Evoken Hits $2B Valuation as AI Aggregator Model Faces Its Next Test
Evoken Technology, parent company of AI creative platform Liblib, has closed a Series B+ round of nearly US$300 million, co-led by Granite Asia, Tencent, and Shunwei Capital, pushing its post-money valuation past US$2 billion — yet the company's model-aggregation strategy is drawing scrutiny over whether subsidized growth can outlast the venture capital cycle.
The round, announced June 29, 2026, makes Evoken the second AI video generation company in China to cross the $300 million single-round threshold, following Aishi Technology. What distinguishes Evoken is precisely what makes investors nervous: unlike Aishi, Evoken develops no proprietary foundation models. Its entire product suite — Liblib AI, the design agent Lovart, and the AI video platform LibTV — is built on third-party model APIs, positioning the company as a high-velocity aggregator rather than a deep-tech bet.
The financing also brings Ant Group and Sequoia China onto the cap table, completing a blue-chip syndicate that signals institutional confidence in the near-term revenue story. Whether that story survives the next wave of model upgrades is a separate question entirely.
Revenue Metrics Justify the Multiple — For Now
Evoken's fundraising pitch rests on three data points that are difficult to dismiss. Total annualized recurring revenue (ARR) across the group has crossed US$300 million. Lovart, the overseas-facing AI design agent launched five months ago, has already reached US$80 million ARR on its own. LibTV, the AI video creation platform that went live in March 2026, reported monthly revenue growing more than 13-fold within its first two months of operation — an acceleration rate that few SaaS products achieve even in their best quarters.
Liblib AI, the original AI asset community launched in 2023, now claims more than 30 million registered users, making it one of China's largest AI creative communities by user base. The platform aggregates mainstream image and video generation models — including ByteDance's Seedance 2.0 via Volcano Engine — and presents them through a unified workflow interface aimed at designers, short-drama producers, and brand creative teams.
The founder profile reinforces the fundraising narrative. Chen Mian, born in 1992, served as global head of commercialization for Jianying and CapCut at ByteDance before founding Evoken, where he held a 4-1 executive designation — ByteDance's internal shorthand for senior leadership. Within two months of incorporation, the company secured angel funding from Source Code Capital, Gaorong Capital, and GSR Ventures, according to corporate registry data from Tianyancha. The Series B+ marks Evoken's sixth financing round in under three years.
Price War Mechanics Drive LibTV's Surge — and Expose Its Fragility
LibTV's explosive user acquisition is directly traceable to a competitor's pricing misstep. Jimeng, ByteDance's own AI video tool, revised its pricing three times in April 2026 alone, with video generation costs rising nearly sixfold. Creators who had been using Jimeng began migrating to LibTV, which accesses the same Seedance 2.0 model at substantially lower out-of-pocket cost — as little as RMB 20 (approximately $2.78) per minute of generated video at off-peak rates, versus rates on Jimeng that editors describe as difficult to match without restrictive usage conditions.
LibTV's annual subscription tiers range from RMB 569 to RMB 8,499 (approximately US$79 to US$1,180), undercutting comparable offerings on first-party platforms. Multiple production teams working in China's fast-growing AI short-drama segment — a market that accelerated sharply after Honguo Short Drama pivoted to AI-generated content following the 2026 Lunar New Year — confirmed they relocated workflows to LibTV primarily on price and queue-time grounds.
The commercial logic underneath this, however, is structurally precarious. Evoken does not own the compute infrastructure it resells. Every token consumed on Liblib or LibTV draws on model providers' data centers, with Evoken earning a margin on the spread between wholesale API pricing and retail subscription revenue. To sustain below-market consumer pricing, the company must either negotiate volume-commitment discounts with providers like Volcano Engine or absorb losses directly — a subsidy dynamic that mirrors the cash-burn playbook of China's mobile internet era, when platforms bought scale before profitability.
A hardware-focused venture investor, speaking on background, framed the structural risk bluntly: once large model providers open APIs more broadly or reduce direct pricing, or release native applications with superior user experience, aggregators like Evoken face direct substitution with limited recourse. Differentiated features built over six months can be rendered obsolete by a single model update.
Content Compliance Adds Regulatory Overhang
Evoken's growth trajectory carries a regulatory footnote that investors cannot ignore. In April 2026, state broadcaster CCTV Finance reported that LiblibAI could be manipulated through obfuscated prompts to generate non-compliant content, bypassing the platform's content moderation filters. Evoken issued a public apology and stated that technical fixes had been implemented.
For a platform operating in China's AI content space — where regulators have moved aggressively to enforce generation standards — a documented compliance failure creates reputational and licensing risk that is difficult to quantify but impossible to dismiss. Platforms prioritizing market-share capture over content governance have historically faced abrupt regulatory intervention in adjacent sectors.
The Aggregator Paradox: Infrastructure Ambition vs. Commodity Risk
The debate framing Evoken's long-term prospects is one that extends well beyond a single company. Anthropic's January 2026 launch of Cowork — a Claude-based tool covering legal, financial, and sales verticals — triggered multiple sell-off sessions in Nasdaq software stocks by demonstrating that foundation model providers can absorb application-layer functionality directly. Anthropic CEO Dario Amodei has argued publicly that as models mature, capabilities such as video generation and graphic design will be embedded in general-purpose base models, eliminating the commercial rationale for standalone third-party tools.
Nvidia CEO Jensen Huang has taken the opposing view, contending that AI agents will be built atop enterprise systems and structured data, and that software vendors face a transformation imperative rather than an extinction event.
The industry has not resolved this disagreement, and Evoken sits squarely at its fault line. Analysts who are skeptical of the aggregator model describe platforms like Liblib as "transitional products of the pre-convergence era" — viable while foundation models remain fragmented and user workflows remain unintegrated, but vulnerable once a dominant model provider closes that gap.
The more constructive read is that execution velocity is itself a defensible asset. LibTV achieving 13x monthly revenue growth in two months, during a window when short-drama production demand and competitor pricing errors aligned simultaneously, reflects an organizational capability that is genuinely scarce. The question Evoken's investors are ultimately pricing is whether that velocity can compound into durable infrastructure — a creator utility with switching costs deep enough to survive the moment Jimeng, or any other first-party platform, decides to compete on price again.
At a US$2 billion valuation on US$300 million ARR, the market is assigning Evoken a roughly 6.7x revenue multiple. That is a reasonable entry point if the company successfully transitions from model reseller to creator operating system. It is an expensive one if the window closes before the moat is built.