Skymizer
Skymizer is a Taiwan-based IP company that designs AI accelerator (LPU) cores and a proprietary compiler architecture (LISA) for on-device and on-premise LLM inference, licensing its technology to semiconductor firms and selling reference inference chips to enterprises.
- Company typePrivate
- Founded2013
- HeadquartersTaipei, Taiwan
- Headcount11–50
- GTM typeB2B
- OfferingSoftware
What Skymizer does
Skymizer Taiwan Inc. is a software-hardware co-designed intellectual property company founded in 2013 and headquartered in Taipei, with an engineering office in Hsinchu. The company develops AI accelerator IP for large language model inference, organized around a proprietary Language Instruction Set Architecture (LISA) and two LPU (Language Processing Unit) IP platforms: EdgeThought, a compiler-centric LPU IP targeting on-device and edge inference on cost-efficient silicon, and HyperThought, a higher-performance platform built on LISA v3 that supports multimodal, concurrent multi-model, and agentic AI workloads. In April 2026 the company announced HTX301, its first reference inference chip built on HyperThought, which integrates six LPU chips with 384GB of memory on a single PCIe card and enables 700B-parameter LLM inference at approximately 240W using a mature 28nm process and standard LPDDR4/LPDDR5 memory rather than HBM.
Skymizer monetizes through three revenue streams: IP licensing of LPU cores and the integrated software stack to semiconductor companies and device manufacturers, licensing of its compiler and runtime tools, and hardware product sales of the HTX301 inference accelerator via an enterprise early-access program. The go-to-market combines direct enterprise sales for flagship accounts with channel partnerships, including JFE Shoji Electronics in Japan, LoMicro in China, and CIRCLE Design Solutions in South Korea. The company is pursuing three primary customer segments: semiconductor companies and OEMs licensing LPU IP for integration into custom SoCs, enterprises seeking on-premise AI inference to avoid hyperscaler GPU cluster costs and address data-sovereignty requirements, and automotive and smart-mobility customers building GenAI-powered cockpits and in-vehicle assistants. Named deployments and partnerships include a knowledge-management deployment with New Taipei Metro (co-developed with the Institute for Information Industry), an in-vehicle GenAI cockpit solution with Inventec and VicOne, and a chiplet-integration collaboration with VIA NEXT Technologies.
The company is privately held and venture-backed, having raised approximately $18M in May 2025 from TGVest Capital, Leadbond Capital, and Taiwania Capital. Leadership combines deep semiconductor industry pedigree: CEO and Chairman Jim Lai (formerly president of Global Unichip Corporation and TSMC North America), CTO and Founder Luba Tang (compiler architect from Marvell and MediaTek), and CMO and EVP William Wei (formerly CTO at Foxconn and contributor to core Apple iOS/MacOS technology), with former Intel Corporate VP Wen-Hann Wang and former Microsoft Cloud AI leader Yu-Ting Kuo on the board. Revenue is not publicly disclosed, and the company employs 11–50 people.
Skymizer firmographics
Firmographics- Name
- Skymizer
- Legal name
- Skymizer Taiwan Inc.
- Website
- https://skymizer.com
- Company type
- Private
- Founded year
- 2013
- Operating status
- Operating
- Headcount range
- 11–50 employees
- Short description
- Skymizer is a Taiwan-based IP company that designs AI accelerator (LPU) cores and a proprietary compiler architecture (LISA) for on-device and on-premise LLM inference, licensing its technology to semiconductor firms and selling reference inference chips to enterprises.
- Ownership category
- akta.pro rank
Skymizer industry classification
Industry- Product category
- AI Accelerator IP
- NAICS
- Computer and Peripheral Equipment Manufacturing (3341), Custom Computer Programming Services (541511), Computer Systems Design Services (541512)
- SIC
- Electronic Computers (3571), Services-Computer Programming Services (7371)
- akta.pro industry
- AI Hardware IP & Chiplet/Packaging for Accelerators (EDA/IP, advanced packaging)Hardware, Technology & Telecom/Artificial Intelligence & Machine Learning/AI Infrastructure & Compute (GPUs/ASICs, accelerators)AI Compiler, Runtime & Kernel Optimization Software (CUDA/ROCm/XLA, graph compilers)Hardware, Technology & Telecom/Artificial Intelligence & Machine Learning/AI Infrastructure & Compute (GPUs/ASICs, accelerators)Model Serving, Inference & Deployment Platforms (APIs, Edge/On-Prem)Hardware, Technology & Telecom/Enterprise Platforms & Databases/Artificial Intelligence Platforms for EnterpriseOn-Device Inference Runtimes & SDKs (mobile/embedded)Hardware, Technology & Telecom/Artificial Intelligence & Machine Learning/Edge AI & On-Device Machine LearningCompute & Acceleration ICs (CPU/GPU/AI/FPGA/ASIC)Model Deployment, Serving & Inference PlatformsHardware, Technology & Telecom/Artificial Intelligence & Machine Learning/ML Platforms, MLOps & Model Lifecycle Management
Keywords
Where Skymizer is headquartered
LocationHeadquarters
- HQ city
- Taipei
- HQ country
- Taiwan
- HQ region
- Asia
Offices2 records
Markets served
Skymizer business model
Business model- GTM type
- B2B
- Offering type
- Software
- Cost components
- Technology or R&D, Personnel, Operations, Marketing or Sales
Revenue model
- IP Licensing: Core business model centered around licensing AI accelerator IP (LPU IP) and integrated software stacks to semiconductor companies, device manufacturers, and system integrators. Partners receive essential building blocks and comprehensive software stacks needed to develop high-performance, cost-effective AI solutions.
- Software Licensing: Licensing of compiler and runtime software technologies that enable AI models to operate with maximum efficiency on diverse computing platforms. Includes model conversion tools, simulators, and development support tools.
- Hardware Product Sales: Sale of HTX301 inference chip - the first reference chip built on the HyperThought platform. Enterprise-grade PCIe AI accelerator for ultra-large model inference targeting on-premise deployments.
Pricing tiers
| Model | Billing | Price |
|---|---|---|
| Other | Multi-year contract | Enterprise Early Access Program |
Go-to-market motion3 records
Distribution channels3 records
Marketing channels4 records
Skymizer product offering
Product offeringCore offering
Skymizer designs and licenses AI accelerator intellectual property (LPU IP) and compiler technology for large language model inference. Its product portfolio includes the EdgeThought LPU IP for on-device inference, the HyperThought LPU IP platform for enterprise and on-premise deployments, and the HTX301 reference inference chip that enables 700B-parameter LLM inference on a single PCIe card at approximately 240W. The company licenses these offerings along with integrated software stacks to semiconductor companies, device manufacturers, and system integrators.
Product overview
Skymizer is a software-hardware co-designed IP company offering a portfolio of AI accelerator intellectual property. The product lineup centers around two LPU (Language Processing Unit) IP platforms: EdgeThought for on-device edge inference and HyperThought for high-performance enterprise and on-premise deployments. HyperThought powers the flagship HTX301 chip—the company's first reference silicon—enabling ultra-large 700B-parameter LLM inference on a single PCIe card. All products are built on Skymizer's proprietary LISA (Language Instruction Set Architecture), which provides the foundational instruction set for language processing acceleration across the portfolio.
Differentiator
Problem solved
Functional benefit
Products and services
- EdgeThought EdgeThought is Skymizer's compiler-centric Language Processing Unit (LPU) IP designed for on-device LLM inference. It leverages software-hardware co-design with dynamic decompression, resource-efficient architecture, and compiler optimization to enable complex LLM inference on cost-effective silicon at the edge. It is targeted at chip designers and semiconductor companies building resource-constrained devices from IoT to AI PCs.
- HyperThought HyperThought is Skymizer's next-generation LPU IP platform for high-performance LLM inference, optimized for multimodal, real-time, and agent-based AI workloads. Built on LISA v3 architecture, it supports concurrent multi-model execution, prefill/decode disaggregation, and scalable deployment from edge to on-premise environments. It targets semiconductor companies and device manufacturers building high-performance AI solutions.
- HTX301 HTX301 is Skymizer's first inference chip built on the HyperThought platform. It is an enterprise-grade PCIe AI accelerator enabling 700B-parameter LLM inference on a single card at approximately 240W. Features six HTX301 chips with 384GB of memory, decode-first silicon architecture, and unified prefill/decode orchestration. It targets enterprises seeking to deploy ultra-large model inference on-premise.
Quantifiable outcome
- 384 GB total memory capacity on single HTX301 PCIe card
- +6 more outcomes
Companies that use Skymizer
Customer profileNamed customers3 records
Segments3 records
Ideal customer profiles3 records
Skymizer technology and API
TechnologyTechnology focussed Yes
API detail
- Has API
- No
- API docs
- API detail
Core technology
AI maturity
App detail
Integration7 records
AI capability6 records
Feature5 records
Skymizer partnerships and signals
Strategic signalPartnerships
Seven partnerships are on record, tiered flagship, strategic, core and regional.
- Institute for Information Industry (III)flagshipStrategic collaboration to advance Taiwan's transition from Edge AI to enterprise-grade On-Premise AI solutions. Partnership combines Skymizer's HyperThought IP with III's standardized middleware and heterogeneous-resource orchestration. Flagship product is HTX301 AI inference chip purpose-built for on-premise environments. Deployment includes knowledge management for New Taipei Metro.
- VIA NEXT TechnologiesstrategicCollaboration to develop next-generation AI inference infrastructure combining HyperThought LPU platform with VIA NEXT's chiplet integration and system-level validation capabilities. Focus on advanced packaging, EVB/DVB development, and modular architectures for enterprise-scale AI inference deployments. Highlights Taiwan's role in global AI semiconductor and advanced packaging ecosystem.
- JFE Shoji ElectronicscoreJFE Shoji Electronics serves as Skymizer's official distributor in Japan, responsible for promoting and selling AI acceleration IP and development solutions. JFE provides initial customer support and facilitates communication with Skymizer's technical team. Japanese customers gain access to HyperThought LPU architecture, EdgeThought platform, and hardware-software integration tools.
- InventecstrategicCollaboration to deliver robust in-car GenAI security solution for smart cockpits. Inventec's AAOS IVI powered by Qualcomm SoC provides in-vehicle experience with multi-screen support, while Skymizer accelerates LLM inference at the edge. Combined solution addresses automotive AI-powered cockpit needs.
- VicOnestrategicCollaboration with Inventec and Skymizer to deliver GenAI security solution for smart cockpits. VicOne's AI Guardian feature protects GenAI-based applications from sensitive data leaks, phishing, and unexpected responses. Safeguards automotive ecosystem from AI-specific security risks.
- LoMicro Information TechnologyregionalListed as agency partner in China. LoMicro is a professional Silicon IP and EDA distributor helping Chinese design houses with innovative technologies.
- CIRCLE Design SolutionsregionalListed as agency partner in South Korea. CIRCLE offers complete ASIC turnkey services including design, manufacturing, packaging, and testing.
Scale indicators3 records
Recent moves6 records
Expansion highlights5 records
Skymizer competitors and assessment
Company assessmentDirect peers
- Groq: Groq builds LPU (Language Processing Unit) inference accelerators — the closest direct competitor to Skymizer's HyperThought LPU. Both target ultra-low-latency LLM inference with custom silicon and decoder-optimized architectures.
- Cerebras Systems: Cerebras designs wafer-scale AI accelerators for inference and training. Competes with Skymizer in the high-end on-prem AI inference market targeted by HTX301, though with a fundamentally different silicon approach.
- SambaNova Systems: SambaNova offers enterprise AI inference systems using custom RDU chips with integrated software stacks. Direct competitor in the on-prem enterprise LLM inference market Skymizer targets with HTX301.
- Tenstorrent: Tenstorrent designs RISC-V-based AI accelerators and licenses IP — overlaps with Skymizer's IP licensing model while also competing on the hardware side in the on-prem inference market.
- FuriosaAI: Korean AI chip startup building server-class inference accelerators (RNGD). Competes in similar on-prem LLM inference market as Skymizer, and is distributed in Korea by Skymizer partner CIRCLE Design Solutions.
- Rebellions: Korean fabless AI chip startup focused on data center and edge inference accelerators (ATOM, REBEL). Regional peer pursuing similar custom-silicon inference strategy as Skymizer.
- Esperanto Technologies: Esperanto develops RISC-V-based AI accelerator chips for inference. Direct peer offering custom silicon IP and reference chips for on-prem LLM workloads, similar market positioning to Skymizer's HyperThought/HTX301.
Emerging players
- Etched: Etched builds transformer-specific ASICs for ultra-fast LLM inference. Targets the same on-prem inference economics story as Skymizer with a transformer-architecture-specialized silicon approach.
- Lightmatter: Lightmatter builds photonic AI accelerators for data center inference. Adjacent competitor pursuing alternative silicon approaches (photonic vs. LPU) for the same on-prem AI inference opportunity.
Broad incumbents
- NVIDIA: NVIDIA's GPU stack (H100, Blackwell, TensorRT-LLM) is the dominant incumbent in AI inference. Skymizer's on-prem single-card pitch is positioned against NVIDIA's GPU-cluster architecture, though NVIDIA offers overlapping capability rather than direct niche competition.
Market position
Strengths5 records
Weaknesses4 records
Competitive moat5 records
Key risks5 records
Key highlights7 records
Customer concentration
Skymizer social profiles
Digital presenceSkymizer financial estimates
Financial estimateRevenue estimate
Valuation estimate
Skymizer leadership team
Management profileNumber of profiles
Profiles7 records
Skymizer funding detail
Funding detailFunding overview
Funding rounds2 records
Investors2 records
Funding detail is available on the Subscription and Enterprise plan.Contact sales →
Skymizer M&A and investment
M&A and investmentM&A
Investments
M&A and investment is available on the Subscription and Enterprise plan.Contact sales →
Frequently asked questions about Skymizer
What does Skymizer do?
Skymizer designs and licenses AI accelerator intellectual property (LPU IP) and compiler technology for large language model inference. Its product portfolio includes the EdgeThought LPU IP for on-device inference, the HyperThought LPU IP platform for enterprise and on-premise deployments, and the HTX301 reference inference chip that enables 700B-parameter LLM inference on a single PCIe card at approximately 240W. The company licenses these offerings along with integrated software stacks to semiconductor companies, device manufacturers, and system integrators.
Is Skymizer a public or private company?
Skymizer is a private company. It is classified as venture growth investor backed and is currently operating.
When was Skymizer founded?
Skymizer was founded in 2013. It employs 11 to 50 people.
Where is Skymizer based?
Skymizer is headquartered in Taipei, Taiwan, in the Asia region.
How does Skymizer make money?
Three revenue lines are on record. IP Licensing is the primary driver. The others are software Licensing and hardware Product Sales.
Who are Skymizer's main competitors?
Direct peers on record are Groq, Cerebras Systems, SambaNova Systems, Tenstorrent, FuriosaAI, Rebellions and Esperanto Technologies. Emerging players are Etched and Lightmatter. NVIDIA is listed as a broad incumbent.
Does Skymizer have an API?
No public API is recorded for Skymizer.
What industry is Skymizer in?
Skymizer's product category is AI Accelerator IP. Its primary akta.pro industry code is HDAAAAAL, AI Hardware IP & Chiplet/Packaging for Accelerators (EDA/IP, advanced packaging), with a secondary code of HDAAAAAI, AI Compiler, Runtime & Kernel Optimization Software (CUDA/ROCm/XLA, graph compilers). Its NAICS code is 3341 and its SIC code is 3571.