AI Video Enters the Revenue Race: Sora Exits, Kling Nears $500M ARR
FutureX Research · AI Lab · 2026.06.23 · 6 pp · preview 2 pp · Revised 2026.11.02
Listen · Audio Summary
5-8 min · AI narration in English · abstract + all key findings
Bottom Line
AI video now competes on revenue; as per-second prices fall, products wired into ad placement and performance data keep pricing power.
- Why now
- On October 6, 2026, Bloomberg reported that Kling is preparing a Hong Kong IPO of at least $1B; a listing would force it to disclose revenue, gross margin and losses under exchange rules.
- What it means
- Founders need verifiable revenue and per-second costs to raise; investors should price revenue quality and gross margin, not benchmark scores.
Abstract
AI video now competes on revenue, and the ability to close the commercial loop separates valuations. OpenAI has shut Sora down; Sora reportedly cost about $1M a day to run against roughly $2.1M in lifetime in-app purchases. Kling closed a ~$3B round at $18B post-money on July 2 and booked over RMB 850M in Q2 revenue; Runway reportedly reached $200M ARR. Shengshu's Vidu Q4 Preview starts at $0.014 per second, and unit prices keep falling. Kling is reportedly preparing a Hong Kong IPO of at least $1B, and listing disclosures would test the quality of its revenue. This is not investment advice. (Data current as of 2026-10-11)
Key Points
- 01Kling's revenue is rising and Kuaishou's profit is paying for it: Q2 results showed Kling revenue above RMB 850M, up 200%+ YoY, while Kuaishou's adjusted net profit fell 30.3% to RMB 3.9B (August 19-20, Verified).
- 02Capital pays a premium for revenue: Kuaishou disclosed on July 2 that Kling closed nearly $3B at an $18B post-money valuation, about 36x its ~$500M March ARR (Per company disclosure).
- 03OpenAI took five months to leave video generation: Sora's consumer app shut April 26 and its API on September 24 (Verified); Sora reportedly cost ~$1M a day to run against ~$2.1M in lifetime in-app purchases (Reported).
- 04Independent startups join the revenue race: Bloomberg reported Runway's ARR at $200M on September 8 (Reported); Higgsfield raised a $400M Series B at a $5.4B valuation on August 17 (Verified).
- 05Per-second prices keep falling: Shengshu launched Vidu Q4 Preview on October 7 at launch pricing from $0.014 per second, or about $0.14 for a 10-second clip (Per company disclosure).
Three sets of events turned the revenue race from inference into observable fact
Our core thesis holds: AI video is moving from a parameter race to a revenue race, and the ability to close the commercial loop is what separates valuations. In mid-to-late July, three sets of events turned that thesis from inference into observable fact.
First, Sora's exit. OpenAI shut Sora's consumer experience on April 26 and retired the API on September 24, with no replacement model named. Sora reportedly cost about $1M a day to run, against roughly $2.1M in lifetime in-app purchases. It is the first time a top player has conceded, by leaving, that a model race without a commercial loop cannot last.
Second, Kling's close. On July 2 Kuaishou announced that Kling had closed a round of nearly $3B at an $18B post-money valuation (Per company disclosure), about 36x its ~$500M March ARR. Capital still pays a premium for video models with revenue, but measured against the rumored $20B, pricing discipline is back.
Third, a squeeze from both ends. Earnings week pushed Big Tech's combined 2026 capex guidance to roughly $500B, while on July 27 Moonshot AI put the full 2.8T-parameter Kimi K3 weights on Hugging Face for free. Compute costs are rigid and rising; marginal model capability is free. The middle layer of creative tooling is caught between the two.
The report uses a neutral value-chain framework and is not investment advice. Developments since August are collected in the final section, Latest developments.
Through August 1: capital pays for revenue, compute costs keep rising, and AI-content labeling arrives on schedule
Data through August 1, 2026. Seven events in the period:
- July 2 · Kuaishou disclosed that Kling AI closed a ~$3B round at $15B pre-money and $18B post-money, led by CPE, with Tencent, Alibaba Cloud and Baidu in the same round. Kuaishou keeps 68.33%, employee incentives take 15% and outside investors 16.67%; the agreement requires an IPO by October 30, 2031.
- April 26 · OpenAI shut Sora's web and app experiences, with the API set to retire on September 24 and no replacement listed on the official deprecations page; costs reportedly far exceeded revenue.
- July 27 · Moonshot AI released the full Kimi K3 weights as promised: 2.8T parameters, 1M-token context, a ~1.56TB download, with day-0 hosting from Together AI and Modal. It is the largest open-weight model to date.
- July 22-29 · Earnings week: Alphabet raised its 2026 capex guidance and Microsoft's Azure topped $100B in fiscal-year revenue for the first time, while Meta's profit fell and its stock slid. Figures are in the table in the next section.
- July 10 · Apple sued OpenAI over alleged theft of hardware trade secrets; OpenAI denied the claims on July 14. The fight over hardware entry points now runs through both products and courts.
- June 16 / June 29 · The EU Parliament and Council passed the Digital Omnibus amendments to the AI Act, deferring high-risk obligations to December 2, 2027 and August 2, 2028. Article 50 transparency duties, including labels for AI-generated and deepfake content, still apply from August 2, 2026, alongside a new ban on tools that generate non-consensual intimate imagery.
- Mid-July · Anthropic reportedly began IPO investor meetings at a $965B valuation, listing as early as October; Zhipu was up about 13x in the six months after its Hong Kong listing.
The full report covers these across six chapters.
Six hard facts: capital still pays a premium for models with revenue, but pricing discipline is back
1. Kling's price shows both enthusiasm and discipline. Kuaishou's Q1 report showed Kling's Q1 revenue above RMB 650M, up more than 300% YoY (Per company disclosure). At $18B post-money, Kling is priced at ~36x its ~$500M March ARR: above Anthropic's ~20x public anchor, below the earlier rumored $20B.
2. OpenAI confirmed, by leaving, that a model race without a commercial loop cannot last. The Wall Street Journal reportedly put Sora's running cost at ~$1M a day against ~$2.1M in lifetime in-app purchases, and the Disney partnership reportedly ended about three months after it was announced.
3. Open weights cut production costs across creative tooling. Since July 27 the full 2.8T-parameter Kimi K3 weights have been free on Hugging Face, which opens up front-end and code generation capability.
4. Compute costs are rigid, and the market now separates spending with returns from spending with stories. The three companies' combined 2026 capex guidance is near $500B.
| Company | Disclosed in earnings week (July 22-29) |
|---|---|
| Alphabet | 2026 capex guidance raised to $195-205B; Q2 alone $44.9B |
| Microsoft | Quarterly capex plus finance leases $41B (+69%); Azure fiscal-year revenue above $100B for the first time |
| Meta | 2026 guidance $130-145B; net income down 14% YoY; stock down about 8% in a day |
xAI's $1B+ acquisition of APR Energy (1GW+ of mobile turbines) was confirmed in regulatory filings and cleared, writing energy into the value chain.
5. Synthetic-content labeling is now hard compliance for AI video in Europe. High-risk duties were deferred, but Article 50 transparency duties apply on schedule from August 2.
6. Public markets give private AI assets a visible anchor.
| Asset | Measure | Figure |
|---|---|---|
| Anthropic | Reported IPO valuation; public P/ARR anchor | $965B; ~20x |
| Zhipu | Gain since its January 8 Hong Kong listing; mid-July market cap; cumulative funds raised after the July 9 follow-on | ~13x; ~HK$731.2B; ~HK$36.4B |
| MiniMax | Market cap after its January 9 listing | Briefly above HK$410B |
Latest developments, newest first (data current as of 2026-10-11)
- 2026-10-09 · Kuaishou (01024.HK) closed at HK$30.10, up 5.47% on the day, for a market value of about HK$129.6B; it had closed at HK$28.54 on October 8 · Public market data (Tencent Securities)
- 2026-10-07 · Shengshu Technology launched Vidu Q4 Preview at launch pricing from $0.014 per second, or about $0.14 for a 10-second clip. It takes up to 15 image and 3 audio references, outputs up to 4K with 10-bit color, and is live on the web and via API. The company says the same budget yields up to 5x as much video under comparable specs and billing; the full Q4 follows the preview · Per company disclosure (PR Newswire)
- 2026-10-06 · Kling AI picked CICC, Goldman Sachs and UBS for a Hong Kong IPO raising at least $1B, with a listing as early as 2027; size and timing could still change. Goldman Sachs and UBS declined to comment; Kling and CICC did not respond · Reported (Bloomberg; relayed by TechNode on October 8; The Information reported the same day)
- 2026-10-06 · On the day of the Bloomberg report, Kuaishou rose 2.31% to HK$30.18 · Public market data (Tencent Securities)
- 2026-09-30 · Runway launched Runway Ads in its web app, an advertising video agent that recreates existing ads, generates variations, localizes them by market, publishes human-approved versions to Meta, Google and TikTok, and pulls performance data back; Lovable is the launch partner · Reported (Adweek)
- 2026-09-28 · Kling AI announced Kling 4.0. The Flash version is in closed beta for Ultra annual subscribers only, with 3- to 20-second clips at 720p in 8-bit SDR. The full model is expected in October with no date given, supporting clips up to 30 seconds, up to 4K, up to 10 keyframe references and 15 multimodal inputs · Reported (Tech Times, September 30; Bloomberg, September 28)
- 2026-09-25 · AdExchanger questioned Higgsfield's figure in a piece titled "Higgsfield's $1 Billion Run Rate Comes With An Asterisk" · Reported (AdExchanger)
- 2026-09-24 · The Sora API stopped serving requests as scheduled; OpenAI's help center posted a discontinuation page telling users to export their material. Counting from the April 26 consumer shutdown, OpenAI took five months to exit video generation · Verified (OpenAI Help Center; advance reports from NewsBytes and others on September 21)
- 2026-09-24 · Higgsfield said it expects $1B in sales over the next 12 months, a company-stated forward figure that is not audited · Reported (Bloomberg)
- 2026-09-17 · Pika released a tool that decides which underlying model generates each element of a video and turns basic user input into an editable shot list that cuts prompting work, plus apps for relighting and for swapping characters or outfits · Reported (Semafor)
- 2026-09-08 · Runway's ARR reached $200M, driven by enterprise customers; the Bloomberg story is paywalled, so the figure follows its headline and same-day accounts from Investing.com and PYMNTS · Reported (Bloomberg)
- 2026-09-08 · SMZDM's September pricing review: a 5-second 768P clip costs about 0.5 yuan on MiniMax H3, and a 5-second 720P-class clip costs 3 yuan on Alibaba Wan 3.0, a gap of about 6x. Wan 3.0 supports 30-second single-shot output, with pricing published at its August public test. Per-second prices are in the table below · Reported (SMZDM)
- 2026-09-03 · A MiniMax open platform announcement said H3 Max generates a 5-second 768P clip with audio in under 3 seconds, faster than playback; the third-party model page AI-bot.cn puts its throughput at about 35x that of H3 · Reported (SMZDM on September 8, citing the announcement)
- 2026-08-19 · Kuaishou Q2 results: Kling AI's quarterly revenue topped RMB 850M, up more than 200% YoY, about $470M annualized and in line with the ~$500M March ARR; global users passed 100M and enterprise customers neared 50,000. Kuaishou's adjusted net profit fell 30.3% to RMB 3.9B · Verified (National Business Daily, Caixin, August 19-20)
- 2026-08-17 · Higgsfield raised a $400M Series B at a $5.4B valuation, about 4x in eight months, with most revenue from enterprise customers · Verified (TechCrunch, ContentGrip)
Per-second API pricing for video generation
| Model | Resolution | Price per second | Evidence |
|---|---|---|---|
| MiniMax H3 | 768P / 2K | 0.1 / 0.8 yuan | Reported (SMZDM) |
| MiniMax H3 Max | 768P | About $0.04 ($2.40 per minute) | Reported (third-party page) |
| Alibaba Wan 3.0 | 480P / 720P / 1080P | 0.3 / 0.6 / 1.2 yuan | Reported (SMZDM, official pricing) |
| Shengshu Vidu Q4 Preview | Up to 4K | From $0.014 (launch promotion) | Per company disclosure |
What This Means for Founders
Investors now look for verifiable revenue: Higgsfield raised a $400M Series B at a $5.4B valuation in August, with most revenue from enterprise customers. As per-second prices fall, products that charge only for generation volume will see margins squeezed, so product comparisons should add generation latency and cost per second. Differentiation is moving from model quality to workflow: Pika orchestrates models, Kling 3.0 and Seedance 2.5 stress creative control, and Runway Ads pushes video into Meta, Google and TikTok campaigns and pulls performance data back. Signals to watch: whether full Kling 4.0 opens in October as planned and beyond Ultra annual subscribers, and whether Runway Ads reports verifiable revenue from customer campaigns.
What This Means for Investors
Capital still pays a premium for video models with revenue: Kling's $18B post-money is about 36x March ARR, above Anthropic's ~20x public anchor. The revenue is real, and so is the profit toll: Kuaishou's Q2 adjusted net profit fell 30.3%. If Kling proceeds with a Hong Kong listing, exchange rules will require it to disclose revenue, gross margin and losses, and that is the next test. The revenue race now includes independent startups, but company-stated figures such as Higgsfield's $1B run rate are unaudited and should wait for independent verification. This is not investment advice.
Where We Could Be Wrong
- 01If Kuaishou's next earnings report shows Kling's quarterly revenue below the RMB 850M of Q2, the claim that the revenue race has posted results does not hold.
- 02If Kling has not filed for a Hong Kong listing within 2026, its revenue and gross margin get no public test, and our view of its valuation needs to be revisited.
- 03If Shengshu's standard price after the promotion is clearly above the $0.014-per-second launch rate, the view that unit generation prices keep falling needs correcting.
Key Questions
Why did OpenAI shut down Sora, and when does it fully retire?
Sora's web and app experiences were shut down on April 26, 2026, and the API was retired on September 24, with no replacement listed on the official deprecations page. The Wall Street Journal reportedly put Sora's daily operating cost at ~$1M against only ~$2.1M in lifetime in-app purchases, and the Disney partnership reportedly ended about three months after announcement — the first top player to confirm, by exiting, that a model race without a commercial loop is unsustainable.
What is Kling AI's latest valuation, and how big is its ARR?
Kuaishou disclosed on July 2, 2026 that Kling AI completed a ~$3B raise at $15B pre / $18B post-money, led by CPE with Tencent, Alibaba Cloud and Baidu rarely investing side by side; Kuaishou retains 68.33% and the agreement requires an IPO by October 30, 2031. Kling's March ARR was near $500M, with Q1 revenue above RMB 650M (+300%+ YoY), implying ~36x P/ARR — above Anthropic's ~20x public anchor.
When do the EU's AI-content labeling obligations take effect, and what do they mean for AI video products?
Article 50 transparency obligations of the EU AI Act — including labeling of AI-generated and deepfake content — apply from August 2, 2026, with a new prohibition on non-consensual intimate imagery tools from December 2, 2026. While the Digital Omnibus amendments (passed Parliament June 16, Council June 29) defer high-risk obligations to December 2, 2027 / August 2, 2028, the marking duty got only a four-month grace (to December 2, 2026) for systems already on the market, making synthetic-content labeling hard compliance for AI video products entering Europe.
Watch & Listen
▶ The Next Gold Mine in AI Investing Is HereYouTube · needs VPN in China
▶ The Moat of AI-Native ContentYouTube · needs VPN in ChinaIn China: search WeChat Channels for 「倩姐投AI」; full library → Qian on AI
Sourcing and standards
Compiled from public sources; data current as of 2026.10.11. The text separates four evidence grades: verified, reported, per company disclosure and public market data, and states the derivation behind every estimate. When we get something wrong, the correction is written into the report body with the original call left visible, and logged publicly.
Research standards & corrections →📄 Full Report
Full report: 6 pages · provided to professional investors & partners only
This is the public preview. The full report includes the sections below. For compliance reasons it isn't posted publicly or offered as a free download. To request a copy, contact the FutureX team.
- 🔒1. The Model Race: From "Can Generate" to "Can Produce" — Open Weights Land in Full
- 🔒2. Capital Markets: Valuation Anchors Collide with Narrative Premium
- 🔒3. Platforms & Giants: After Sora's Exit, a Two-Sided Squeeze on Creative Workflows
- 🔒4. Distribution & Devices: The Hardware Entry Battle Goes Product-and-Courtroom
- 🔒5. Value Chain Breakdown: Which Links Benefit (A Neutral Framework)
- 🔒6. Risks & Watchpoints: August 2 on the Regulatory Calendar
Where we stand on this
Value lands in the application layer and at the open-source chokepoints; the model layer cannot hold it
This cycle's AI value lands in two places: ARR in the application layer, and the open-source chokepoints every model passes through. The model layer cannot hold it: large models have no network effects and no switching costs, so pricing power dissolves as capability converges. The bubble sits in the middle layer, which carries the most debt and lives on refinancing.
Full argument and falsification tests →Verifiability times feedback cycle decides the order in which AI remakes industries
The order in which AI remakes industries equals verifiability times feedback cycle. Where a result checks itself in seconds, AI moves first; where a human stays in the loop and feedback takes months, it moves last. Each sector gets a 12-to-18-month window after capability crosses its threshold and before enterprise budgets arrive.
Full argument and falsification tests →Questions people ask next
Related Research
Industry research from FutureX Capital's AI Lab, compiled from public information; not investment advice; contains no fund performance, AUM, or offer to raise capital.