LUMUT, PERAK — 16 September 2026 — Quiver Dive Team Lumut Sdn. Bhd. today announced two additions to its industry AI platform, now presented under the name AI Logistics Cloud: a GPU cloud service covering training and inference capacity, and a multimodal token service that meters text, image, video and voice usage on a common basis.
The company confirmed that its positioning has shifted. AI services are now the main offer, while freight forwarding, transportation and logistics agency work and electronics trading continue in operation as the source of the scenarios, documents and imagery that its models are built and evaluated against.
The GPU cloud service provides multi-GPU training nodes with checkpointing and resume, dedicated queues for LoRA and supervised fine-tuning jobs, elastic inference capacity for latency-sensitive endpoints, and scheduled batch throughput for document and imagery processing. Jobs are scheduled by priority agreed in advance so that production traffic is protected during peak periods. Utilisation, queue depth and cost per job are reported to the customer console. Where data cannot leave customer premises, the same scheduling layer runs on customer-side hardware under a private arrangement.
The multimodal token service responds to a practical problem: a single shipment file mixes documents, photographs, video clips and voice calls, and metering all of them as text tokens gives customers no usable view of where cost is generated. Under the new service, image usage is measured by resolution and processing depth, video by duration and sampling rate, voice by audio length and language pair, and text by input and output tokens per model version. A cross-modal task that reads a document, checks a photograph and drafts a reply is recorded as one traceable unit with its component costs. Separate quotas can be set per modality, which matters because image and video workloads are usually the fastest-growing cost line.
The platform architecture is now described in five layers: data and knowledge, model, compute, capability and service. The compute layer sits between models and capabilities so that capacity planning, model selection and agent design are handled together rather than in separate procurement conversations.
The roadmap was updated accordingly. The near term covers model APIs, GPU inference capacity, token metering and a sandbox for customer evaluation. The mid term adds hosted agent templates, multimodal token metering and GPU training queues for fine-tuning engagements. The longer term covers shared industry knowledge bases and cross-scenario agent orchestration with carriers, warehouses and customs service partners.
As with the rest of the platform, the company stated that model output supports decisions rather than replacing them, that evaluation results are reported per scenario against an agreed evaluation set, and that no model or service is presented as delivering guaranteed outcomes.