Nano-core Chip Raises Over RMB 1 Billion for AI Inference

Direct answer: Hangzhou-based AI chip developer Nano-core Chip has raised more than RMB 1 billion (around $140–150 million) in a Series C2 funding round led by JianTou Investment, to develop and prepare for volume production of its compute-in-memory AI inference chips. The company says its first three chips have completed tape-out, with customer sampling and initial shipments planned for early 2027, and that the round takes its fundraising past RMB 2 billion in two months and RMB 3 billion in six months. It is the latest large bet on a simple idea: if AI is going to run everywhere, the chips that run it need to use far less energy and cost far less than today’s hardware.

What happened

Nano-core Chip announced the completion of its Series C2 round on 28 October 2026, according to TechNode’s report of the announcement, published on 30 October. The round raised more than RMB 1 billion. JianTou Investment led the round, with participation from Hangzhou Capital, Kaitai Capital, Xiaochi Capital, Yantai Caijin, Shengkun Investment, SDIC Taikang, Lingyang Fund and other institutions, while several existing shareholders increased their stakes.

The company did not disclose a valuation for the round, nor the exact amount above the RMB 1 billion floor, nor each investor’s contribution. PitchBook’s company profile records the Series C2 deal at $149 million, which is consistent with a round just above RMB 1 billion at current exchange rates, and puts Nano-core’s total funding to date at about $506 million. Those figures are worth stating carefully: the “more than RMB 1 billion” number is the company’s own account, and the valuation remains undisclosed.

Context makes the round more significant than a single number. The C2 round follows a Series C1 round of RMB 1 billion announced in September 2026, in which industrial investors including Huaqin Technology and Meige Intelligent participated. Combined, the company says it raised more than RMB 2 billion within two months and a cumulative RMB 3 billion within six months. That pace of fundraising — three billion yuan in half a year, before a single production chip has shipped — tells you how hot the AI inference hardware race has become, and how much capital investors are willing to commit before revenue proves the thesis.

Nano-core Chip’s formal name is Hangzhou Weina Hexin Electronic Technology Co., Ltd. (also rendered in some coverage as Hangzhou MicroNano Core Electronic Technology Co.). It was founded in 2021 and incubated at the Zhejiang Province institute of Peking University’s information technology research organisation. Hangzhou Capital, one of the round’s participants, has described Nano-core as the only compute-in-memory AI chip company among the “Hangzhou New Eight Steeds”, a local grouping of the city’s prominent young technology firms. Hangzhou, home to Alibaba and DeepSeek, has become one of China’s most important AI hardware hubs, and Nano-core is now one of its best-funded chip bets.

The money will be used for product development iterations and supply-chain preparation for mass production, the company said. That phrasing matters: the next twelve months are about turning taped-out designs into shippable, tested, supported products — the stage where many chip startups stumble, however good their architecture looks on paper.

How the technology actually works

To understand why investors are paying this much attention to a company with no shipping product, it helps to understand the bottleneck Nano-core is attacking. It is one of the oldest problems in computing, and AI has made it dramatically worse.

The memory wall

Traditional computers keep processing and memory in separate places. A processor fetches data from memory, performs a calculation, and writes the result back. For ordinary software, that round trip is fast enough to ignore. For AI inference — running a trained model to answer a question, recognise an image, or generate text — the model itself can contain billions of parameters, and every single output requires shuffling enormous quantities of data between memory and processor. The shuttling, not the arithmetic, becomes the bottleneck. It burns energy, generates heat, and limits speed. Engineers call this the von Neumann bottleneck, or the “memory wall”.

Compute-in-memory attacks the wall directly. Instead of moving data to the processor, it performs calculations inside or immediately beside the memory array itself. Data barely travels, so energy use falls and throughput rises. The idea is decades old, but it has become commercially interesting only recently, because AI inference workloads are so memory-bound, and because manufacturing techniques have finally made these designs practical at scale.

Nano-core’s 3D-CIM architecture

Nano-core’s approach is a self-developed architecture it calls three-dimensional compute-in-memory, or 3D-CIM. It combines three elements:

• DRAM-based 3D near-memory computing. DRAM is the dense, cheap memory used in most computers. Placing compute close to stacked DRAM lets the chip work through large models without constantly crossing the chip boundary.
• SRAM-based processing-in-memory. SRAM is faster and more expensive than DRAM, and it is well suited to the precise, repeated multiply-and-accumulate operations at the heart of neural networks. Doing those operations inside SRAM arrays removes another round trip.
• RISC-V heterogeneous computing. RISC-V is an open, royalty-free instruction-set architecture, increasingly popular with Chinese chip designers because it avoids dependence on proprietary architectures controlled by Western firms. A heterogeneous design lets the chip assign different tasks to the compute element best suited to them.

The company says this combination delivers high-efficiency, low-cost AI inference using mature manufacturing processes. That last point is strategically important: by designing for mature process nodes rather than the most advanced fabrication available, Nano-core can get chips made more cheaply, with better yields, and with less exposure to the export controls that restrict China’s access to cutting-edge chipmaking equipment. It is a bet that clever architecture on ordinary manufacturing can beat brute-force silicon on exotic manufacturing — at least for inference, where cost and energy matter more than raw peak performance.

Inference, not training

Nano-core’s chips are inference chips. That distinction is the key to its whole strategy. Training a frontier AI model — the months-long process of teaching a model from vast datasets — demands enormous clusters of the most powerful accelerators available, a market dominated by Nvidia. Inference is everything that happens afterwards: every ChatGPT answer, every photo enhancement, every voice transcription, every on-device assistant response. Inference already accounts for a large and growing share of total AI compute, and it is far more cost-sensitive, because it happens billions of times a day, on devices and servers where energy bills and hardware costs are felt directly.

Nano-core’s product lineup reflects that split. Its chips target edge devices such as AI smartphones and AI PCs, as well as cloud-side inference scenarios such as intelligent computing centres. An AI phone that can run a capable assistant locally, without sending every query to a distant data centre, needs an inference chip that is fast, cheap, and miserly with battery. A cloud provider running millions of inference requests needs chips that deliver the most answers per watt and per yuan. Both are Nano-core’s target markets, and both are enormous.

The company’s first-generation trio of chips has completed tape-out — the milestone where a design is sent to a foundry for manufacturing — and is now in the transition from tape-out to customer verification. Customer sampling and initial volume shipments are planned for early 2027. Until those samples are in customers’ hands and performing to specification, the architecture remains a promise, however well-funded.

Comparisons with rivals

Nano-core is not alone in chasing efficient inference, and the competitive field explains both the excitement and the risk. Several recent funding rounds show how crowded this race has become.

The closest comparison for UK readers is one this site covered days ago: South Korea’s FuriosaAI raised $620 million in a Series D round for its RNGD inference chips, aimed squarely at enterprise AI inference demand. FuriosaAI, like Nano-core, is betting that inference is a distinct market where a focused challenger can win against general-purpose giants. Our coverage of that round is here: FuriosaAI’s $620M Series D for AI inference chips.

Within China, Nano-core’s direct neighbours include Yizhu Technology, which is developing processing-in-memory chips based on 3D DRAM technology and announced financing exceeding RMB 1 billion on 23 October 2026, and Houmo Intelligence, whose SRAM-based processing-in-memory M30 and M50 chips reached mass production in mid-October, with the M50 securing more than 150 design wins. Houmo’s progress is the benchmark Nano-core will be measured against: a rival architecture, already shipping, already designed into customer products. Another Chinese inference-chip firm, Sunmmio, raised $277 million for 3D AI inference chips, covered here: Sunmmio’s $277M for 3D AI inference chips.

Outside China, the field includes London-based Fractile, which according to Companies House filings allotted around £577 million in Series C preferred shares between September and October 2026, with reports of an initial $250 million chip order from Anthropic — a reminder that the inference-chip race is not only a Chinese story, and that the UK has its own horse in it. And above all of them sits Nvidia, whose dominance in AI accelerators means every inference startup must answer the same question: why would a customer choose a specialist chip over the incumbent’s ecosystem, software, and scale? The honest answer, for most inference challengers, is cost per answer and energy per answer — the two numbers Nano-core’s architecture is designed to win on.

Timeline

• 2021: Nano-core Chip founded in Hangzhou, incubated at a Peking University research institute in Zhejiang.
• January 2022 onwards: Early-stage funding rounds build the company; PitchBook records a total of about $506 million raised across its history.
• May 2026: Series B3 round completed, continuing the company’s steady fundraising climb.
• September 2026: Series C1 round of RMB 1 billion announced, with industrial investors including Huaqin Technology and Meige Intelligent.
• 28 October 2026: Series C2 round of more than RMB 1 billion announced, led by JianTou Investment.
• 30 October 2026: TechNode publishes its report of the round; wider coverage follows the same day.
• Early 2027 (planned): Customer sampling and initial volume shipments of the first three chips.

What it means for UK small businesses

For a small business in Manchester, Leeds, or London, a chip funding round in Hangzhou can feel remote. It is not. The chips inside the phones, laptops, and cloud services your business buys determine what AI features you get, how much they cost, and how much of your data must leave your premises.

The most direct effect is cost. Inference is the recurring cost of AI: every time a customer uses an AI feature, somebody pays for compute. If compute-in-memory chips deliver the efficiency gains their designers promise, the per-query cost of AI services falls, and providers can offer more capable features at lower prices — or the same features with better margins. For a small firm using AI for customer service, document handling, or marketing content, that translates into cheaper subscriptions and more features included in the base tier, over a horizon of one to three years. Do not expect a price cut next month; chip cycles move slowly, and Nano-core’s own products are not due for sampling until early 2027.

The second effect is on-device AI. Phones and PCs with capable inference chips can run assistants, transcription, translation, and image tools locally, without sending audio, documents, or photos to a cloud server. For businesses handling client-confidential material — solicitors, accountants, clinics, financial advisers — that is a meaningful privacy and compliance advantage. It also means AI tools keep working when connectivity is poor, which matters for field teams, shops, and tradespeople as much as for offices. When you next refresh devices, ask vendors specifically about on-device AI capability and whether it runs locally or in the cloud; the distinction will matter more with every refresh cycle.

The third effect is supply-chain and geopolitical. China’s push for “independent and controllable” chip technology, using open RISC-V architectures and mature processes, is explicitly designed to reduce dependence on Western-controlled technology. For UK buyers, a more diverse chip supply base is generally good news — more suppliers, more competition, less single-point-of-failure risk. But it also comes with complications: export controls, shifting trade rules, and questions about which vendors’ hardware your customers and partners are comfortable with. A prudent approach is to stay vendor-flexible where you can, avoid locking critical workflows to a single chip vendor’s proprietary features, and keep an eye on how your software providers source their hardware.

Fourth, there is a lesson in Nano-core’s fundraising itself. More than RMB 3 billion raised in six months, before shipping, is a sign of how much capital is chasing AI infrastructure — and how much of that capital is betting on efficiency rather than raw scale. The UK’s own inference-chip efforts, including Fractile in London, are part of the same wave. If your business is in or adjacent to hardware, electronics, or AI services, the talent and investment flowing into this sector are worth watching: partnerships, hiring, and acquisition activity will follow the money.

Finally, keep expectations grounded. Nano-core has taped out three chips and raised a great deal of money, but it has not yet shipped a production chip, disclosed a valuation, or named a single customer. The history of chip startups is littered with elegant architectures that stumbled on yield, software support, or customer qualification. The right stance for a small business is interested but patient: track the early-2027 sampling milestone, watch whether named customers and design wins follow, and let the hardware prove itself before building plans around it.

Honest limits / what to watch

Several limits deserve emphasis, because the announcement’s headline number invites over-reading. First, the exact size of the round above RMB 1 billion is undisclosed, as is the valuation. “More than RMB 1 billion” is a floor, not a figure, and comparisons with other rounds should respect that uncertainty. Second, the company’s claim of RMB 3 billion in six months aggregates multiple rounds, including the September C1; it is a fundraising total, not revenue, and it says nothing about burn rate or runway. Third, tape-out is a milestone, not a product: chips can fail verification, miss performance targets, or arrive late, and customer sampling in early 2027 is a plan, not a result.

There are also questions the announcement does not answer. Which foundry manufactures the chips, and on which process node? What software stack will developers use to target them — the area where Nvidia’s CUDA ecosystem is hardest to displace? Which customers are sampling, and under what terms? Until those answers emerge, Nano-core’s competitive position is best described as well-funded and architecturally interesting, not proven.

Geopolitics is the other watch item. Compute-in-memory and RISC-V are attractive partly because they sidestep some export-control pressure, but the rules keep changing, and a chip company operating across US, EU, and Chinese markets faces compliance complexity that can affect availability and pricing in the UK. None of this makes Nano-core a bad bet — but it does mean UK buyers should treat supply-chain resilience as a live question, not a settled one.

One more caution: funding waves create noise. Not every well-funded inference startup will survive, and the eventual winners may consolidate the market quickly, as has happened in previous chip cycles. For small businesses, the practical hedge is the same as ever: buy capability, not narratives; prefer tools that work across hardware; and revisit your AI costs and device plans annually, when the hardware landscape has actually changed, rather than reacting to each funding headline.

FAQs

What is Nano-core Chip?

Nano-core Chip is a Hangzhou-based chip developer founded in 2021 and incubated at a Peking University research institute in Zhejiang. It designs AI inference chips based on compute-in-memory technology.

How much did Nano-core raise in its Series C2 round?

More than RMB 1 billion, according to the company’s announcement on 28 October 2026. The round was led by JianTou Investment, and the company did not disclose a valuation.

What is compute-in-memory technology?

It moves calculations into or beside the memory that holds the data, instead of constantly shuttling data between separate memory and processor chips. That can cut energy use and speed up AI inference, the work of running trained models.

When will Nano-core’s chips reach customers?

The company says its first three chips have completed tape-out, with customer sampling and initial volume shipments planned for early 2027. No production chip has shipped yet.

Why does this funding round matter for UK businesses?

Cheaper, more efficient inference chips feed directly into the cost of AI phones, PCs and cloud services that UK firms buy. A competitive Chinese chip sector also affects supply chains, prices and the geopolitics of AI hardware.

Is Nano-core competing with Nvidia?

Only partly. Nano-core targets inference, especially on devices and in cost-sensitive cloud settings, rather than the large-scale AI training systems where Nvidia dominates. Its nearer rivals are other inference-chip specialists.

Sources: TechNode, 30 October 2026; RecodeX Pro analysis of the round; PitchBook company profile for Nano-Core Chip.

CTA: Get a Free AI Consultation