the other hand, extended services can help customers reduce costs and get what they want more conveniently.

In fact, the semiconductor industry has also tried many methods when it comes to dealing with chip price fluctuations brought about by the cycle. For example, Micron announced a new pricing model called a "forward pricing agreement." Unlike existing volume-based agreements, this one takes into account both the size of the purchase and the price.

Micron Chief Commercial Officer Sumit Sadana said one of the top 10 customers has signed this new pricing agreement with us, which will generate more than $500 million in annual revenue over the three-year agreement.

What are the chances of success?

Renting out the chip is definitely a bold attempt. However, Samsung and Intel face different leasing methods and have different challenges.

In Intel's On Demand program, turning functionality on and off in local servers is a challenge because Intel may not have access to such installed servers.

Jim McGregor, principal analyst at Tirias Research, said: "Hyperscale operators and IT service providers are more likely to charge for it. This makes more sense. If you go to AWS or Azure, you pay for what you need, or you pay by the minute, or you pay for resources. It's a bit difficult to do this at the chip level. ”

And there are also customers who have expressed doubts: "The list of features enabled through software updates can be endless, from more cache memory to DRAM memory, to additional processing cores, to additional GPUs for gaming applications, to auxiliary cellular (possibly 6G) antennas." "No one can say for sure why such services cost extra.

Attempts at new business models are not always successful.

Historically, Intel also tried to launch a paid upgrade of desktop CPUs for consumers in 2010, allowing some low-end desktop CPUs to pay a fee to get an activation code to unlock additional features.

Initially launched for the Pentium Pentium G6951 desktop CPU (2.8 GHz), the processor enables an additional 1 MB of cache and hyperthreading (the upgraded CPU model is called the Pentium G6952) for a fee of $50, bringing its performance close to that of the higher-end Core i3-530 CPU.

Because at the market price at the time, the Core i3-530 CPU was only $15 more expensive than the standard Pentium G6951... There was no need for consumers to spend $50 to upgrade the G6951, so the program didn't have much response.

At present, the most likely to adopt the chip leasing model is cloud service vendors and HPC vendors. Such manufacturers often need to face scenarios with high computing intensity, high concurrency, and complex applications, and chip performance is extremely important to them, and high-performance chips often need to occupy a large amount of expenditure.

However, Amazon, which is also a cloud server manufacturer, does not seem to buy chip leasing. Amazon says that producing chips in-house will provide customers with more cost-effective computing power than renting processors made by Intel, Nvidia or AMD.

For the high-end DRAM leasing proposed by Samsung, there are also optimistic voices in the industry, believing that this model is cost-effective for Samsung and its potential customers. Samsung can withstand cold downcycles, and customers can reduce capital expenditure risks and have more flexibility to meet diversified DRAM needs.

Samsung's choice of this time is also in HPC and cloud such areas as storage requirements, logically speaking, for Amazon, Google and other manufacturers, the business model of cloud services involves a variety of "XaaS" leasing services, and they will be more accepting of high-end DRAM leasing models.

Overall, the chip leasing model has just entered the market, and it remains to be seen whether there are enterprises to accept it. However, this kind of approach does give customers who have been pressured by soaring prices of dedicated chips in the past some respite.

Because semiconductor design and production cycles are usually very long, if you start selling products with built-in functions like Intel, it will be a great test for enterprises, because it is necessary to predict the future trend in a few years to ensure that the additional functions are still needed by customers after the product is enabled in production.