Moonshot's $30B path opens after Kimi K3

Moonshot AI released an open-weight model last week that followed by a selloff in semiconductor stocks that some market commentators compared to prior AI-model-related market reactions. Days later, the same company reportedly asked its shareholders to approve a Hong Kong listing at a pre-money valuation near $30 billion. In a separate corner of the market, CuspAI raised $450 million at a $2.6 billion valuation to aim AI at chipmaking materials, with Jeff Bezos among the backers.

‍

The lab that shook the chip market is reported to be heading to Hong Kong

On July 16, at the World AI Conference in Shanghai, Beijing-based Moonshot AI released Kimi K3, a 2.8-trillion-parameter open-weight model and the largest of its kind at launch. In third-party benchmarks it reportedly matched or beat several leading US models. Some market participants compared the release to the DeepSeek market reaction, and semiconductor stocks sold off around the same period: the Philadelphia Semiconductor Index fell into correction territory, with reporting putting the drop in global chip-related value in the trillions. (The Next Web)

Within days, the company reportedly circulated a resolution to shareholders seeking approval for a Hong Kong IPO, and began unwinding its offshore holding structure to clear the way, a process that once started often runs about six months. The timing, valuation, underwriters, and completion of any listing remain uncertain. DeepSeek, for all the noise it made last year, never went that route. Bloomberg-sourced reporting puts the pre-money valuation near $30 billion, with Goldman Sachs and CICC in talks on the underwriting and reported annualized revenue climbing from about $200 million in April to $300 million in June. (Yahoo Finance / Bloomberg)

For private-market readers, the selloff matters less than the listing. A Hong Kong listing, if completed, could provide an additional public reference point for investors evaluating parts of the China-AI cohort, though any read-across would be limited by differences in business mix, cost structure, capital commitments, liquidity, governance, and disclosure. Zhipu and MiniMax already trade there; Moonshot would be the first of the three to compete directly with the US labs at the frontier. Any read-across to OpenAI or Anthropic still has limits, given differences in business mix, cost structure, capital commitments, and disclosure.

Reported private-company financials and secondary-market indications may be unaudited, incomplete, non-standard, or based on limited transaction activity. They should not be relied upon as fair value, executable pricing, or a basis for any investment decision.

The AI money moved toward materials

One of the week's larger rounds went to a company using AI on materials rather than language. CuspAI, a two-year-old British startup applying models to discover new materials for chipmaking, energy, and manufacturing, raised $450 million at a $2.6 billion valuation, roughly five times its $520 million mark nine months earlier. Jeff Bezos's Bezos Expeditions joined the round alongside Kleiner Perkins and NEA. (CNBC) (Tech Startups)

Alongside the raise, CuspAI launched an "AI Materials Foundry," a coalition it says includes Nvidia, Meta, Hyundai, Samsung, and dozens of other companies, built to design materials the semiconductor and energy industries need.

Capital that spent two years chasing software agents is now paying up for AI aimed at materials, energy, and chips. Underwriting a materials lab looks nothing like underwriting a chatbot, and it points at a different set of companies. For a secondary desk, these are names that will likely surface in tender activity over the next 12 to 18 months, long before any of them are liquid, but any tender activity, timing, pricing, availability, or liquidity remains uncertain.

Reported private-company financials and secondary-market indications may be unaudited, incomplete, non-standard, or based on limited transaction activity. They should not be relied upon as fair value, executable pricing, or a basis for any investment decision.

📈 Data Point of the Day

$1.65 trillion

That's reportedly the estimated off-balance-sheet debt across Alphabet, Amazon, Meta, Microsoft, and Oracle, according to a Nikkei study published this week — up roughly eightfold since 2022, and larger than the ~$1.35 trillion of debt those five carry on their balance sheets. Most of it is long-term commitments for data-center leases, GPUs, and servers not yet delivered, which accounting rules keep off the books. It landed the same week Kimi K3 revived the question underneath the whole AI trade: if frontier capability stops being scarce, the spending that justifies these commitments gets harder to defend. For AI-side private names whose marks assume the buildout continues, it belongs next to the valuation. (Nikkei Asia) (Semafor)

💨 Quick Takes

• A US judge approved Anthropic's $1.5 billion author settlement — the largest known US copyright settlement, at roughly $3,000 per work, with claims filed on more than 92% of the ~480,000 works covered.
• Microsoft and Mistral expanded their partnership into a multibillion-dollar European infrastructure deal — Microsoft will draw on Mistral's Europe-based data centers, which the French lab is expanding with thousands of Nvidia GPUs.
• OpenAI says Codex and ChatGPT Work reached 10 million users — roughly double the tally from two weeks earlier, though the company hasn't defined "active" or split the two products.
• Sony Music filed a second suit against Udio over 30,117 recordings — the last major label without a Udio license, pushing the AI-music company's potential statutory exposure sharply higher.
• US VCs have deployed a record ~$412.7 billion in 2026 so far — with AI absorbing the overwhelming majority of it, and most other startups seeing little of the record.

🎓 Manual

VIE (Variable Interest Entity) structure

A variable interest entity is an offshore holding arrangement Chinese companies have long used to take foreign capital and list abroad without granting outsiders direct ownership of the mainland operating business. Control runs through contracts rather than equity. Regulators on both sides have grown wary of it, and a company preparing a Hong Kong listing may need to unwind or restructure its VIE first, as Moonshot is reportedly doing. For anyone reading a China-AI listing story, the VIE is a reminder that the corporate plumbing can matter as much as the valuation: what a foreign shareholder buys may sit a contract or two away from the business itself.

👀 What We're Watching

• The rest of the Chinese AI labs, and their listing clocks.Moonshot's reported move follows Zhipu and MiniMax, both already public in Hong Kong. If more China frontier labs begin listing prep, the secondary market gets a widening set of public comps for a cohort that until recently priced only in private rounds. (Yahoo Finance / Bloomberg)
• How closely the AI-infrastructure debt gets read. The $1.65 trillion figure landed while chip stocks were already wobbling on the Kimi K3 reaction, and markets steadied only partway through the week. Additional attention to how these commitments are disclosed, if it develops, may add context for how the buildout is being financed. (Semafor)
• Whether "physical AI" becomes its own funding lane. CuspAI's $2.6 billion mark and its multi-company materials coalition sit alongside a broader tilt toward AI applied to chips, energy, and materials. More rounds in that cluster, if they occur, would point to a hard-tech cohort worth tracking before any of it reaches the secondary market. (CNBC)

‍