AI funding rounds keep getting bigger as edtech and security bills come due

Lovable, the Swedish vibe coding startup, is closing a funding round that values it at $6.6 billion, CNBC reported on 16 December. Two sources with knowledge of the deal told the network that U.S. venture firm Accel participated, and that Khosla Ventures was also in the round.

That is more than triple the $1.8 billion valuation Lovable reached after its previous round in July. It is the company's third raise of 2025.

None of the parties had responded to CNBC's requests for comment when the article went live, and Accel's involvement had not been previously reported. Lovable was founded in 2023. In November it reported $200 million in annual recurring revenue, the company's filing of its own figures shows, less than a year after it first hit $1 million in ARR. It says 100,000 projects were being built on its platform every day at the time it published those numbers.

The numbers behind the hype

The company is based in Stockholm and is opening offices in Boston and San Francisco. Its platform uses models from providers including OpenAI and Anthropic to let users build apps and websites from text prompts, without writing code themselves. The July round brought in $200 million and included Creandum, Klarna founder Sebastian Siemiatkowski, ElevenLabs founder Mati Staniszewski and Synthesia founder Victor Riparbelli.

Comparisons in the same sector are not small either. Anysphere, which makes Cursor, raised $2.3 billion at a $29.3 billion valuation in November, according to CNBC.

Replit reached a $3 billion price tag in September after picking up $250 million, and Vercel closed a $300 million round at a $9.3 billion valuation.

Where the money stopped going

The picture looks very different one category over. Global edtech investment peaked at $16.7 billion in 2021 and had fallen to less than $3 billion by 2025, according to Tracxn, a Bengaluru platform that tracks startup funding. Rest of World reported the figures on 23 April 2026.

The pool of founders has shrunk with the money. Just 645 education companies launched in 2025, against almost 10,500 in 2020. High customer acquisition costs, long institutional sales cycles and low retention rates plagued the sector, according to an analysis by Loot Drop, a database of more than 1,700 startup closures.

Byju's, once the world's most valuable education startup at $22 billion, collapsed under a financial crisis, and Nigerian startup Edukoya shut down in 2025. HolonIQ, a research firm that advises governments and investors on education, wrote in a 6 February post that capital was concentrating in AI-enabled products, workforce-aligned platforms and K-12 operations tools that address cost and staffing pressures. Corporate workforce development and certification prep, the Loot Drop analysis concluded, look more promising than general K-12 education.

When the AI gets it wrong

There is a harder bill attached to the AI boom as well. MeetingTV sued Palo Alto Networks and its Koi Security threat-intelligence unit over a December 30 blog that linked the video conferencing startup to a Chinese corporate espionage operation, The Register reported on 2 July 2026.

The complaint alleges Koi used a large language model to generate the report, that the system hallucinated findings and that the company published them as facts. Koi's blog, since edited to remove references to MeetingTV's Zoomcorder product, called the recording service a public-facing front for a criminal operation. Security providers around the world then blocked MeetingTV's domains as malware and command-and-control infrastructure.

MeetingTV founder and CEO Michael Robertson told The Register that Koi never contacted the company, and that he learned of the report only when providers began blocking his traffic. A Palo Alto Networks spokesperson said the company was aware of the lawsuit and believed Koi's research reflected its commitment to identifying threats, adding that the dispute should be resolved through the legal process.

Robertson's argument is less about one blog than about process. He said the extension the report relied on, a supposed Twitter X Video Downloader, does not exist, and that without the software even basic analysis was impossible. His broader point: AI systems hallucinate, and their findings should not be treated as fact without human review.