Verdancia Capital’s first SPV Investment Targets the AI Inference Market & Exhibits Advantages of Supply-Chain Embeddedness

As AI permeates everyday life, its compute paradigm remains anchored in two core functions: training and inference. While training has benefited from years of investment in high‑power GPUs, purpose‑built inference hardware has received less focus yet is deemed the next durable growth frontier. Some have estimated the AI inference, conservatively, to be a USD 100 billion market by 2035.[1]

There has long been a technical issue associated with the hardware dedicated to inference: memory wall. This is the bottleneck created when processors perform computations far faster than data can be supplied from memory, causing the hardware to sit idle while waiting for data, something that even the best hardware still can’t fully address.[2]

Verdancia’s first SPV investment is targeting a Taiwan‑originated founder team that has developed a new line of inference processors built with a novel technology: “wafer-on-wafer”. To address how memory and processor can function well together, this approaches stacks layers of memory AND the SoC layer as a single unit, thus ensuring exceptional inference performance in terms of available memory, accessible bandwidth, and compute throughput. This technology has been lauded by industry leaders, and our SPV company will be amongst the first to bring it to market.[3]

This unique position is achieved because the project company is fully embedded within Taiwan’s semiconductor supply chain. Wafer-on-wafer is a craft that requires deep collaboration along the conception, engineering, and fabrication processes, something that outside competitors cannot readily access. Then there is the issue of foundry and component scarcity, which industry observers cite as a key bottleneck in the development of the inference market.[4] Due to longstanding supply chain relationships, our SPV target company has secured resources allowing it to expand relatively uninhibited.

By being part of the supply chain fabric, the SPV company gains access to real manufacturing feedback loops and the ability to align its product with the practical constraints and capabilities of upstream and downstream partners. This is the foundation that makes the transaction viable: Verdancia can underwrite not just the technology, but technology that the ecosystem itself has chosen to support.

The proposed SPV aligns directly with Verdancia’s investment thesis: (i) innovation rooted in the depth of Taiwan’s tech ecosystem, and (ii) strong embeddedness within Taiwan’s world‑class semiconductor fabrication. Given the scale of the coming inference requirement, this SPV offers a truly compelling opportunity to secure an early, meaningful position in a critical market that will be exceptional in scale.

The proposed size of the investment is USD 50 million for 20% shareholding of the target company; the investment period is projected to be six years, in a 4 + 2 configuration.

Please refer to the full Verdancia SPV1 Overview for details.

Verdancia is now structuring its second SPV investment, centered on a Taiwan originated company that manufactures advanced materials used in semiconductor substrates and packaging. These materials are exceptionally lightweight and exhibit superior thermal dispersion properties, making them highly relevant to next generation packaging requirements. The company is a growth stage business that has already secured vendor code status with a prominent semiconductor supply chain partner and has received trial orders, signaling that the product is validated and ready to scale. Verdancia will introduce this new investment opportunity, SPV2, in 4Q 2026.


  1. https://www.globemarketresearch.com/reports/ai-inference-chip-market
  2. https://wccftech.com/micron-says-ai-memory-wall-worsening-compute-outruns-hbm-bandwidth-by-3x-every-two-years/
  3. https://www.jonpeddie.com/news/qualcomm-pushes-compute-next-to-memory/
  4. https://www.businesstimes.com.sg/international/global/whats-stopping-asias-chipmakers-taking-nvidia-answer-lies-bottleneck