Introducing Frontier Rounds: The Frontier Raises in Rounds

TL;DR

Frontier Rounds is a record of how frontier AI gets funded — not a news feed that forgets, but structured funding histories for the companies defining this era of AI, every round traced to a source. The name is the thesis: the frontier raises in rounds, and we record every one. The site launches with three connected sections, free to read.

A Funding Leaderboard of 26 AI companies sorted by cumulative capital — every entry graded Disclosed or Reported and linked to its source.
16 Product Histories: company-by-company funding timelines, round by round, with the story that connects them.
An Investor Atlas: 22 profiles of the funds and individuals financing the frontier, cross-linked to the companies they backed.
A published sourcing standard: three confidence tiers for every number, backed by a maintained source whitelist. If a number cannot be traced, it does not get published.

1. Why we built Frontier Rounds

Frontier Rounds exists because the largest financing wave in the history of technology is being recorded with tools built for gossip: fragments that surface, get shared, and sink without ever becoming a record. The money moving into AI companies is measured in the hundreds of billions of dollars, the decisions it funds will shape the field for decades, and the trail it leaves lives in press releases, paywalled scoops and social posts that decay almost as fast as they appear. We thought that deserved better, so we built it.

Start with how a funding event actually gets covered. A company announces a round, a dozen outlets publish within hours, and every one of those posts tells you the same narrow thing: the newest round. The amount, the valuation, maybe the lead investor, one quote about how the mission has never mattered more. What none of them tell you is how the company got here. Anthropic has raised roughly $118 billion across a sequence of rounds that includes an FTX-led round that later became a distress sale, dual cloud investments from Google and Amazon, and a Series H priced just under a trillion dollars — a genuinely dramatic story. The coverage of its latest round contains none of that context, and six months from now the post itself will be buried under the next one.

Databases were supposed to solve this. In practice they flatten it. A typical database entry gives you a table of rounds — dates, amounts, maybe investor names — with no narrative connecting them and, on free tiers, a paywall between you and half the cells. A funding history is not a table. It is a story with sequence and causality: why a lab's valuation multiplied sixteen-fold in fifteen months, why a Chinese lab went public before any of its Western rivals, why the largest seed round in European history went to a company with no product. Those questions have answers, but a row of cells cannot hold them.

Then there is the quietest failure: numbers without provenance. Aggregators copy aggregators, blogs copy aggregators, and a figure that began life as a rumor gets quoted enough times to look confirmed. Sometimes the number is right and nobody can show why. Sometimes it is wrong and has been circulating for years. In a market where a ten-billion-dollar difference in valuation changes how founders, policymakers and journalists read an entire sector, "probably fine" is not an acceptable standard for a number.

Who needs the record? Anyone whose work depends on reading the frontier correctly. Investors tracking a market where a single round can outsize another sector's entire year. Journalists who need the context behind an announcement rather than the announcement itself. Founders calibrating their own raise against the landscape. Analysts and policymakers trying to understand where capital believes the next breakthrough will come from. For all of them the raw material is the same: a complete, sourced, structured account of who raised what, from whom, on which terms. That is what did not exist, and that is what we set out to build.

I have spent years advising AI companies, close enough to this industry to watch the same handful of funding stories get retold — sometimes accurately, often not — while nobody built the structured record those stories deserved. Frontier Rounds is the record I kept wanting to cite and could not find. It is a site where the funding story of every significant AI company gets written once, written completely, and written with its sources attached.

2. The name

The name of this site is not decoration. It is the entire thesis in two words, and it is worth unpacking, because if you understand the name, you understand what we are building.

Frontier is the term of art for the leading edge of AI research. Frontier models are the most capable systems in the field — the ones that define what the rest of the industry builds on. Over the past few years the word has widened from a technical label into a map: frontier labs, frontier safety, frontier capabilities. When we say the frontier, we mean the moving edge of the field, and the small set of companies racing along it.

Rounds is the other half, and it is more precise than it looks. A funding round is the atomic unit of private finance: a discrete, datable, attributable event in which capital takes a position. Who led it, who followed, at what valuation, on what terms — each round is a vote, cast by people whose job is to price the future, on a specific company at a specific moment. One round is an opinion. A sequence of rounds is a record of conviction.

There is a third, quieter sense of the word we also mean. In finance, a round is an event; in geometry, a round is an orbit — something that closes, completes, accumulates. We like that the word carries both. The financing record is circular by nature: each new round circles back over the same companies, the same investors, the same questions, one layer further out. A good record does not merely note that each orbit happened. It keeps every lap.

Put them together and you get the thesis, which is also the tagline: the frontier raises in rounds, and we record every one. The logo is the same idea drawn instead of written — concentric arcs orbiting a core inside a black square, one ring deliberately left open, a single point already past the gap. The record accumulates outward, and the next round is always escaping the ring. That is the site in an image: not a feed, which forgets, and not a ledger, which flattens, but an orbit that keeps its history.

3. The Funding Leaderboard

The Funding Leaderboard is the snapshot layer of the site: where the frontier's capital stands right now. It lists 26 AI companies sorted by cumulative funding, and together they have raised more than $400 billion. The spread is the point. OpenAI leads with roughly $142 billion raised, including a single $122 billion round at an $852 billion valuation in March 2026 — the largest round we have on record, in any industry. Anthropic follows at about $118 billion, xAI at $37 billion, Databricks at $25 billion. The lower half of the board is as revealing as the top: Sarvam AI's $350 million building sovereign AI for India, or Groq's $4.26 billion bet that inference, not training, is where the money lands.

The board also refuses to be a single-country story. Four of the twenty-six entries are Chinese companies — Zhipu, Moonshot AI, MiniMax and Unitree — alongside France's Mistral and India's Sarvam, because the frontier's capital map is global whether or not the coverage of it is. The sectors range across foundation models, autonomous systems, coding agents, legal AI, robotics, voice and silicon, which is itself a finding: the financing of AI is no longer one financing story but many.

It is worth saying what the leaderboard is not. It is not a hot list, and it is not a ranking of the best AI companies — capital raised measures conviction, not quality, and a company can be heavily funded and still be wrong. Sorting by cumulative funding is simply the most honest ordering a snapshot can offer: it reflects every decision every investor has made to date, rather than our opinion of who will win. The leaderboard records the market's revealed position, not our predictions about where it moves next.

Each entry carries the numbers we would want if we were reading it: total raised, largest single round, latest valuation, the investors who led, and the date of the most recent close. And every entry is graded — twenty-one of the twenty-six are Disclosed, confirmed by the company or a filing; the remaining five are Reported, traced to established coverage. What the leaderboard deliberately does not do is tell you how any of these companies got here. That is the snapshot's honest limit, and it is the next section's entire reason to exist.

4. Product Histories

Product Histories are the narrative layer: sixteen company-by-company funding timelines, each an attempt at the definitive account of how a company financed its way to where it is. Every round on a history page carries its amount, valuation, date, lead and participating investors, and a link to its source. Between the rounds sits the story — the sections that explain what the numbers meant — because a timeline without interpretation is a spreadsheet that reads itself slowly. The full set, from OpenAI and Anthropic to World Labs and Higgsfield, lives at Product Histories.

The clearest way to show why the format matters is an example. Lovable, the Stockholm-born app builder, raised a $7.5 million pre-seed in October 2024. Twenty-two months later it closed a $400 million Series C at a $13.3 billion valuation, with roughly $953 million raised in total along the way. A news post tells you the endpoint; a table tells you the endpoints. Only a history lets you see the acceleration — pre-seed to seed to a $200 million Series A to $330 million to $400 million, each round arriving faster and pricing higher than the last — which is the actual phenomenon worth recording.

Or take Zhipu, the Beijing lab behind the GLM model family. Its record runs from a 40-million-renminbi angel cheque in 2019 through eight pre-IPO rounds backed by more than fifty shareholders — venture funds, strategic giants, local government vehicles — to a Hong Kong listing in January 2026 that made it the first publicly traded large-model company anywhere, and onward to post-IPO raises, including a two-part placement and convertible-bond package reported at roughly $5 billion in September 2026. There is no news post that can hold that story, and no database row that can explain it. A history can.

The histories at the top of the market carry the same weight. OpenAI's page runs from a 2015 founding promise through the $1 billion Microsoft investment of 2019 and the $10 billion of 2023, onward to the $122 billion round of March 2026 — a decade-long arc in which each round repriced the entire sector. Read as a sequence, it is a history of how the market's imagination about AI grew. Read as a table, it is ten numbers. We think the difference matters.

Histories also do something no single page of coverage can: they make patterns visible. Read enough of them side by side and the mega-seed era jumps off the screen. Mistral's €105 million seed in 2023 was startling at the time. Thinking Machines Lab raised $2 billion at seed in 2025, the largest first round on record. In 2026, Ineffable Intelligence and AMI Labs each raised about a billion dollars at seed. In three years the price of entry at the frontier inflated roughly twenty-fold — a fact about this market you can only see when the histories sit next to each other.

5. The Investor Atlas

Capital is a character in every one of these stories, and the Investor Atlas gives it a cast list: twenty-two profiles of the funds and individuals financing the frontier. Each profile covers who the investor is, what its thesis is, and — most importantly — which of the companies in our records it has backed, cross-linked so that a name inside a round entry resolves to a full record of that investor's behavior.

The Atlas deliberately spans the whole spectrum of AI capital. There are the AI-dedicated funds — Radical Ventures, Air Street Capital, Conviction, AIX Ventures, Gradient — firms that do nothing else. There are the generalist giants that made AI the center of their portfolios: Sequoia, Andreessen Horowitz, Lightspeed, Index Ventures, General Catalyst, Menlo Ventures. And there are the individuals — Elad Gil, Nat Friedman, Naval Ravikant, Eric Schmidt — whose personal checks move earlier and faster than any fund's process.

Why track investors separately at all? Because a round tells you what happened, while the investor tells you what it meant. When a fund leads a company's seed and then returns for every round that follows, that sequence — visible only across histories — is one of the strongest signals private markets offer. The Atlas and the histories together give you that view for free, which is roughly what a mid-tier database subscription promises and, in our experience, rarely delivers.

An example of the Atlas and the histories working together: Mistral AI. Lightspeed led its seed and did not miss a round after; Index Ventures was present from the first check as well; General Catalyst led the Series B; ASML, the Dutch lithography monopoly, later led a Series C that made it the company's largest shareholder, before Samsung Electronics led the most recent round. Each of those facts lives on the company's history page; each investor's wider pattern lives in the Atlas. Neither view alone tells you how deliberately the cap table of Europe's AI flagship was assembled.

6. How we source

A funding site's real product is trust, and trust is not a tone of voice — it is a sourcing policy, published and enforced. Ours has two parts: a three-tier confidence system for every number we publish, and a whitelist of the domains we accept as evidence. Both are applied with a bias we state plainly: a missing number is better than a wrong number.

The tiers work like this. Disclosed means the company itself announced the figure, or it appears in a filing — the strongest evidence there is. Reported means an established news organization — Reuters, Bloomberg, CNBC, TechCrunch and their peers — has stated it, which is strong but secondhand. Estimated means the figure was inferred, by us or by a third party, and where an estimate is the best available, we say so explicitly rather than let it masquerade as fact. On the leaderboard today, twenty-one of the twenty-six entries are Disclosed and five are Reported, and every entry carries its grade on its face.

The whitelist is the part that does the quiet work. We maintain a list of the domains we accept as sources — official company announcements first, established media second, a small set of research databases third — and a claim that cannot be traced to a whitelisted source does not get published. Not published with a caveat. Not published because everyone reports it. Left out. Aggregator sites that copy one another are exactly how a bad number becomes industry consensus, so aggregators are excluded on principle.

The policy also governs what we refuse to record. A round that is "in talks" or rumored is not a round; it enters our data only when it closes and can be sourced. Amounts originally disclosed in other currencies keep their disclosed terms, with any conversion marked as such. Where two credible figures conflict, we choose the more conservative one and say so. None of this is complicated — it is simply a refusal to let a record drift toward whatever number is most repeatable.

One more rule is worth naming, because almost nobody else follows it: we do not count ourselves as a source. A page on this site linking to another page on this site is navigation, not evidence. And the blog you are reading holds itself to the same standard — same tiers, same whitelist, same preference for an honest "we do not know yet" over a sourced-looking guess. Every number on this site is graded, and when it can be traced to a source we can stand behind, it links there, so you never have to take our word on faith. That, more than any feature, is the product.

7. What's next

This post is the first entry in a blog that will do for analysis what the archives already do for data. Three formats are planned. Funding Deep Dives will take a single company's financing history and give it the long-form treatment: the full narrative, the cap-table politics, what the rounds reveal about strategy. Quarterly AI Funding Reports will read the leaderboard as a trend surface — where the money concentrated, which rounds broke records, which investors accelerated. Round Recaps will cover the months in between, recording significant rounds as they close.

The cadence will follow the market rather than a content calendar; we would rather publish one complete thing than four timely ones. The blog runs on an open-source publishing engine we built, because the tooling for a site like this should be as inspectable as the data on it. What will not change is the discipline: every claim traceable, every number graded, every history written once and written completely.

Conclusion

Frontier Rounds started from a small frustration that would not go away: the most consequential financing in the history of technology, recorded nowhere in full. The site is our answer — a leaderboard for the snapshot, sixteen histories for the stories, an atlas for the capital, and a sourcing policy that treats every number as a claim until it has been traced. Everything is free, everything is linked, and everything is written to be read months and years from now, when the news posts have sunk and the context is what matters. The frontier raises in rounds. We record every one.

Frequently asked questions

Why should I trust Frontier Rounds?

Because every number we publish is graded and traceable, and the grading rules are published too. Each figure on the site carries a confidence tier — Disclosed, Reported or Estimated — and is drawn only from a maintained whitelist of official announcements and established media; where a source meets the bar, it is linked on the page. When we cannot trace a number to an acceptable source, we leave it out rather than publish a guess.

Where does the data come from?

From primary sources wherever possible: company announcements, filings and official posts first, reporting from established news organizations second, and a small set of research databases as a last resort. We maintain an explicit whitelist of acceptable source domains and exclude aggregator sites that republish one another's figures, because untraceable copying is how bad numbers become consensus.

How is this different from Crunchbase?

Crunchbase is a breadth play — a database that tries to cover every company in every industry, and largely does. Frontier Rounds is a depth play: we cover AI only, and for those companies we publish full funding histories with narrative, graded figures and source links, free of charge. If you need the widest coverage across all industries, a general database serves well; if you want the complete, readable financing record of a frontier AI company, that is what we built.

How often do you update?

As rounds close and can be verified. The recent-rounds feed on the homepage is updated as rounds are recorded, and histories are revised whenever new primary information surfaces. We do not publish on an artificial schedule; a number added too early is a number that may later have to be retracted.

What do Disclosed, Reported, and Estimated mean?

Disclosed means the figure comes from the company itself — an announcement or a filing. Reported means an established news organization has stated it. Estimated means the figure is an inference, ours or a third party's, and it is labeled as such. The tier appears alongside every number, so you always know how strong the evidence behind a figure is.