Why We Invested in C2i and Unlocking Power Delivery for AI

Ravi Jain, Vasan Churchill, Aakriti Singh, TDK Ventures

AI is beginning to reshape our world, both digitally and physically. The computing infrastructure on which AI is so reliant is stressed for power. One of the core issues is to figure out methods to get power to xPUs, especially GPUs, in the most efficient way possible. It was this premise above all, and our internal explorations around it, that led the TDK Ventures team to C2i Semiconductors (C2i). Founded in June 2024, C2i is developing a software-defined Integrated Vertical Voltage Regulator (V2R) platform with a clear vision to deliver grid-to-core power solutions that make power delivery more reliable, efficient, and intelligent. In this article we aspire to describe the landscape of power delivery today, the value proposition that C2i provides, and ultimately why we invested.

The Problem: Power Delivery in the Age of AI

There are a few macro changes happening in AI computing overall.

a) The data center energy consumption is expected to be upwards 1000GWh. This is a massive increase in a short period of time.

b) As new generation of GPUs such as Feynman Ultra go into production in 2029–30, the absolute amount of current that is consumed by each GPUs moves beyond 5000A. This is ~5x change from the current generation of Blackwell GPUs in production today.

c) Each AI compute rack in the data center goes from 10s of kW to 1MW.

In other words, we expect to see significantly more power flowing into much smaller die area and more densely packed sever motherboards in the racks.

To achieve and package those power levels, current density, is expected to rise from less than 1 A/mm² to around 4–6 A/mm². In short, to meet the rise of AI, computing requires more power than ever in ever smaller packages which amount to a physics environment which very simply today’s architectures were not designed to handle.

Figure 1. The progression of characteristic size and operating voltage continue to decrease and the current draw increase

The fundamental problem is the “last inch.” Current power delivery solutions are placed at a distance from the chip, forcing current through a relatively long path. This innately introduces resistive driven losses that scale with the square of current (P = I²R) [4]. For example, a system moving from 875A to 4500A, those losses grow by at least 25x under typical power delivery topology.

The size of the opportunity is massive. A 10% improvement in grid-to-core energy efficiency is a $10B/year energy savings opportunity. A good chunk of these energy savings can be unlocked by new & efficient modalities of vertical power delivery and VRs. Alongside energy loss, these longer paths also produce larger voltage transients and slower response to load steps, potentially reducing GPU utilization and accelerating device aging over time [5]

Naturally there are ways that standard topologies could be engineered to reduce or mitigate these issues, but truly solving the “last inch” problem is more a matter of disruption rather than iteration: reimagine the topology altogether, which is precisely what C2i is attempting to do.

Figure 2. Conceptual layout of a power delivered on chip, wherein the distance away inherently leads to resistive losses proportional to current squared.

Our Investment Thesis

The main points of our investment thesis come from our deep exploration of the voltage regulation and IVR space included:

• Vertical power delivery is likely to be adopted near-term, with backside and landside placement first and substrate-embedded approaches following.

• The 12V intermediate bus converters (IBC) will start to regulate at lower voltages

• Scalable, modular solutions that can serve general compute, HPC, and AI workloads have the potential to create large, durable platform businesses.

• IVR solutions, which innovate at the level of control and power-delivery topology (while relying on proven passive technologies), are the most likely to reach mass production and generate meaningful revenue over the next decade.

• Approaches that depend heavily on novel ferromagnetic inductor materials face long reliability and qualification cycles that could delay monetization well beyond the critical 2026–2030 window.

Our “King of the Hill” threshold was then set as: control loop latency; passive components under 1mm total height; a roadmap to meet both types of IVR technologies; target current densities in the 2–4 A/mm² range scaling to 10,000A; and a team that was already building alongside anchor customers rather than waiting for a “perfect” solution..

Creative Disruption with C2i

C2i met all of our criteria and was built around exactly the kind of topology-first innovation we were looking for. C2i has developed two core proprietary IP blocks that together underpin their entire product family. The first, called Manas, is a software-defined controller.

It does this by rethinking the control loop algorithm and topology, achieving voltage-source-like wide-band response without the efficiency penalties that typically accompany that approach. The second IP block, called Sarayu, is a smart power stage that breaks the classical trade-off between switching frequency, efficiency, and electromagnetic interference (EMI) ringing. By precisely controlling switching frequency in a novel way, Sarayu enables higher switching frequency without efficiency loss, lower EMI, and better thermal performance.

C2i’s platform is projected to deliver over 96% power conversion efficiency, compared to 94% for incumbent solutions, enabling GPUs to run around 4°C cooler and reducing aging rate to 10% of what we’d currently see. As an example, consider a 100MW AI data center running Blackwell-class GPUs. The combination of efficiency gains, reduced heat removal load, and higher input voltage capability is estimated to translate to approximately $12 million in annual energy savings. Importantly, C2i achieves these results without relying on novel magnetic materials, with credible industry partners.

The core TAM is projected to grow from approximately $2.7B currently to over $12B by 2032, reflecting a nearly 29% growth rate [6]. The Data Center and AI segment alone (which today represents almost $1B) is expected to grow to $9B by 2032 at a 44% CAGR, driven by the explosion in GPU and accelerator power demand. Beyond the VR market itself, the structural tailwinds are compelling. Vertical Power modules are already becoming non-negotiable for high-performance chips. In other words, this is a busy market, and we believe C2i, will be at the center of engagement clearing the way for the future as AWS, Google, Meta, and Nvidia all work toward their own next generation (TPU programs, infrastructure, Feynman GPUs, etc.) [8].

Making all of these possible is the C2i team. C2i was founded in June 2024 by four former Texas Instruments technologists: Ram Anant (CEO, ex-Director TI Power Management Products 30+ years in power semiconductors), Preetam T (CTO, ex-Chief Technologist TI, 25+ years, 80+ US patents), Vikram Gakhar (Lead Mixed Signal, 25+ years, 20 US patents), and Dattatreya BS (Lead Power, 25+ years, 15 US patents). The founders have worked together for several years prior to starting C2i and collectively hold deep expertise in the exact domain they are now addressing. Beyond their credentials is their track record in execution.

In roughly 18 months from founding, the team has grown to 65 people, filed 9 patents (with 22 more pending), signed agreements with leading foundries and other technology and engaged meaningfully with a high potential customer pipeline.

C2i team and Advisors

Synergy with TDK

TDK is honored to have been a global leader in advanced passive components including thin-film capacitors, inductors, and related technologies that are integral to any high-performance power delivery system. C2i’s expertise lies in the intelligent control topologies and power stages that govern how those components perform. These two capabilities are not merely adjacent — they are deeply complementary, and together they open a natural path toward co-developing next-generation vertical power modules that neither company could bring to market as effectively on its own.

Just as importantly, the relationship gives TDK direct insight into the power requirements being defined by hyperscalers and AI OEMs today — ensuring our own component roadmap stays tightly aligned with where the market is heading. We see this as a genuine equal-win partnership, and one of the clearest examples of “TDK Goodness” we have encountered in this investment cycle.

Looking Ahead

The transition to AI-scale computing is one of the most significant infrastructures build-outs of our generation, and it will require innovation at every layer of the stack — including, critically, the layer that delivers power to silicon. We believe C2i has built the right technology, with the right team, at the right moment to help lead that effort. What excites us most is not just the opportunity in front of them, but our ability to actively contribute to their success.

This is precisely where “TDK Goodness” comes in. Beyond capital, we are committed to bringing the full weight of TDK’s global capabilities, manufacturing expertise, and deep customer relationships to bear as C2i scales. Whether that means co-developing next-generation passive components, opening doors with hyperscalers and AI OEMs, or helping navigate the complex supply chain realities of bringing a semiconductor platform to production — we intend to be a genuine partner in every sense of the word. The C2i team has already shown us what they are capable of with limited resources. We are proud to be part of what comes next.

References

[1] F. Sukhoi, “Where Will the World’s Electricity Come From in 2030?,” Impakter, Mar. 2, 2026. [Online]. Available: https://impakter.com/where-will-the-worlds-electricity-come-from-in-2030/

[2] Goldman Sachs Global Institute, “Rising power density disrupts AI infrastructure,” Goldman Sachs, May 2025. [Online]. Available: https://www.goldmansachs.com/insights/articles/rising-power-density-disrupts-ai-infrastructure

[3] M. Studer, “The energy challenge of powering AI chips,” Robeco, Nov. 6, 2023. [Online]. Available: https://www.robeco.com/en-int/insights/2023/11/the-energy-challenge-of-powering-ai-chips

[4] J. Morra, “Can integrated voltage regulators take on AI power delivery?” Electronic Design, Apr. 2024. [Online]. Available: https://www.electronicdesign.com/technologies/power/article/55021283/electronic-design-can-integrated-voltage-regulators-take-on-ai-power-delivery

[5] J. Morra, “Empower’s voltage-regulator IC enables vertical power delivery for AI chips,” Electronic Design, Oct. 2024. [Online]. Available: https://www.electronicdesign.com/technologies/power/article/55237347/electronic-design-empowers-voltage-regulator-ic-enables-vertical-power-delivery-for-ai-chips

[6] Transparency Market Research, “Integrated voltage regulator (IVR) market: Global industry analysis, size, share, growth, trends and forecast, 2024–2031,” Transparency Market Research, Jun. 2024. [Online]. Available: https://www.transparencymarketresearch.com/integrated-voltage-regulator-market.html

[7] KPMG Corporate Finance, “Semiconductor industry and M&A update — Summer 2024,” KPMG, Sep. 2024. [Online]. Available: https://corporatefinance.kpmg.com/us/en/insights/2024/semiconductor-industry-update-summer-2024.html

[8] A. Shilov, “Nvidia enterprise GPU and CPU roadmaps: Rubin, Rubin Ultra, Feynman, and silicon photonics,” Tom’s Hardware, Aug. 2025 [Online]. Available: https://www.tomshardware.com/tech-industry/semiconductors/nvidia-enterpise-roadmap-rubin-rubin-ultra-feynman-and-silicon-photonics