ReportGem ReportGem 中文

GLOBAL RESEARCH ARCHIVE

Stark Technology Inc (2480 TT, NT$150) - Order backlog reaches record high

Published: 2026-05-15Institution: Fubon Securities Investment Services Co., Ltd.Company / ticker: 2480.TWPages: 7Original language: 英语Evidence page: 1

Research evidence excerpt

Stark Technology Inc (2480 TT, NT$150) - Order backlog reaches record high

e

electronics and semiconductor manufacturing sectors, marking the highest level since the

What's changed

company's founding. About 64% of these are expected to be fulfilled within this year.

New Old %

Rating Buy Buy

Target price 180 180 0 Small language models gradually becoming mainstream in enterprise applications

EPS forecast 9.4 9.34 1 STI noted that over 80% of enterprises do not need model training but require AI

Target PE 19.1 19.3 inference, and not all enterprise applications need the massive computing power of large

language models (LLM). Small language models (SLM) have lower hardware requirementsFinancial summary Year-end: Dec 31

NT$mn FY25 FY26F FY27F and are better suited to most enterprises’ actual computing environments. With advances

Net sales 8,394 9,880 10,291 in model technology, some SLMs with only about 27 billion parameters now outperform

YoY growth (%) 11.4 17.7 4.2

models with hundreds of billions of parameters in specific industry applications. ForOp profit 997 1,187 1,272

YoY growth (%) 8.6 19.1 7.2 enterprises with strict data confidentiality requirements, SLMs, due to their moderate size,

Net profit 851 1,000 1,030 are easier to deploy inside enterprises than LLMs, avoiding the risks of sensitive data

YoY growth (%) 6.7 17.5 3.0

being uploaded to the cloud. When SLMs are combined with dedicated inference chips

EPS (NT$) 8.00 9.40 9.68

DPS (NT$) 7.8 8.4 8.8 (such as LPU) or general hardware (such as AMD or Intel chips), they offer higher cost-

PE (x) 18.8 16.0 15.5 performance ratios and much lower costs than renting large-scale computing platforms.

PB (x) 4.5 4.3 4.2 As major AI vendors (such as Amazon, Google, and OpenAI) reach a certain stage of

Div.

The English excerpt is extracted automatically from the cited source page and may contain layout or recognition errors. It is never batch translated.

Open report viewer