FPT AI Factory

Agentic coding is pushing LLM inference beyond traditional chat.
But it wouldn’t be a problem with this new architecture.

In Part 2 of our 3-part tech blog series “How To Solve ‘Broken’ LLM Inference In Agentic Coding”, we’ll take a closer look at the solution: separating prefill and decode into dedicated GPU pools - giving each phase its own hardware, and turning repeated context into a more efficient memory lookup.

👉 See how it works under the hood in Part 2.
https://short.factory.fpt.ai/0ZfW7

How Can AI Transform Engineering and Manufacturing?

FPT recently hosted a private AI workshop for Rinnai Corporation, a leading global manufacturer of kitchen and household appliances, welcoming nearly 100 participants from its engineering, IT, DX, and business teams.

The workshop explored how FPT AI Factory can enable AI-powered engineering and manufacturing, covering practical use cases across knowledge search, workflow automation, engineering data utilization, CAD/CAE, and Physical AI. Through live demonstrations, participants saw how these AI capabilities can help unlock technical know-how, streamline workflows, and accelerate product development.

The session also provided an opportunity for Rinnai and FPT to exchange perspectives on the evolving role of AI in manufacturing and explore potential areas for future collaboration.

VOTE FOR FPT AI FACTORY TO MOVE FORWARD IN AI AWARDS 2026

👉 Vote now: https://short.factory.fpt.ai/binhchonFPTAIFactory

Competing in the Solution category at AI Awards 2026, FPT AI Factory is proud to represent FPT in this year’s competition and following in the footsteps of FPT’s AI solutions that have made their mark at previous AI Awards competition.

Help FPT AI Factory to advance further in the competition with 3 easy voting steps
1️⃣ Scan the QR code shown in the picture or visit the link above
2️⃣ Click “Vote” and log in as instructed
3️⃣ Fill in the required information to complete your vote

Voting period: From 12:00 PM on September 07 to 12:00 PM on September 30, 2026

⭐️ FPT AI Factory is the AI Cloud platform developed by FPT Corporation. Acting as an “AI factory,” the platform enables businesses and developers to build and deploy AI more easily, efficiently, and flexibly, while contributing to a distinctive “Make in Vietnam” AI ecosystem shaped by Vietnamese enterprises.

Show your support and vote for FPT AI Factory today!

FROM BOTTLENECK TO SCALE: HOW TO SOLVE “BROKEN” LLM INFERENCE IN AGENTIC CODING

There is a challenge that many engineers are facing now: Large-scale Agentic Coding workloads can place significant pressure on LLM inference.

Duong Nguyen Thai, our AI Engineer of FPT AI Factory, points out, there are three actions we can take to solve the problem: routing requests based on worker cache state and current load, separating prefill and decoding into dedicated GPU pools, and implementing KV cache offloading.

When combined at the system level, these approaches help control latency, improve concurrent request handling, and use GPU resources more efficiently as Agentic Coding workloads continue to scale.

If you are facing the impending challenge, check out Duong’s full article via the link in the comments 👇

[ANNOUNCEMENT] NEW CUSTOMER SERVICE HOTLINE

FPT Smart Cloud announces a change to our 24/7 hotline, applying to all services: FPT.AI, FPT Cloud, FPT AI Factory, FPT CFS, and FPT Data Suite.
🔹 New Hotline: 1800 6139 (Toll-free for domestic calls within Vietnam)

Transition Timeline:
📍 01/09 – 31/10/2026: Both numbers remain active
New hotline: 1800 6139
Current hotline: 1900 638 399
📍 After 31/10/2026: 1900 638 399 will be discontinued. 1800 6139 becomes the sole hotline for all support.

Please save and update the new hotline 1800 6139 for seamless communication in the future. Thank you for your continued trust and partnership!

NVIDIA RTX PRO 6000 Blackwell is now live on FPT AI Factory!
Available ondemand as a GPU Virtual Machine, starting from $2.19/hour for early access. No long-term commitment needed.
👉 Try it now: factory.fpt.ai/gpu-virtual-machine

Built on NVIDIA's latest Blackwell architecture with 96 GB of GDDR7 memory, this GPU powers:
🔸 AI Inference: Run LLMs and generative AI up to 6x faster
🔸 Physical AI & Simulation: Power robotics and digital twins with NVIDIA Omniverse
🔸 Rendering & 3D Graphics: Photorealistic rendering up to 4x faster

Agentic coding is pushing LLM inference beyond traditional chat.

Coding agents repeatedly send large amounts of context back to the model, creating new challenges for keeping inference fast, efficient, and scalable.

In Part 1 of our 3-part tech blog series, our expert from FPT AI Factory unpacks why a straightforward LLM deployment can struggle with agentic coding workloads and what it takes to handle them more effectively.

👉 See what’s happening under the hood in Part 1.
https://short.factory.fpt.ai/a8HQO

GLM-5.3 - One of the strongest coding and security-capable models is ready to run on FPT AI Factory. No GPU infrastructure, no setup overhead.

👉 Request access to GLM-5.3 today: https://short.factory.fpt.ai/GLM_5.3

Why run it on FPT AI Factory:
✅ One API, one environment for GLM-5.3, Qwen, MiniMax, Gemini, and more
✅ Straightforward, enterprise-ready billing across every model
✅ Dedicated technical support to help you get started fast

Building Sovereign AI Capabilities for Banking - Finance with FPT AI Factory

At the 2026 Banking Sector Digital Transformation, FPT AI Factory showcases its all-in-one AI Cloud ecosystem to turn AI into business values
🚀 Robust, NVIDIA-accelerated GPU Cloud to run every AI workload
📈 Production-ready Token Factory to build, deploy, and orchestrate AI applications at scale
🤖 Enterprise-grade AI Agents to automate 100+ use cases

Visitors had the opportunity to experience the complete AI journey, from training and fine-tuning to deploying models on infrastructure fully under their own control.

With this ecosystem, FPT AI Factory reaffirms its commitment to powering banking institutions with the latest technology to stay ahead in the AI-first era.

GLM-5.3 is coming soon to FPT AI Factory

Released Aug 14, GLM-5.3 brings a major coding leap plus unexpected cybersecurity gains — all from post-training on the same base model as GLM-5.2.
✅ +50% coding on Z.ai Code Bench, open-source SOTA on Terminal-Bench 3.0
✅ 34.5% at Max effort using ~75K tokens (vs. 23.4% at 96K for GLM-5.2)
✅ SOTA on CyberGym, 2× GLM-5.2's exploitation performance

Soon on FPT AI Factory — same API, same billing as every other model. Stay tuned. 👀

Learn more https://factory.fpt.ai/news/glm-5-3-purpose-built-for-code