Lovable raises $400M at $13.3B, doubling valuation in eight months
Money moved on three fronts at once: Europe's hottest vibe-coding startup doubled its valuation, DeepSeek's long-rumored agent harness surfaced under a black-whale logo, and Mistral started charging a premium for EU data residency.
Lovable, the Stockholm-based startup whose AI app builder helped turn "vibe coding" into a mainstream workflow, has raised $400 million at a $13.3 billion valuation — roughly double the $6.6 billion it commanded in December, per a Wall Street Journal report. The round makes Lovable one of Europe's most valuable startups and the latest evidence that investor appetite for AI coding tools hasn't cooled. The company's pitch is simple: describe an app in plain language and the platform builds it, and that loop has made it a fixture of the current application layer. Doubling in about eight months, with no public revenue figures attached, is a statement of faith in the category as much as the company — the bet that building software by conversation becomes the default way products get made, with the market pricing the winners before their financials catch up.
DeepSeek's long-rumored agent harness is out of stealth: a "DeepSeek Harness" WeChat account sporting a black-whale logo went live, and Chinese media report internal testing is underway ahead of a launch expected alongside the V4 official release. The account, registered July 6, fronts a Harness effort led by Cui Tianyi — the former Jane Street quant who joined DeepSeek in March — that the lab has been hiring for since spring. Chinese outlets report that testers sign non-disclosure agreements, that early internal feedback is strong, and that Cui has said Harness ships together with V4's official release, which 36Kr reports would add native image reasoning, agent-focused upgrades and peak/off-peak API pricing. DeepSeek hasn't commented publicly, so the timeline is reported, not confirmed. But a first-party harness would put the lab directly in the agent wars against OpenAI's Codex and Anthropic's Claude Code — a sign that China's frontier labs now see the execution layer, not just the model, as the competitive battleground. We covered the V4 family's early benchmark run last week — Independent run confirms DeepSeek V4 Flash's 82.7% score.
Mistral is now selling EU data residency and a paid fast lane: regional inference endpoints are generally available and a Priority Tier is in public preview, both for a surcharge — with third-party open models joining the platform starting with Z.ai's GLM-5.2. Customers can pin requests to European or US endpoints at a 10 percent premium so processing stays in-region, or buy Priority Tier at 1.75x for a 99.5 percent uptime SLA and negotiated rate limits when traffic peaks. The limits matter for enterprises doing due diligence: only function calling works on regional endpoints — agents, batch and file APIs are excluded as "stateful" features — and account, billing and API-key data can still be processed outside the chosen region. That makes "sovereignty" a compute-level guarantee rather than a platform-level one, and it is a striking arrangement: Europe's flagship lab hosting a Chinese lab's open model under EU residency rules while it leans on the 1GW compute coalition we covered yesterday — Mistral rallies ASML and partners behind 1GW European AI compute.
What to watch: whether Lovable's new mark holds as AI-coding rivals raise behind it — and whether DeepSeek confirms a Harness launch date before the week is out.
Vibe-coding startups keep doubling their valuations — is the AI app-builder boom the frothiest corner of the market, or the real deal? Tell us in the comments.
