Investing in AI chips: Nvidia competitors before their IPOs

Close-up of an AI chip with the label "AI" on a circuit board, surrounded by smaller components and gold conductive tracks

You ask ChatGPT a question, and somewhere in a data centre a chip from Nvidia is very likely doing the work. Perhaps you are wondering who wants to take this position from Nvidia and whether you can hold a stake in these companies before they go public. In May 2026, Cerebras became the first major challenger to list on the stock exchange. Many other AI chip makers are private and fund themselves through large rounds: Tenstorrent, for example, raised around 1.5 billion US dollars in October 2026.

AI chips at a glance

  • AI models run on specialised chips. Nvidia leads with its graphics processors, while challengers build chips for individual tasks such as answering queries.
  • Cerebras has been listed since May 2026. As of November 2026, Tenstorrent, Ayar Labs, SambaNova and Etched are private companies.
  • Tenstorrent develops AI processors based on the open chip architecture RISC-V and licenses its technology to groups such as Samsung and LG.
  • Ayar Labs connects chips with light so that data in large AI systems flows faster and with less energy.
  • Tenstorrent and Ayar Labs are held in the ARK Private Innovation ELTIF, in which you can invest through NAO with a one-off investment or a savings plan.
  • Chip start-ups need a lot of capital, and their valuations fluctuate sharply. In a fund, your money is tied up for a long time.

Why AI needs new chips

A language model goes through two phases. In training, it learns from huge amounts of data over several weeks, with thousands of chips computing at the same time. Then comes inference: the finished model answers queries every time someone types in a question. Both phases consume vast amounts of computing power and electricity.

Nvidia originally developed its graphics processors, or GPUs, for video games. They carry out many simple calculations in parallel, which is exactly what AI models need. On top of this comes the CUDA software, which developers have used for years to program Nvidia chips. Anyone who wants to switch has to rebuild part of their software, and that keeps many customers with Nvidia.

The more people use AI in everyday life, the larger the share of inference becomes. This is where the challengers see their opportunity. A chip tailored to a specific task is meant to complete it faster or with less power than a general-purpose processor. There is also a bottleneck between the chips: in large AI systems, data constantly flows back and forth between thousands of processors, and copper wiring is reaching its limits.

Our article Investing in AI shows which AI companies you can already reach today before an IPO. 

Nvidia competitors: who builds AI chips alongside the market leader

Nvidia faces competition from established chip groups such as AMD and Intel, from technology groups such as Google and Amazon, which develop their own AI chips for their data centres, and from a number of young specialists. This section is about these start-ups. We name them for orientation only, and this does not constitute a recommendation to buy.

Cerebras: the first major IPO

Cerebras, based in California, builds chips the size of a dinner plate. According to the company, its current chip, the WSE-3, is 57 times the size of the largest GPU. A growing share of its business runs through its own data centres, where Cerebras rents out computing power as a cloud service. In January 2026, OpenAI announced a contract with Cerebras worth around 20 billion US dollars, and Amazon Web Services has also been using the chips since March 2026.

A first IPO filing in 2024 failed, partly because Cerebras depended heavily on a single customer from the United Arab Emirates at the time. The company has been listed since 14 May 2026. For investors who wanted to get in early, the only route beforehand was through private investments. Our article Pre-IPO investing explains how such investments before an IPO work.

SambaNova and Etched: private and highly valued

SambaNova from Silicon Valley builds chips and complete systems that companies can use to run AI models in their own data centres. In July 2026, the company closed the first part of a 1 billion US dollar funding round at a valuation of 11 billion US dollars.

Etched takes specialisation one step further. Its chip, Sohu, works exclusively with transformer models, the architecture on which ChatGPT is also based. According to its own figures, Etched has collected pre-orders worth around 1 billion US dollars. There are no independent performance tests yet, and Sohu is not yet being delivered in large quantities.

A deal from December 2025 shows how sought-after the technology of these companies is: Nvidia paid around 20 billion US dollars for the technology of the inference specialist Groq. According to media reports, further candidates for an IPO include Rebellions from South Korea and d-Matrix. None of these companies can be invested in through NAO.

Tenstorrent and Ayar Labs in the ARK Private Innovation ELTIF

As of November 2026, two private chip companies are held in the ARK Private Innovation ELTIF. They work on different parts of AI hardware: Tenstorrent on the processor itself, Ayar Labs on the connections between processors.

Tenstorrent: AI processors based on RISC-V

Tenstorrent is based in Santa Clara and led by Jim Keller. Among other things, Keller helped develop the Zen processor architecture at AMD and Tesla's chip for autonomous driving. Tenstorrent sells AI computers in various sizes, from a single plug-in card to large server racks in its Galaxy range.

On the technical side, the company relies on RISC-V. This is an open instruction set, the basic language in which a processor understands commands. Anyone may use it without licence fees, whereas the rights to the widespread x86 and Arm architectures belong to individual companies. Tenstorrent licenses its RISC-V processor TT-Ascalon and its AI compute cores to companies that design their own chips, including Samsung and LG. Tenstorrent provides the accompanying software as open source, so that developers can run their models on it without being tied to one manufacturer.

Ayar Labs: transmitting data with light

Ayar Labs, based in San Jose, was co-founded by Mark Wade, who also runs the company. It develops small optical components, known as chiplets, that sit directly next to the computing chip in the same package. They convert electrical signals into light and send the data over optical fibre to the next chip. Experts call this co-packaged optics. Compared with copper wiring, this is meant to let more data flow while using less energy.

Ayar Labs works together with the market leader on this. Nvidia, AMD, MediaTek and the chip designer Alchip took part in the funding round in March 2026. The chiplets are manufactured by the semiconductor maker GlobalFoundries. With the fresh capital, Ayar Labs is expanding its production and testing capacity, including in Hsinchu in Taiwan.

Episode #60 of our podcast Diversify (in German) explains how ARK Invest selects companies like these and builds a fund from them.

Valuation and funding rounds

Private companies have no share price. Their value is set by investors in funding rounds: whoever provides new money negotiates the price at which they receive shares. This determines the valuation of the whole company.

At Tenstorrent, this valuation has risen significantly within just under two years:

  • December 2024: funding round at a valuation of around 2.6 billion US dollars.
  • October 2026: Series E of 1.5 billion US dollars at a valuation of 6.2 billion US dollars. The round was led by Trace Capital and Innovation X, with participants including Fidelity and Morgan Stanley.

Ayar Labs raised money twice in 2026:

  • March 2026: Series E of 500 million US dollars led by the asset manager Neuberger Berman, at a valuation of around 3.75 billion US dollars. The Qatar Investment Authority also took part.
  • September 2026: a further 150 million US dollars to expand series production. In a sale of existing shares at the same time, Ayar Labs was valued at more than 5 billion US dollars.

As of November 2026, neither company has named a date for an IPO. Valuations from funding rounds are snapshots. They can also turn out lower in the next round, for example if a product is delayed or market sentiment shifts.

AI chip stocks or private investment?

If you want to invest in semiconductors, you will find the big names on the stock exchange: chip developers such as Nvidia and AMD, contract manufacturers such as TSMC and, since May 2026, Cerebras. Semiconductor ETFs bundle such shares in one fund. With them, you hold stakes in groups that are already listed. As a rule, they do not include young chip start-ups such as Tenstorrent or Ayar Labs. Our article What are private markets? explains what distinguishes public and private markets.

In the NAO app, you will find the ARK Private Innovation ELTIF, which our investment team reviewed, as it does every fund in the app. ARK Invest launched it in January 2026. The fund invests in private technology companies in later growth stages. In AI hardware, these are Tenstorrent with its processors and Ayar Labs with its optical connections. The fund also holds AI developers such as OpenAI and Databricks as well as companies from other sectors such as SpaceX. AI chips are therefore one focus among several.

This is how an investment through NAO works:

  • You open an account in the NAO app and select the ARK Private Innovation ELTIF.
  • You invest with a one-off investment or a savings plan.
  • Your order is executed on the fund's dealing date.
  • A lock-up period applies until 26 February 2027, during which you cannot redeem your units. After that, redemptions are possible on the dates set out in the fund documents.

With a fund unit, you hold a stake in a portfolio whose composition can change at any time. You do not receive individual shares in Tenstorrent or Ayar Labs. You can find the fund on our private equity page, and our overview of the ELTIF explains how an ELTIF is structured.

Risks of investing in AI chip start-ups

If you hold a stake in chip companies through a fund, you should be aware of these risks:

  • Capital requirements: A new chip often costs hundreds of millions of US dollars before it is ready for series production. Many start-ups have to raise new money again and again, which can reduce the weight of early stakes.
  • Nvidia's market power: Many customers have built their software around Nvidia chips. A technically strong chip therefore does not automatically win through.
  • Manufacturing: The start-ups have their chips produced by a small number of contract manufacturers, most of them in Asia. Bottlenecks or export restrictions can delay deliveries.
  • Technology risk: It often takes years to get from announcement to delivery in large quantities. Whether a chip delivers what the manufacturer promises only becomes clear when customers use it.
  • Valuation: Valuations from funding rounds can fluctuate sharply. The fund values its holdings on the basis of funding rounds and models, and there is no daily market price.
  • Capital lock-up and loss: A lock-up period applies to the ARK Private Innovation ELTIF until 26 February 2027, and even after that, redemptions are only possible on set dates. Losses up to and including total loss are possible.

What does this mean for you?

Nvidia remains the top dog in AI chips, yet billions are flowing to its challengers. Some of these companies are private, and you can only hold a stake in them before an IPO through a fund. Before you decide, you can ask yourself:

  • How much money can you invest for several years without needing it?
  • What share of your wealth should go into young technology companies?
  • Do you already hold chip groups through shares or ETFs and want to add private companies?
  • What does the fund cost, and when can you get your money back? The answers are in the key information document.

Do you have questions about the fund? Our customer service team is there for you in person on working days. Feel free to write to us.

FAQ

Who are Nvidia's biggest competitors in AI chips?

The biggest competitors include the chip groups AMD and Intel as well as Google and Amazon with their own AI chips. Among the young manufacturers, Cerebras, SambaNova, Etched and Tenstorrent are some of the best known.

Can you buy Tenstorrent shares?

No. As of November 2026, Tenstorrent is a private company and not listed on a stock exchange. You can hold an indirect stake through the ARK Private Innovation ELTIF, which holds Tenstorrent.

Is there an Ayar Labs stock?

No. Ayar Labs is private and has not announced an IPO. Its most recent valuation was more than 5 billion US dollars in September 2026. You can hold an indirect stake through the ARK Private Innovation ELTIF, which holds Ayar Labs.

Which AI chip makers are already listed?

Listed companies include Nvidia, AMD, Intel, Broadcom and the contract manufacturer TSMC. Cerebras has been listed since 14 May 2026 as one of the first specialised AI chip makers.

What is RISC-V?

RISC-V is an open instruction set for processors that anyone may use without licence fees. Tenstorrent builds its AI processors on it and licenses its own RISC-V designs to other chip makers.

How risky is investing in AI chip start-ups?

Very risky. The companies need a lot of capital and compete with a dominant market leader. In a fund, your money is tied up for a long time.

This article is for information purposes only and does not constitute investment advice. Investments in private markets are long-term, capital is tied up for the term, and losses up to and including total loss are possible. You can find the fund documents and the key information document in the NAO app.

‍

‍

‍

Start now

Ready to join the NAO community?

Start today with access to exclusive private markets. Start investing like a millionaire, starting at just €1.

Create account now