AI Startups

AI Startups to Watch (Sep 22): Harvey's Open-Weight Pivot & Spott

Harvey's gross margin fell to negative 50% before it moved to open-weight Kimi K3, plus Spott, BCI-Sonics, CellPoint, Benford and AI Hospitality raised.

Daniyal

· 5 min read
Editorial illustration of a rising bar chart built from glowing circuit lines beside an ascending rocket on a deep navy background

The most important startup story of September 22 was not a funding round. It was a $15.6 billion legal AI company admitting that renting frontier models had broken its unit economics, and fixing the problem with open weights trained in part outside the United States.

Below are the startups worth watching from September 21 and 22, starting with the one that changed how investors should think about every AI wrapper on their books.

The Story of the Day

Harvey went from 50% gross margin to minus 50% and swapped its model

Bloomberg reported that Harvey, the legal AI company valued at $15.6 billion, saw its gross margin collapse from about 50% to negative 50% in roughly six months as agentic usage drove inference costs up faster than revenue. The trigger was a March 2026 product update that pushed customers into far heavier usage. Source: Bloomberg

The fix is the interesting part. Harvey moved its flagship off frontier API models and onto Harvey Tenet, an in-house model post-trained with Fireworks on top of Kimi K3, the open-weight model from China's Moonshot AI. Harvey says open weights give it cheaper per-token economics and control over the stack. Margins turned positive again after Tenet launched in August 2026. Source: AI Weekly

Why this matters for founders: the margin problem Harvey hit is the margin problem every agentic application has, just at different scale. If your product's usage grows faster than your price, your gross margin is a countdown. Harvey's answer was to stop renting intelligence and start owning it, even when the base weights came from a competitor ecosystem. Treat the margin figures as Bloomberg's reporting, not audited numbers.

September 22 Rounds

Spott raises €18.3 million for an AI-native recruiting operating system

Belgium's Spott closed a €18.3 million Series A led by Balderton to build an AI-native recruiting operating system. The company deliberately keeps consequential candidate decisions in human hands, which is now a selling point in hiring software rather than an afterthought. Source: Tech Startups

BCI-Sonics banks RMB200 million for ultrasound brain interfaces

Shanghai's BCI-Sonics raised RMB200 million, roughly $29.9 million, in a pre-Series A co-led by HSG and Yunqi. The company builds a non-invasive brain-computer interface around ultrasound and neural decoding, with clinical work planned from early 2027. This is the hardest technical bet on the list, and it needs regulatory permission, clinical data and hardware to line up at once. Source: Tech Startups

CellPoint adds $34 million to put AI decisions inside travel payments

London's CellPoint raised another $34 million to push AI deeper into travel payments, where its Zenith engine makes routing and commercial decisions for airlines and hotels. Payments is a domain where a wrong answer creates a real financial loss, which is exactly the pattern of this funding cycle. Source: Tech Startups

Benford raises €5 million to become the auditor, not the vendor

Norway's Benford landed €5 million at pre-seed to operate as a licensed, technology-driven audit firm on its AuditOS platform. That is a structural bet: instead of selling automation to audit firms, Benford takes on the audit engagement and the liability that comes with it. Source: Tech Startups

AI Hospitality Group takes $7.5 million to run hotels, not sell dashboards

Dallas-based AI Hospitality Group raised $7.5 million to build a hotel-management company around autonomous back-office agents. It wants the management agreement and the profit and loss statement, not a software seat. Source: Tech Startups

Also funded in the same window: TacnIQ.ai, which is building tactile foundation models because robots cannot operate on vision alone, and OpenDebt, which raised about 22 crore rupees for AI agents that handle debt recovery conversations in India. Source: Tech Startups

Why the Money Is Moving This Way

Three patterns ran through the September 22 rounds, and none of them is about building a bigger model.

First, capital is paying for control points. The rounds that got funded sit where data access, deployment location, regulatory permission or transaction verification create a moat. The Wall Street Journal reported that one enterprise AI security vendor has raised $1.94 billion since June 2025 alone and has just landed its first customer spending more than $10 million a year, which is what a control point looks like in a bank account. Source: The Wall Street Journal

Second, the funding is shifting from software seats to service companies. Benford becomes the auditor. AI Hospitality Group becomes the hotel operator. OpenDebt talks to borrowers directly. The revenue pool is larger and the operational responsibility is much heavier.

Third, the cost of capital is real. The Fed raised rates by 25 basis points on September 16 to a 3.75% to 4.00% range, so a €5 million pre-seed and a RMB200 million clinical bet are not being funded by accident. Investors are reserving large checks for bottlenecks they believe cannot be removed. Source: Tech Startups

Quick Comparison

Startup Raise What It Actually Sells Source
Harvey Repriced model stack, $15.6B valuation Legal AI, now on in-house Tenet weights Bloomberg
Spott €18.3M Series A AI-native recruiting operating system Tech Startups
BCI-Sonics RMB200M pre-Series A Non-invasive ultrasound brain interfaces Tech Startups
CellPoint $34M AI decisioning for travel payments Tech Startups
Benford €5M pre-seed Licensed audit firm on AuditOS Tech Startups
AI Hospitality Group $7.5M Hotel management run by agents Tech Startups

Frequently Asked Questions

What are the top AI startups to watch today?

Harvey, Spott, BCI-Sonics, CellPoint, Benford and AI Hospitality Group are the AI startups to watch from September 22, 2026. Harvey leads the list for switching its flagship model to open-weight Kimi K3 weights after gross margins fell to negative 50%, while Spott raised €18.3 million and BCI-Sonics raised RMB200 million.

Why did Harvey's gross margin go negative?

Harvey's gross margin fell from about 50% to negative 50% as agentic usage pushed inference costs above what the company charged. A March 2026 product update increased customer usage sharply. After launching its in-house Tenet model on open-weight Kimi K3, Bloomberg reported that the margin turned positive again in August 2026.

What do the September 22 funding rounds have in common?

Most of the money went to companies that own a regulated workflow rather than a model. Spott keeps hiring decisions with humans, Benford holds an audit license, CellPoint decides payment routing, and AI Hospitality Group takes on hotel profit and loss. Investors are paying for accountability and access, not raw intelligence.

Bottom Line

Harvey's margin story is the one to remember from this week. It says out loud what most AI application companies are discovering quietly: if your product works well, usage grows, and if usage grows while you rent the model, your best customers can become your worst line items. The startups funded on September 22 are mostly selling the opposite of that. They own the workflow, the license or the liability. More AI startups and the rest of this week's coverage are on the blog. If you missed the next day's edition, it covered Snorkel, Cyera and Firecrawl.

Deeper context from the archive on the companies and technologies in this story:

Building an AI agent, automation, or backend workflow? AsaanKaroobar builds custom AI solutions, apps, and digital experiences. Get in touch →

More on AI Startups

Browse every AI Startups story →