BORACAY — Choosing an AI server requires more than comparing hardware specifications. Businesses first need to identify the work the system will handle, because that determines the computing power, electricity, and cooling required, according to ASUS Philippines.
Testing an application on a smaller, local device can help companies assess those needs before investing in larger equipment. The starting task could be preparing finance reports or searching company records, with the results helping guide plans for wider use.
“First, they need to know how they want to really involve AI in their operations,” said Edcel Emerenciana, country product manager, NUC & Server, Open Platform Business Group, ASUS Philippines.
In an interview with Back End News on the sidelines of the VSTECS CXO Tech Summit, Emerenciana said understanding the application allows ASUS to assess its computing workload and recommend the resources needed to run it.
The challenge begins before the hardware purchase. Companies need a clear purpose for AI, whether that means reducing the time spent preparing reports, helping employees find information, or supporting customer service agents.
Each task gives the business something specific to develop and assess. It also provides a basis for deciding whether a local device is enough or whether the application needs a larger system.
Edge AI processes information close to where it is collected or used. A company can install a supported AI model on a local device and use it to develop an application.
ASUS offers the Ascent GX10 for this work. Emerenciana said companies can use it to develop tools for a particular department or explore applications for the wider organization.
Supported models can also run offline once they are installed, he said. This allows businesses to perform those tasks locally without sending each request to an online AI service.
For companies still developing their applications, the device provides a starting point. They can work on the task they want AI to perform before deciding what equipment they need for broader use.
Emerenciana described an AI agent that could search company files to answer an employee’s question. For example, the employee could ask how many products were shipped during the previous quarter.
The agent would check the relevant records, reducing the manual work of opening spreadsheets, finding entries, and adding up figures. It would first need to be set up for the task and given access to the appropriate files.
Emerenciana distinguished general-purpose servers, which host company applications and databases, from systems used to run AI applications. He cited tools that help finance teams prepare reports as one possible use.
Larger AI systems enter the discussion when companies have applications ready to put into use across departments. At that stage, the hardware assessment needs to account for the computing work those applications will require.
There is no single upgrade point in the approach he described. The decision depends on the application, rather than a fixed sequence in which every business starts with an edge device and eventually buys an AI server.
A smaller device may meet a particular need. Expanding an application may require a different setup, but companies first need to assess what that expansion involves.
The operating requirements matter alongside computing capacity. Emerenciana said identifying the workload helps ASUS determine power needs. More demanding installations, such as AI data centers, may also require cooling systems that use water.
This makes the intended application central to the spending decision. It helps establish both the equipment a business needs and the resources required to operate it.
Local processing is another consideration. Emerenciana said companies may want to keep information within their own systems instead of uploading it to an external service.
“The processing of their data is within their reach, so your control is very important in terms of data security,” he said.
That control must also cover what AI applications can do with the information. Emerenciana stressed the need for approval rules when agents take actions, including seeking permission before sending information.
An agent that searches company records needs access to files to perform its task. Businesses therefore need to define its permissions alongside the application itself. Keeping information on local hardware does not replace those controls.
Whether the task involves customer calls, finance reports, or company records, the equipment decision follows the application. Defining the work first gives businesses a clearer basis for assessing server capacity, power, cooling, and data controls.