The Tables Have Turned: Which AI Company Could Be the Biggest Capital-Markets Winner?

On 26 April 2026, OpenAI took Sora offline. The product that exploded globally in 2024 with a single video did not die because the technology failed. It died because the unit economics did not work. Media reports put daily compute costs at roughly USD 15 million, with revenue far short of covering them.

Sora’s shutdown is a watershed. It put the real question of the AI-video race on the table: once “generating a clip” is no longer scarce, the winner is whoever can turn generation into cash first.

At that watershed, global players split into three paths — big platforms distributing through existing traffic (ByteDance, Kuaishou, Alibaba); independents financing their way to an exit (Kling racing toward an IPO at an USD 18 billion post-money valuation, Runway pivoting to world models, Evoken’s LibTV monetizing at the application layer); and listed pioneers (MiniMax, Zhipu) now providing a public-market scoreboard for the sector.

The hand is now clear: the AI-video game has switched from “who has the stronger model” to “who first closes a commercial loop and who first sits down at the capital-markets table” — and Chinese players are taking the seat OpenAI vacated. 

EX.IO Research’s view: Sora’s death was the death of a business model, not of the technology. Burning about USD 15 million a day without enough revenue to cover it is an equation that does not close. No matter how strong the model, it gets cut. That draws a red line for the industry: a model without a commercial loop is just an expensive demo.

The Chinese players now picking up the baton are not winning on parameters. They are winning on cost structure and commercialization path. Kling has Kuaishou’s traffic. Seedance has Douyin / CapCut distribution. LibTV has orders from nearly a thousand short-drama teams. They win because generation can become revenue immediately — not because they win the leaderboard.

Capital has already voted with its feet: Kling’s USD 18 billion is a cut from USD 20 billion. That 10% haircut is the market’s discount on the “China takes over AI video” story. It both endorses the handoff and asks a harder question: can model advantage actually convert into subscription and API revenue?

One Race, Three Ways to Survive 

Sort the main global players by capital path and a clear split appears:

1. The Exit Camp: Too Expensive to Keep Burning 

• Sora (OpenAI): OpenAI took Sora’s web/app experience offline on 26 April 2026 and said the API would stop on 24 September. The official line was “unsustainable cost.” Media reports put daily compute at about USD 15 million, with revenue far from covering it. Sora’s exit is, in essence, OpenAI cutting a loss-making, copyright-contentious line before IPO and concentrating resources on a story that can be sold.

2. The Big-Tech Camp: Fold Video into Multimodal, Do Not Finance It Separately 

• Veo (Google DeepMind): Veo 3.1 (January 2026) leads with native audio-visual sync and was called the “strongest all-rounder” in independent testing (Zapier). It has been folded into Gemini Omni. Video is one slice of Google’s multimodal franchise, with no standalone capital process.

• Seedance (ByteDance / Doubao): Seedance 2.5 launched on 31 July — 30-second one-take generation, native 4K, up to 50 reference inputs, and timestamp-level local edits. ByteDance distributes inside the Douyin / CapCut / Jimeng ecosystem and likewise does not finance the model as a standalone company.

• Tongyi Wanxiang / Wan (Alibaba): Wan 2.7 sits inside Alibaba and is aligned with the e-commerce ecosystem.

• Grok Imagine Video (xAI): Under Elon Musk; Imagine Video 1.5 is an adjunct to the multimodal stack.

3. The Independent Camp: Raise Capital and Monetize via Listing 

• Runway (US): The Western independent leader. In February 2026 it closed a USD 315 million Series E at a USD 5.3 billion valuation (led by General Atlantic) and is shifting from video generation toward world models.
• Pika (US): A lighter creative path; Pika 2.5.
• Luma (US): Luma Ray 3.2.
• Vidu (Shengshu Technology, China): Closed an A+ round of more than RMB 600 million in February 2026. Vidu Q3 is positioned as “built for drama” and ranks first in China and second globally on the Artificial Analysis leaderboard.
• Kling (Kuaishou, China): See below. The player with the clearest capital path in the race — and the closest to an IPO.

Capital Expectations: Who Is Aligning with Whom? 

This is the part of the 2026 race most worth watching — valuation anchors are being recalibrated against one another at speed. 

1. Kling at USD 18 Billion: A Post-Money Price, and a Marked-Down One 

According to ChinaBizInsider, Kling completed about USD 3 billion of pre-IPO financing at a post-money valuation of roughly USD 18 billion (about RMB 124.2 billion) — about USD 2 billion below the original ~USD 20 billion target in April, or nearly 80% (about 78%) of parent Kuaishou’s then market cap. That has two implications:

• The bull case: Kling is one of the faster-commercializing Chinese text-to-video names. Spinning it out for a standalone raise already shows that Kuaishou wants an independent capital-markets price on the asset.

• The risk case: ByteDance’s Seedance 2.5 (30-second one-takes, 50 references) is a direct parameter-level squeeze, and ByteDance has Douyin / CapCut distribution. LibTV has also shown that application-layer revenue can scale faster than the underlying model. Whether Kling can hold an USD 18 billion valuation depends on converting “model advantage” into subscription / API revenue before it files.

2. MiniMax’s A+H Path: A Template for Still-Private Players 

After listing in Hong Kong in January 2026, MiniMax started STAR Market IPO tutoring on 29 May (CITIC Securities). Public reports point to an A+H dual-listing structure. One of the key risk factors in the prospectus is loss pressure from elevated compute costs. That is a reference point for everyone else: listing is not the finish line. Compute-cost control plus revenue conversion is what the secondary market prices.

3. Evoken / LibTV: Capital Is Rotating from Models to Applications 

Evoken (演语科技) closed a nearly USD 300 million B+ round in June (jointly led by Granite Asia, Tencent and Shunwei Capital, with HT Investment and Times Capital following, and Gaorong, Ant and other existing shareholders adding on), at a valuation above USD 2 billion and ARR above USD 300 million. Its product LibTV crossed USD 1 million in daily revenue in its first month and, by May, was more than 13× the first month, serving nearly a thousand short-drama teams. This is a signal that primary-market money is rotating: capital is moving from foundation models toward applications that can collect cash directly. 

4. Who Is Short of Cash — and Who Is Not? 

• Sora: Media reports of ~USD 15 million in daily compute costs and revenue far short of covering them; ultimately shut down.

• Runway: Closed Series E at a USD 5.3 billion valuation and is turning to world models for a second growth curve.

• Kling: A single ~USD 3 billion round — a record for a Chinese AI-video raise.

Market Timing Is the New Battlefield 

In sum, EX.IO Research believes the decisive variables in this race are no longer “the model,” but the commercial loop and capital timing. Whoever can balance compute cost against subscription / API revenue will get the better secondary-market price.

It is also clear that Chinese players are taking over the global video-generation vacuum left by Sora’s shutdown — and Hong Kong is the main financing channel for that takeover. Zhipu and MiniMax are already listed; Kling’s Hong Kong IPO progress is worth watching. Kling’s ~USD 18 billion post-money valuation is, in substance, a stress test against “nearly 80% of Kuaishou’s market cap.” It is both a valuation reference for Chinese video models and a pressure transmission to other players still in fundraising.

Another trend worth watching: the application layer (short-drama industrialization, as represented by LibTV) may reach a self-reinforcing loop faster than foundation models. That is a new variable in the “model company vs. application company” valuation debate.

Sources

1. OpenAI Help Center — Sora discontinuation timeline: https://help.openai.com/en/articles/20001152-what-to-know-about-the-sora-discontinuation 
2. Seedance 2.5 official announcement: https://seed.bytedance.com/en/blog/one-take-creation-flexible-referencing-introducing-seedance-2-5 
3. Kling ~USD 3B raise, USD 18B post-money, ~78% of parent market cap (ChinaBizInsider): https://chinabizinsider.com/kuaishous-kling-ai-secures-3b-mega-round-ahead-of-hong-kong-ipo/ 
4. Evoken B+ round of nearly USD 300M, ARR USD 300M (TMTPost): https://en.tmtpost.com/news/8033194 
5. Evoken / LibTV revenue details (Sohu / ChooseAI): https://www.sohu.com/a/1038245105_118792 
6. Runway USD 315M Series E at USD 5.3B valuation (TechCrunch): https://techcrunch.com/2026/02/10/ai-video-startup-runway-raises-315m-at-5-3b-valuation-eyes-more-capable-world-models/ 
7. MiniMax A-share IPO tutoring (East Money / CSRC filing): https://finance.eastmoney.com/a/202605303755066810.html 
8. Zhipu and MiniMax Hong Kong listings (Sina Finance): https://finance.sina.cn/2026-01-05/detail-inhffmxv8207565.d.html 
9. Sora shutdown and ~USD 15M daily compute cost (Neuronad): https://neuronad.com/kling-vs-sora/ 

Disclaimer

 

This article is for general market information and educational reference only. It does not constitute investment advice, legal advice, compliance advice, a trading recommendation, a solicitation or an endorsement. It is not an offer to issue, sell or subscribe for any securities or shares, and is not a public offer of securities in Hong Kong or any other jurisdiction.

Company valuations, financings and listing plans cited herein are drawn from public sources and may change with time and company disclosure. Data are as of the publication date and are not warranted to be accurate, complete or current. This article contains forward-looking statements based on currently available public information and market expectations; actual results may differ materially for many reasons.

The author and EX.IO do not represent that they do not hold, or will not establish, any interest in the companies discussed. Digital-asset and related technology-company prices are volatile; past data are not indicative of future performance. Readers should not make trading, product or business decisions solely on the basis of this article. This piece is addressed to a general audience and is not tailored to any particular investor’s financial circumstances. For legal, compliance, tax or investment judgments, consult a qualified professional adviser.