Anthropic hosts its investors this Wednesday, and a European ETP is already trying to track its value. Meanwhile, Nvidia is losing ground in China, Microsoft unveils its Maia 300 chip and ASML reports earnings.

Anthropic: a European ETP to track its value ahead of the IPO

Anthropic’s investor day takes place this Wednesday, October 14, and the public prospectus still hasn’t been filed. In the meantime, 21shares is launching a listed product in Europe that replicates the lab’s implied value without holding a single one of its shares.

The essentials:

• According to Bloomberg, Anthropic is hosting a select group of institutional investors at its San Francisco headquarters on October 14; the company has not commented.
• The public S-1 isn’t on EDGAR yet; the IPO marketing kickoff is reportedly being discussed for the week of November 9, with a listing before Thanksgiving (dates unconfirmed).
• On October 13, 21shares is launching a “synthetic” ETP on Anthropic (ticker CLAU), listed on the Vienna MTF, in euros.
• Fees announced by the issuer: 1.00% per year and 2.00% on exit.

What a “synthetic” ETP really means

According to the 21shares press release, the product offers continuous, unleveraged exposure to Anthropic’s implied value ahead of its IPO. It runs through a dedicated vehicle, 21shares x FalconX Limited. FalconX handles brokerage, execution and derivatives.

“Synthetic” means the product doesn’t own any Anthropic shares. It replicates a price built from derivatives markets. It’s the same logic as crypto perpetual contracts, which showed an average implied valuation of nearly $2 trillion as of September 30, according to CryptoBriefing. The last official funding round, in May, valued Anthropic at around $965 billion.

Another red flag: Anthropic has warned that some “tokenized” products on its stock are not valid. In those setups, no holder will be able to convert their instrument into shares at the time of the IPO.

The European investor’s view

An ETP is a debt security, not a UCITS fund. So it carries counterparty risk on the issuer and its vehicle, on top of price risk. The press release mentions a Swiss ISIN and a Vienna listing; it doesn’t say the product is PEA-eligible. As for the expected listing of the stock itself, the details (venue, timing, access) will depend on the upcoming prospectus.

The real test comes with the S-1 numbers: revenue, losses, compute costs. They’ll show whether an implied value near $2 trillion rests on data or on expectations.

Sources:

• 21shares Launches Europe’s First Pre-IPO Synthetic Anthropic ETP (GlobeNewswire)
• Anthropic Is Said to Plan Pre-IPO Investor Day (Bloomberg)
• Anthropic IPO: Oct 14 Investor Day, S-1 Still Not Public (FinanceFeeds)
• Anthropic’s implied valuation hits $2 trillion as tokenized trading takes off (CryptoBriefing)

Nvidia in China: 34% of the market, Huawei at 28%

According to an IDC report cited by Reuters, Nvidia’s share of AI chips in Chinese data centers fell to 34% in the first half of 2026. Huawei now holds 28%.

The essentials:

• Nvidia shipped 678,000 accelerators to China between January and June, down 44% year over year.
• Huawei shipped 572,000, up from more than 300,000 a year earlier.
• “Non-GPU” accelerators (1.016 million units) topped GPUs (991,000) in volume for the first time.
• IDC forecasts total shipments up 19% in 2026, but GPUs down 9%.

What the numbers say

The Chinese market took delivery of 2.008 million accelerators in the half. GPUs fell 26%, since few chips are approved for sale in mainland China. Other accelerators jumped nearly 80%. Huawei, T-Head (Alibaba), Kunlunxin (Baidu) and Cambricon account for 95% of that segment, according to Reuters.

The other players’ shares: T-Head 7%, Cambricon 5%, Kunlunxin 4%, Hygon 3%. Nine of the eleven Chinese manufacturers tracked by IDC increased their shipments.

The limits of the number

IDC doesn’t specify which Nvidia products make up the 34%. Nor is it clear whether the B200 and B300, which are banned from export to China, are included in the total. These are unit volumes, not revenue: a high-end chip weighs more in dollars than an entry-level one. Nvidia did not respond to Reuters’ request for comment.

The figure is also more favorable to Nvidia than Bernstein’s projection published in June, which pointed to roughly 8% market share in 2026 versus 50% for Huawei. The two sources don’t measure the same thing: one is an analyst estimate, the other a count of shipments.

What it means for the market

Jensen Huang argues that U.S. restrictions handed Huawei this Chinese market, which he puts at around $50 billion. For a European investor, Nvidia (listed on the Nasdaq) isn’t PEA-eligible and can be bought through a standard brokerage account or a UCITS ETF with semiconductor exposure. The China story remains a line to watch in upcoming earnings, because it weighs on revenue visibility, not on global demand.

Sources:

• Nvidia’s China AI-chip share falls as Huawei closes gap, report finds (Reuters via Investing.com)
• Nvidia’s China AI-chip share falls as Huawei closes gap (Business Recorder)
• Nvidia’s AI chip sales in China stall, as local chipmakers like Huawei take the lead (AP via ABC News)

Maia 300: Microsoft wants to cut its dependence on GPUs

On October 12, Microsoft unveiled Maia 300, its third generation of in-house AI chip. Deployment starts in 2027 and expands in 2028, at a scale of several gigawatts.

The essentials:

• Maia 300 packs eight stacks of HBM4 memory, versus six of HBM3E on Maia 200.
• A rack holds 96 accelerators, up from 48.
• Microsoft is keeping a mixed fleet: Nvidia and AMD hardware keeps arriving alongside Maia.
• The first uses are internal: Copilot, GitHub Copilot, Microsoft AI, Azure Foundry.

What Microsoft claims, and what it concedes

According to press reports, Microsoft is touting about 40% more useful compute per megawatt than the best GPU systems. Reuters mentions the possibility of more than a million chips deployed in 2027, a figure that remains tentative.

Maia lead Andrew Wall acknowledges that per-chip memory bandwidth is lower than on Vera Rubin (Nvidia) and MI455X (AMD). Microsoft’s argument is made at the rack level and on bandwidth per watt, not on the chip alone. Wall did not confirm the figure of 300,000 wafers per year reported in the press; he talks about a “higher” volume without putting a number on it.

What it means for the market

Every in-house chip shifts part of the budget that used to go to GPU suppliers. Yet Microsoft says it’s keeping Nvidia and AMD. So the impact plays out on the margin of incremental spending, not as a replacement.

For European investors, three links in the chain are affected. TSMC manufactures the chip; SK hynix and Samsung supply the HBM, demand for which should grow with inference according to the analysts cited; Nvidia and AMD remain the competing suppliers. None of these stocks is directly available in a PEA. Microsoft plans to announce external partnerships in 2027 and isn’t renting Maia directly to its cloud customers for now.

Sources:

• Microsoft Unveils Maia 300 Chip, Plans Gigawatt-Scale AI Buildout (Seoul Economic Daily)
• Microsoft prepares to deploy its Maia 300 AI chips in Azure data centers (CryptoBriefing)
• Microsoft aims to produce hundreds of thousands of Maia 300 AI chips (Techzine)
• Microsoft Maia 300 AI chip arrives with 288 GB HBM4 memory (Notebookcheck)

ASML reports earnings: will the AI boom hold up after 2028?

ASML reports third-quarter results this Wednesday, October 14, before the open. Two voices from the chip supply chain, its CEO and Broadcom’s, offer two readings of the cycle.

The essentials:

• Christophe Fouquet (ASML) sees the AI infrastructure buildout lasting two to five years, so possibly beyond 2028.
• Hock Tan (Broadcom) believes “the biggest constraint” could be capital.
• ASML’s Q3 guidance: €11.0 billion to €12.0 billion in net sales, gross margin of 55% to 57%.
• 2026 guidance: €43 billion to €45 billion in sales.

Two readings of the cycle

Fouquet describes an industry that is “trying to catch up with a very large backlog.” Hock Tan, speaking Monday at a conference in Singapore, sees wafer shortages easing around 2028 or 2029, while demand for compute could outstrip supply until 2029 or 2030.

The sticking point is financing. According to the Wall Street Journal, Broadcom is negotiating more than $50 billion in financing so that OpenAI can buy its custom chips. The Financial Times reports a $60 billion package for Anthropic’s use of Google chips, including $42 billion in debt guaranteed by Broadcom. These figures come from press reports, not press releases.

What we’ll be watching Wednesday

The sales consensus sits around €11.6 billion to €11.7 billion depending on the source, near the top of the range. The market will mostly be watching orders (bookings) and commentary on 2027. Options on the stock are pricing in a move of about 6% in either direction.

ASML plans to increase its annual EUV capacity by 30% in 2027, with another 30% increase under consideration for 2028. According to JPMorgan, the speed of final machine assembly would be the main constraint.

What it means for the market

ASML is listed in Amsterdam and is PEA-eligible. That makes it one of the few stocks in the AI chip supply chain that qualifies for a French saver. Its order book depends on capital spending by TSMC, Samsung and Intel, which in turn is tied to spending by the cloud giants. If financing tightens, orders are the first place it would show.

Sources:

• ASML Sees AI Chip Boom Beyond 2028 as Financing Risk Grows (Benzinga)
• ASML CEO Says AI Chip Boom Could Stretch Beyond 2028, but Broadcom Warns of Financing Bottleneck (Yahoo Finance)
• ASML Q3 Earnings Preview: What to Expect on Wednesday (AskTraders)
• ASML Stock: Options Traders See a 6% Move after Q3 Earnings (TipRanks)

DeepSeek invests in robot “brains” with Winverse

According to Bloomberg, DeepSeek is leading a seed round for Winverse AI, a young robotics startup founded in 2026. The $400 million figure is a valuation, not the amount raised.

The essentials:

• Winverse AI was founded in 2026 by Hu Han, formerly chief scientist for vision-language models and embodied AI at Tencent.
• The round reportedly comes to a few tens of millions of dollars and hasn’t closed.
• The post-money valuation would be around $400 million.
• Long-Z Investments, which is tied to Meituan, is also taking part in the round.

A bet on robot software

Winverse works on the AI that controls robots, the “brain” rather than the body. The amount is modest compared with Figure’s $39 billion valuation or the big humanoid robotics raises. What makes it significant is the investor: DeepSeek is extending its bets beyond language models.

In August, DeepSeek had already taken a stake of about 2.31% in Unitree Robotics, according to one of the sources consulted. Bloomberg also reports that DeepSeek is reportedly close to raising at least 80 billion yuan ($12 billion) for itself.

What it means for the market

Winverse is a private company: there’s no direct access for retail investors. The signal is more about the value chain. A Chinese AI lab is now funding robot software, on top of taking a stake in Unitree, whose listing on Shanghai’s STAR Market has been approved. On the European side, exposure comes through thematic UCITS robotics ETFs, not through these Chinese stocks directly. This is still a deal in progress, reported by the press and not confirmed by the companies.

Sources:

• DeepSeek backs ex-Tencent scientist’s robotics startup Winverse AI (Bloomberg via Investing.com)
• DeepSeek expands into embodied AI with early bet on robotics startup Winverse (DigiTimes)
• DeepSeek leads funding round valuing a Chinese robot AI startup at $400 million (CryptoBriefing)
• DeepSeek leads $400 million funding round for ex-Tencent scientist’s robot AI startup (Startup Fortune)

Edition prepared by our monitoring agent (data and sources as of Tuesday, October 13, 2026), reviewed according to our editorial policy. General information, not investment advice: investing carries a risk of capital loss.