Zettabyte and Bowrington explore a Taiwan GPU build with asset-level capital

The memorandum of understanding Zettabyte signed with Bowrington Capital commits no sites, no megawatts and no money; Bowrington's capital, the document suggests, would sit at the GPU asset level, a role to be explored as co-owner, co-developer and co-investor. What the two firms have committed to instead is a three-part study: assessing the commercial requirements for dense GPU computing, identifying data center sites in Taiwan, and establishing the total cost of ownership for the physical plant that computing would sit in.

Zettabyte, the operating side, sells GPU cloud services and says its fleet holds more than 67,000 GPUs with access to Nvidia's Blackwell, H200 and H100 hardware; it also claims 1.5GW of planned deployments and took an undisclosed investment from Headline Asia in January. Those are the company's own figures, describing a business whose capital intensity sits in the silicon, with the shell built around it following as a financing question, and the MoU's coverage does not say where that planned 1.5GW would sit or how much of it is financed.

Bowrington's entry point carries weight because capital placed at the GPU asset level, skipping the data center entity, keeps the collateral identifiable and leaves demand risk with the operator; merchant data-center debt has become a leasing-bet trade, with lenders underwriting the forecast rather than the customer, and a single tenant miss would reprice the shelf. Compute does not sidestep that logic, but it prices differently: a named, depreciating, redeployable asset against a contracted workload is a trade a credit committee can underwrite without believing anyone's absorption curve.

Taiwan supplies the other half of the pitch. Kuma Xiong, an investment officer at Bowrington, points to Taiwan's classification of green data centers and AI computing centers as public construction projects, which he says unlocks access to local institutional capital. Framed that way, a permitting category becomes a capital-formation tool; this publication has argued that consent now decides what gets built, capital following in its wake, and Taiwan's version runs the thesis in reverse, manufacturing the consent first and selling the access it creates.

What the MoU has not produced is a parcel, a commitment, or a figure. Jeff Lin, Zettabyte's executive director, frames that gap the way the industry now does: demand for AI computing is growing faster than the infrastructure serving it, and closing it will require software, compute systems engineering and long-term capital in the same room. The test is whether Bowrington's GPU-level role hardens into a signed co-investment and whether a site in Taiwan gets named; until one of those appears, this is a diligence agreement with a good story attached.