The Argument in Favor of Tranched Rounds

Before we get to today’s column, the big day is here: Anthropic dropped its long-awaited public S-1 filing Monday morning. Tune in today for stories from my colleagues and me that will be breaking down the most important takeaways from the filing, such as improvements in the company’s compute efficiency, details on its spending and cash burn, and interesting new disclosures that could point to an eventual end to its unusual Long-Term Benefit Trust, a group of non-shareholders who advise the company on safety and social issues and can appoint a majority of its board.

Onto today’s column...

It seems like every AI startup financing these days has multiple tranches in which different investors get in at different valuations. At first, it was unusual to see just two tranches. Now we see extra-hot funding rounds with three tranches or more.

It’s no secret that investors aren’t thrilled about these multi-tranche rounds (specifically, the investors that can’t get in at the lower valuation). Some of them argue that the practice allows bigger venture capital firms to flex their power and force smaller VC firms into the second, more expensive tranche. They also say it allows the startup to claim a bigger valuation, even though some or most of the new capital came in at a lower price.

But founders and many investors tell me that the unique capital needs of AI companies, which are eternally starved of enough servers, is driving these unusual structures.

A young startup that wants to rent around 2,500 Nvidia chips to develop new models might need to raise anywhere between $150 million to $200 million in capital just to rent those graphics processing units from a cloud provider for two years, according to an investor who has backed multiple such companies. The startup then faces two unattractive options: raise capital at a lower valuation to attract institutional investors and heavily dilute existing shareholders, or raise at a $1 billion-plus valuation to minimize dilution but turn off investors that presumably want a bigger return on their investment.

Tranched rounds give founders the best of both worlds, as large corporate investors like Nvidia tend to pile into the higher-valued tranche, the investor said. In many cases, Nvidia will provide a significant chunk of capital in the higher-valued tranche, often more than 25% of the total tranche, they said.

Then the question becomes how the company can get an institutional investor to lead the higher-priced, less attractive round, as Nvidia doesn’t historically lead rounds itself. (That’s for several reasons, including that Nvidia often invests in competitors, doesn’t want to be seen as setting the price for the round and likely doesn’t want to invite regulatory scrutiny for investing in its customers.) In the hottest rounds, some VCs might not have a choice than to invest in or lead the more expensive round.

In other cases, more unusual things might happen. One investor told me their firm will sometimes put the vast majority of their investment into the lower-priced tranche, but then will invest a few million into the higher-priced tranche and “lead” it in name only.

That way, it gives corporate investors the cover to pour millions of dollars into the more expensive tranche without having to take on the responsibility of “leading” it and facing scrutiny if the startup goes south, the investor said.

Here’s what else is going on…

Big Number

DeepSeek has closed its second funding round, raising 106 billion yuan ($15.9 billion) in one of the biggest-ever fundraising deals in China’s AI sector, The Information reported.

OpenAI is aiming to close a $30 billion fundraising round that will value the AI company at up to $1.25 trillion, not including the new investment, before the end of next week.

Deals and Debuts

See The Information’s Generative AI Database for an exclusive list of private companies and their investors.

Huixi Intelligence, a Beijing chipmaker, closed combined Pre-B+ and Series B funding rounds of undisclosed amounts led by the Beijing Artificial Intelligence Industry Investment Fund, taking its 2026 equity financing past 2.5 billion yuan (about $375 million).

Pathos AI, which uses AI to develop oncology drugs, raised $237 million in a Series E funding round at a $3 billion post-money valuation, led by Bristol Myers Squibb.

Warp Sciences, which uses AI and robotics to speed up drug-candidate chemistry, emerged from stealth with $215 million in funding at an $865 million post-money valuation, led by Lightspeed Venture Partners.

Tractian, a 7-year-old physical AI startup based in Atlanta, Georgia, has raised $130 million at a valuation of $1.4 billion in a Series D funding round led by Sapphire Ventures.

Promaxo, which combines AI and robotics for prostate care technology, closed the initial tranche of a Series C funding round — amount undisclosed — with participation from New Science Ventures and Amara Ventures, alongside a separate $100 million loan arrangement with an undisclosed lender.

Formic, a robots-as-a-service platform for manufacturers, raised $50 million in financing, made up of a $40 million Series B led by Mercato Partners.

Cognitive Credit, which provides AI-powered credit-market data and analytics, raised $40 million in a strategic growth investment led by Summit Partners.

Higgsfield, an AI video and image generation platform, extended its Series B funding round with new investment from TCV and NVIDIA.

Enact, a Zurich robotics startup, emerged from stealth with $30 million in pre-seed funding led by Redalpine, Notion Capital and Firstminute Capital.

Maxinsights, which creates training data for robots using footage from cameras worn by contractors, raised $30 million in a seed funding round led by Peak XV.

Mave, which is developing an AI tool to automate security investigations, raised $28 million in a Series A funding round led by Red Dot Capital Partners.

Slingshot Insurance, an AI-native embedded-insurance platform, emerged from stealth with $27 million in seed funding led by Rally Ventures and Skyward Group.

FORT Robotics, which helps develop safety software for physical AI, secured approximately $20 million in new commitments from existing investors, led by Tiger Global with participation from Mark Cuban Companies.

Cuttable, an AI advertising platform, raised A$20 million (about $13 million) in a Series A funding round led by Airtree Ventures.

Modelyo, which lets companies run AI in the cloud without exposing their data, raised approximately $17 million in funding led by OlivTree.

Electric vehicles company Rivian acquired Third Dimension AI, a developer of AI-powered 3D spatial generation, to speed up synthetic data generation and full-world simulation for its driver-assistance and self-driving systems.

Albert Invent, an AI platform for materials R&D, secured a growth equity investment from Applied Ventures, the venture arm of Applied Materials, in an amount the company didn't disclose.

Hugging Face releases Carbon-X, an AI model that lets users ask questions about DNA sequences and their health and get a natural language response.

Moonshot AI on Friday announced its new model, Kimi K3.1 Pro, demonstrating performance that rivals that of top-tier U.S. frontier models on some benchmarks.

Walmart launched Agent Gateway, a connector that allows AI agents to browse Walmart’s website and make purchases. The launch comes as Walmart’s competitor, Amazon, has blocked AI agents.

Canva, a graphic design platform, launched Craft, its first AI model for generating images trained on human-designed templates within Canva.

OpenAI’s systems had multiple vulnerabilities that would allow AI agents to circumvent monitoring and take arbitrary unauthorized actions, evaluators from AI safety nonprofits METR and Redwood Research found during a three-week pilot with OpenAI, according to a report released Wednesday. OpenAI also made its Decisions API model, a classifier model meant to Jev, widely available.

This week on AI Deep Dive

Check out the most recent episode of AI Deep Dive. An up-close look into how frontier AI actually gets built — the models, the money, and the people behind them.

Thank you for reading the AI Agenda Newsletter! I’d love your feedback, ideas and tips: [email protected].

If you think someone else might enjoy this newsletter, please pass it forward or they can sign up here.