Understanding why Singapore’s web hosts are migrating to ARM servers
Understanding why Singapore’s web hosts are migrating to ARM servers
The shift didn’t start with a press release or a product launch. It started with a power bill.
In mid-2023, a mid-sized hosting company operating out of a data centre in Tai Seng received its monthly invoice from the colocation provider and found the figure had climbed roughly 18 percent over the same period the previous year. The operator, who manages around 400 physical servers across two racks, spent the following week running thermal readings and power draw logs on every unit. What he found was not unusual for the industry: the Intel Xeon processors, several generations old now, were pulling between 180 and 220 watts per socket at idle, and the cooling load required to keep them stable added another 30 percent on top. The numbers were not catastrophic. They were simply no longer competitive.
That hosting company is one of perhaps two dozen in Singapore that have quietly begun migrating production workloads to ARM-based servers over the past eighteen months. The migration is not yet a flood. But it has reached a point where the operators who run the city’s web hosting ecosystem are paying attention in a way they were not two years ago.
What ARM servers actually change for a hosting operation
The technical difference is straightforward enough. ARM processors, originally designed for mobile devices and later adapted for data centre use by companies like Ampere Computing and AWS with its Graviton line, use a reduced instruction set architecture. Where traditional x86 processors from Intel and AMD handle a broader range of instructions natively, ARM relies on software to translate more complex operations. The trade-off has historically been compatibility: not everything runs on ARM, and the things that do sometimes run differently. The trade-off in return is efficiency. An Ampere Altra processor running at comparable throughput to an Intel Xeon Gold typically draws somewhere between 40 and 60 percent less power under load, depending on the workload.
For a hosting operation in Singapore, where commercial electricity rates hover around 30 cents per kilowatt-hour and data centre space rents by the kilowatt, that difference is not academic. It is the difference between a ten-centimetre-thick margin and a twelve-centimetre-thick one. Over a three-year server lifecycle, the power savings alone on a single 64-core ARM machine can cover roughly a third of the server’s purchase price.
The catch, and there is one, is that the savings only materialise if the software stack cooperates. A shared hosting environment running predominantly PHP applications, WordPress sites, and static file serving will migrate with minimal friction. A custom Node.js application compiled for x86 will not. The difference between a smooth transition and a painful one depends almost entirely on how much of the stack is open-source and how much is proprietary.
Which workloads move easily and which do not
The operators who have done this already tend to describe the migration in phases. The first phase is always the easiest: static file serving, object storage, caching layers, and any service that runs in a containerised environment with a well-maintained ARM Docker image. These workloads are essentially architecture-agnostic. The hosting provider simply rebuilds the container image for the new platform, deploys it to a test cluster, runs the existing test suite, and promotes to production if nothing breaks.
The second phase is more involved. PHP-based applications, including the vast majority of WordPress installations, generally run without issue on ARM, provided the hosting company rebuilds the PHP runtime with ARM-specific optimisations enabled. The PHP interpreter itself has been architecture-agnostic for years, but some popular PHP extensions — particularly image processing libraries like ImageMagick and certain IonCube loaders — have not always kept pace. A hosting provider in the Jurong East area discovered this the hard way when a customer’s e-commerce site, built on a custom PHP framework, began throwing segmentation faults after the migration. The cause turned out to be an obscure PHP extension for PDF generation that had not been compiled for ARM and was silently falling back to an x86 emulation layer. The emulation worked, but it was slow enough that the site’s page load times doubled. The fix required the provider to patch the extension’s source code themselves, a process that took two engineers the better part of a week.
The third phase, and the one most hosting companies are still avoiding, involves databases. MySQL and MariaDB run on ARM, and the performance benchmarks published by Ampere show the gap between ARM and x86 for database workloads narrowing with each generation. But database migration is a risk calculus that most operators are not yet ready to make for production customer data. The latency-sensitive nature of database queries means that even a small architectural mismatch can introduce unpredictable slowdowns. The general approach among Singapore’s hosting companies has been to move web and application servers to ARM first, leaving databases on x86 hardware, then reassess after a year of operational data.
The pricing reality of ARM server hardware
The economics of acquiring ARM hardware in Singapore bear some examination. A new Ampere Altra 80-core server from a distributor in the Kaki Bukit industrial area, configured with 256 gigabytes of RAM and dual NVMe storage, costs roughly $9,800 Singapore dollars as of early 2025. A comparable Intel Xeon Silver 4510 server with similar core count and memory lands at around $11,200. The upfront saving of about $1,400 per server is noticeable but not decisive. The decisive factor is the ongoing cost: the ARM server, under a typical hosting workload, consumes approximately 280 watts at load compared to 480 watts for the Intel machine. In a data centre charging $0.35 per kilowatt-hour for power, the ARM server saves roughly $600 per year in electricity alone. Over a five-year depreciation cycle, the total cost of ownership works out to about $2,200 less per server.
That calculation assumes the server runs at moderate utilisation. In practice, the ARM machines tend to run hotter in terms of CPU utilisation because operators feel comfortable pushing them harder, knowing the thermal headroom is greater. One operator running a cluster of ARM machines for shared hosting reported average CPU utilisation of 72 percent across the fleet, compared to 55 percent on the equivalent x86 machines. The ARM machines simply did not throttle under sustained load the way the Intel machines did.
Control panel compatibility and the software middle layer
For the companies that sell managed hosting, the control panel is the interface through which customers interact with the service. cPanel, still the dominant panel in Singapore’s shared hosting market, added official ARM support in version 110, released in late 2023. DirectAdmin followed shortly after. Plesk, which has a smaller but loyal following in the region, has not yet committed to a release date for ARM-native builds, though beta versions have circulated among early adopters.
The absence of Plesk on ARM is not a dealbreaker, but it creates a practical constraint. One hosting company in the Suntec City area, which had migrated approximately 60 percent of its shared hosting fleet to ARM by mid-2024, reported that it had to keep a separate cluster of x86 machines solely for customers who insisted on Plesk. Those customers paid a premium of $8 per month for the privilege of staying on the older architecture, a pricing decision the company described as a “legacy tax” that encouraged migration.
The energy-efficiency argument that matters more than the cost one
The hosting companies that have moved to ARM earliest tend not to lead with the cost savings when they talk about it publicly. They lead with the efficiency argument, because it is cleaner and because it aligns with a broader industry push toward sustainability reporting. But privately, the operators acknowledge a different motivation. Singapore’s data centre industry faces a growing constraint on power availability. The government has not issued new data centre construction permits freely since the moratorium imposed in 2019, and while the moratorium was lifted in 2022 for selected projects, the approval process remains stringent. For existing operators, the only way to increase compute density within a fixed power allocation is to use more efficient hardware. ARM servers are, for many workloads, the only way to do that without replacing the entire data centre’s cooling infrastructure.
The migration to ARM in Singapore’s hosting industry is not driven by ideology or vendor preference. It is driven by a simple arithmetic problem: the building has a fixed amount of power, the customers want more performance, and the old processors cannot deliver both. ARM is not a revolution. It is a workaround for a physical limitation that nobody planned for.
The hosting company in Tai Seng that started all of this, the one with the rising power bill, has now moved about half of its fleet to ARM. The other half remains on x86, running the database workloads and the legacy PHP applications that the engineers have not yet had time to rebuild. The power bill is lower than it was. It is not as low as the sales projections suggested. But it is lower, and that is enough for now.

📷 Photos: Brett Sayles (Pexels), Ravish Maqsood (Pexels)