This chip startup just raised $135M in bets that the big bottle of AI doesn’t count – I remember.

Every time you ask ChatGPT a question, your request triggers a data race. Data leaves memory, goes through the CPU to be processed, goes to the GPU for heavy calculations, and then back – and the entire journey repeats every word the AI ​​makes.

The bottle is stable – it means to bypass the most expensive and powerful chips in the industry for every request. What is that inefficiency XCENAstarted with offices in South Korea and the US, they are trying to solve. The four-year-old startup created a chip that puts computing power closer to DRAM — fast, short-lived devices that store data that the processor is currently using — allowing routine tasks to be done close to memory, without expensive round trips between CPUs, GPUs, and memory.

If it works on a large scale, the impact on the cost of AI infrastructure could be huge, which explains the interest of global investors. Indeed, XCENA just raised $135 million in Series B at a cost of $570 million, bringing its total funding to $185 million.

XCENA CEO Jin Kim co-founded the startup in 2022 along with CTO Dohun Kim and CPO Harry Juhyun Kim, both former Samsung and SK Hynix, memory giants that supply the chips that power Nvidia’s GPUs. “CPUs and GPUs have both been smart for years. Memory never did. XCENA wants to change that,” Kim said in an interview with TechCrunch. “The recent rise in prices of memory and related stocks indicates a significant shift from AI architecture to memory architecture,” he added. (This month, the three companies that dominate the global memory chip market — Samsung, SK Hynix, and Micron — each crossed the trillion-dollar mark for the first time.)

XCENA is betting its business on the idea that “estimation is not a numerical problem; it’s increasingly a memory problem,” Kim said.

XCENA’s chip, MX1, communicates with the CPU via CXL (Compute Express Link) – essentially a dedicated channel between the processor and memory – to process data before it needs to leave the memory section. It brings computation to data, not the other way around. The company claims that what used to require 10 servers can be run on just one.

“Although GPUs excel at matrix multiplication – the heavy math behind AI-type learning – most of the round data calls, including corrections, KV cache management (a system that stores previously discussed issues so the model doesn’t have to be retrieved), and data caching, still work on CPUs.

The demand for memory solutions has increased since the second half of last year, and the company believes that the period is working in its favor.

Talks with several global memory vendors are at an early stage, although Kim declined to name them. The company’s ideal customers are hyperscalers who spend billions a year on AI-based products, where a small gain in memory efficiency can mean hundreds of millions in savings.

The MX1 is still a model. Mass-produced chips are expected to roll out of Samsung’s lineup by the end of 2026, with the company hoping to make money from 2027.

While neural processing unit (NPU) manufacturers are racing to compete with Nvidia for data processing, XCENA is looking at the memory unit that sits underneath it all.

XCENA’s biggest rivals include Astera Labs and Marvell, both Nasdaq-listed companies working on next-generation memory interconnects. Marvell is a large, established player already operating in the same space, Kim said, adding that the differentiator comes down to the smart products. “We have thousands of cores,” Kim said. Based on what most people use, Marvell’s method relies on fewer cores for all purposes in comparison.

The cores are built on RISC-V – an open source chip architecture – and optimized for performance., everything is deliberately kept small and practical. Beyond just the cores, XCENA develops its own internal registers, communication buses, and DRAM controller — a level of integration that many chip companies, including its biggest competitors, often outsource.

Seoul-based VC firms Altinum and IMM Investment led the Series B round, along with Corstone Asia and incumbents SBI Investment and Mirae Asset Capital. The company, which has more than 90 employees across offices in Pangyo, a tech hub outside Seoul, and in Sunnyvale, is also in talks with international investors for additional funding.