A useful watchlist explains what each company is trying to make possible and what evidence would show that it works. Funding is evidence of a financing event. It is not a ranking of product quality. The differences between a coding tool, a protein-design platform and defence technology are more informative than placing their round sizes in one league table.

How the ten were selected

Each entry has an identifiable company or investor source and a distinct product question. Seven names from the previous article remain. Three entries whose identities and performance claims could not be adequately substantiated—Div-idy, EcoFusion and NeuroSphere—have been replaced by Harvey, ElevenLabs and Helsing. This is an editorial selection, not a survey of every startup, and the order is not a score.

Some background rounds occurred in late 2024. They are dated explicitly rather than relabelled as 2025 funding. For later announcements, the date is the date of the source, not necessarily a cash settlement date. Company announcements can also contain promotional claims; the analysis below keeps those separate from the financing facts.

Databricks: established infrastructure, not an early-stage newcomer

Source and date: 17 December 2024. The company announced a $10 billion Series J, with $8.6 billion completed at that announcement and a $62 billion valuation. Its data and AI platform makes it relevant to a 2025 technology watchlist, but calling a Series J business a new startup conceals its maturity.

What to assess: The useful question is whether customers can turn their existing data into dependable production workloads. A financing headline does not answer migration cost, governance or the expense of running those workloads.

Mistral AI: model development with a strategic industrial investor

Source and date: 9 September 2025. Mistral announced a €1.7 billion Series C at an €11.7 billion post-money valuation, led by ASML. These are euro figures, not dollar amounts. The announcement connects the financing to research and industrial applications.

What to assess: Watch the fit between a particular model, deployment requirements and the task. The company name alone does not establish the licence or openness of every model; check the specific release rather than describing the entire catalogue as open-source.

Anysphere: Cursor and the cost of reliable coding assistance

Source and date: 6 June 2025. The team behind Cursor announced $900 million at a $9.9 billion valuation. Cursor is a coding product, not a general measure of how much autonomous software development has been solved.

What to assess: A practical evaluation includes the time spent reviewing generated changes and repairing mistakes. A tool that produces more code is valuable only when the resulting work meets the project’s requirements.

TinyFish: agents working through the live web

Source and date: August 2025, as identified in its company account. TinyFish identifies a $47 million Series A led by ICONIQ. Its subject is infrastructure for web agents: executing online workflows rather than merely returning a conversational answer.

What to assess: The relevant test is completion of an authorised workflow under changing interfaces, with failure reporting and access controls. Do not treat later customer or operation counts on the live company page as facts known when the 2025 round was announced.

Neysa: distinguish a round from cumulative funding

Source and date: 22 October 2024. The Mumbai company announced a $30 million Series A after a $20 million seed round earlier in 2024. The two figures add to $50 million; that does not make the Series A itself a $50 million round. Its Velocis platform provides access to AI computing infrastructure.

What to assess: Compute availability is only part of the customer decision. Data location, support, capacity commitments and the total cost of a workload matter alongside the advertised hardware.

Cradle: protein engineering still needs laboratory evidence

Source and date: 26 November 2024. Cradle announced a $73 million Series B led by IVP. The company develops AI-assisted protein engineering; this financing predates 2025 and is included as background to the watchlist, not as a 2025 transaction.

What to assess: A generated protein design and a protein validated for a particular use are different milestones. Look for relevant experimental measurements and the conditions under which they were obtained, rather than treating a financing round as biological validation.

CuspAI: materials discovery in the United Kingdom

Source and date: Investor portfolio record checked 27 October 2026. Investor Hoxton identifies CuspAI as a UK company founded in 2024 and describes a platform for requesting and evaluating materials with specified properties. The original article’s US label was incorrect. This entry does not assign a funding amount from an unverified announcement.

What to assess: The commercial bottleneck is whether a promising proposed structure can be made, tested and used economically. Simulation can narrow a search without eliminating those later steps.

Source and date: 12 February 2025. Harvey announced a $300 million Series D led by Sequoia at a $3 billion valuation. Its focus on legal and professional services makes it a distinct application business rather than another foundation-model laboratory.

What to assess: Source verification, confidentiality and review responsibility are central evaluation questions for legal work. A plausible answer is insufficient when the underlying authority or document does not support it.

ElevenLabs: audio production and control over voices

Source and date: 30 January 2025. ElevenLabs announced a $180 million Series C co-led by a16z and ICONIQ Growth. Its announcement covers speech generation, voice design, dubbing and other audio tools.

What to assess: Evaluate whether the output preserves meaning, pronunciation and appropriate permission to use a voice. Adoption claims do not settle those questions for a particular production or language.

Helsing: defence technology has a different route to adoption

Source and date: 17 June 2025. Helsing announced a €600 million Series D led by Prima Materia. The release explicitly said the financing was subject to certain approvals. The euro denomination and that condition belong with the headline amount.

What to assess: Defence procurement, operational requirements and authorisation make this a different business from a self-service software subscription. Consumer adoption metrics cannot simply be reused as the standard for judging it.

What this watchlist can tell you

The cases suggest three different evaluation routes. For software such as Cursor or Harvey, inspect the completed work and its review burden. For infrastructure such as Databricks or Neysa, compare a real workload under equivalent security and support requirements. For proteins, materials or defence systems, look for the relevant experimental or operational validation before extrapolating from a demonstration.

These are questions to investigate, not findings that every company has already met them. A buyer can ask for a bounded pilot with success criteria; a reader following the sector can look for published evidence against the same criteria. Neither needs to infer success from valuation alone.

Round size, valuation, cumulative funding and revenue describe different things. Valuation prices an ownership transaction; cumulative funding combines past events; a round is a particular event. Dollar and euro amounts should remain labelled unless an exchange rate and conversion date are supplied. We do not combine them into a misleading total.

What changed in this archive correction

The unsupported growth percentages, anonymous investor quotation and unverified company entries have been removed. Neysa’s $50 million cumulative figure is separated from its $30 million Series A, CuspAI’s country label is corrected, and each retained amount has a date. Later rounds and present-day valuations are outside this historical comparison. The original URL and publication history are retained.