TrndX raises $3.175M to make an AI cloud fit in a rack

Why it matters

TrndX is packaging years of founder experience in photonics and resource disaggregation into a full AI rack, targeting the cost and utilization problems created by GPU-heavy infrastructure.

Jitender Miglani has raised $3.175 million for TrndX, his year-old attempt to package compute, storage, networking and infrastructure management into a rack-scale system for AI workloads.

A Form D filed with the Securities and Exchange Commission on September 3rd shows that TrndX sold the full $3.175 million equity offering to 10 investors. The first sale occurred on August 21st under Rule 506(b), a commonly used exemption for private securities offerings.

The filing does not identify the investors or disclose TrndX's valuation. It also offers no revenue, customer or deployment figures. The financing therefore establishes that Miglani found 10 backers for the pitch, while leaving the commercial progress behind that pitch untested in public.

Miglani is returning to an infrastructure problem he has worked on across several generations of data-center technology: expensive hardware is frequently trapped inside fixed server configurations, even when another workload could use it. Public patent records associate him with inventions spanning optical circuit switches, reconfigurable computing clusters, remote PCIe transport and memory disaggregation.

TrndX describes Miglani as a former engineering leader at Drut Technologies, Calient, Juniper Networks, 3Com and Lucent. TrndX also credits him with more than 25 patents and three optical-switch products deployed by hyperscalers including Google. Patent records support his work in optical switching and disaggregated systems, though TrndX's deployment and product-count claims remain self-reported.

A second pass at disaggregated infrastructure

Miglani founded Drut Technologies in 2018 around a related thesis: data-center operators should be able to separate resources such as GPUs and memory from individual servers, connect them through high-speed photonic fabrics and assign them to workloads as needed. Patents assigned to Drut include systems for memory disaggregation and reallocating resources across server racks.

That history makes TrndX less of a fashionable pivot into AI infrastructure than a broader implementation of work Miglani has pursued for years. The AI boom has made the utilization problem easier to sell. GPUs are costly, AI clusters combine several infrastructure layers, and enterprises increasingly want private deployments that behave like public clouds without sending sensitive workloads outside their own facilities.

TrndX's UniFlex Platform is designed as an on-premises, cloud-like rack that combines servers, GPU acceleration, storage, networking and dedicated management controllers. TrndX says the platform can run bare-metal workloads, virtual machines and containers from one control plane, while allowing each hardware layer to scale separately.

Its T-POD software is intended to slice and compose compute, networking and storage for individual AI workloads. The architecture includes 25, 100 and 400 gigabit networking, pooled storage and support for infrastructure tools including Kubernetes, Slurm, Ansible and Terraform. Those specifications describe TrndX's intended product surface, rather than independently measured performance.

TrndX's current website identifies Venkata Prasanna Tumiki as co-founder and CTO and Kanwarjit Sabharwal as co-founder and co-CTO. Tumiki, whose background includes Drut, Calient, 3Com, Lucent and Sun Microsystems, is also listed as a director in the SEC filing. Sabharwal previously worked on networking and data-center systems at Google, Cisco and Calient, according to his TrndX profile.

The filing separately lists Ritu Miglani as an executive officer and director. It does not describe her operating responsibilities. TrndX operates from Nashua, New Hampshire, and lists an additional presence in Hyderabad, India.

The rack is the product

TrndX is entering a market where customers can already buy tightly integrated AI systems from incumbents or composable hardware from specialists. Nvidia sells DGX SuperPOD as a complete AI infrastructure architecture, while Dell offers integrated rack-scale systems. GigaIO and Liqid approach the same utilization problem through fabrics that pool and dynamically allocate resources.

The comparison that matters for TrndX is GigaIO, which describes its FabreX architecture as a way to compose compute, GPU, storage and networking resources at rack scale. GigaIO announced a $21 million first close of its Series B in July 2025, giving it substantially more disclosed capital and an established product line. Liqid, meanwhile, supplies software and fabric technology for systems that pool memory and accelerators across a rack.

TrndX is betting that buyers want a broader control layer spanning hardware orchestration, storage, networking and AI workload management, without being tied to one server or accelerator vendor. TrndX also promotes open-source components and the absence of per-resource licensing fees. No public benchmark yet establishes whether UniFlex can match the performance, reliability or operational simplicity of larger alternatives.

That leaves Miglani with a familiar founder's challenge. The technical pieces have precedents, including some in his own earlier work. The hard part is delivering them as one supported product that infrastructure buyers will trust under production AI workloads.

What the financing has to prove

For a hardware-heavy infrastructure effort, $3.175 million is an opening round rather than a large balance sheet. Rack systems require hardware procurement, integration work, validation and field support before software economics can take over. TrndX will also be selling to customers that tend to demand long tests and credible support plans before putting valuable AI workloads on a new infrastructure stack.

The financing buys Miglani room to turn a technically expansive product into a repeatable deployment. TrndX says UniFlex is available, but the public record does not establish production customers, pricing or performance results. The next useful evidence will come from deployments that show how quickly a rack can be installed, how effectively resources can be reassigned and whether the unified control plane reduces operating work in practice.

Miglani has spent years building the underlying components for composable data centers. Ten investors have now funded his attempt to assemble those ideas into a complete AI infrastructure product. TrndX still has to show that putting the cloud in a rack makes the rack easier to buy.