Nobody saw it coming.
For months, an AI video model called "Happy Horse" had been dominating the global leaderboards for AI-generated video — beating OpenAI's Sora, Google's Veo 3, and every other model anyone could name. The AI research community had been puzzling over who built it. The model was stunning: cinematic quality, precise motion, photorealistic output.
Last week, Alibaba confirmed it was behind Happy Horse — formally known as Wan Video 2.5. The model had been submitted anonymously to benchmark sites. It ranked first globally before anyone knew it came from China's largest tech company.
That single revelation sent shockwaves through the AI industry. And if you run a business in New York City, it should matter to you too.
What Actually Happened
Alibaba's research team built a video generation model that, by almost every measurable standard, outperforms anything available from American AI companies. It generates video from text prompts, from images, from rough sketches — up to 60 seconds of high-resolution footage, with accurate physics, smooth motion, and the kind of production quality that used to require a production crew and a budget.
The "stealth" aspect wasn't a PR stunt — Alibaba researchers submitted the model to independent evaluation sites under a neutral name to get unbiased benchmark results. It won. Then they announced it.
This follows a pattern that's becoming impossible to ignore: Chinese AI companies are no longer playing catch-up. DeepSeek's R1 model shocked the industry in January. Alibaba's Qwen series has quietly become one of the most-used model families in the world. And now Wan Video 2.5 has topped the video generation category that OpenAI spent hundreds of millions trying to own.
The AI war has two fronts now. American companies are not pulling away from the pack — they're in a dead heat, and in some categories, losing.
Why This Matters for NYC Businesses
Here's the direct translation for anyone running a business in the five boroughs:
Video advertising is no longer the exclusive domain of companies with production budgets. It never was, exactly, but the gap between what a bootstrapped small business could produce and what a funded brand could afford has just collapsed to near-zero.
A restaurant owner in Astoria can now generate a 30-second video ad — real food, real motion, real production values — in under an hour, with no camera crew, no editor, no studio. A Brooklyn boutique can create seasonal content that looks like it came out of a Manhattan ad agency. A Queens-based contractor can produce before-and-after video testimonials without coordinating a shoot.
This isn't hypothetical. These tools are live, and several are free or nearly free for basic use.
The Tools You Can Use Right Now
Wan Video 2.5 (Alibaba) is available through Hugging Face and several API platforms. For most small business owners, the more accessible path is through tools that have already integrated competitive video AI:
Kling AI — one of the most polished video generators available, free tier included, developed by Kuaishou (Chinese video platform)
Hailuo AI / MiniMax — another Chinese-backed model with strong motion quality
Runway ML — US-based, still competitive, free tier available
Pika Labs — US-based, consumer-friendly, integrates with Canva
The common thread: within about 30 minutes of free account creation and some trial-and-error with prompts, a non-technical person can generate video content that would have cost $5,000–$50,000 to produce professionally in 2022.
The Platform That Makes This Actionable
Generating the video is step one. Getting it in front of customers is step two — and that's where most small business owners stall.
The standard assumption is that video ads require massive spend on platforms like YouTube or Meta. That's outdated. Connected TV (CTV) — meaning apps on Roku, Fire TV, Apple TV, and smart TVs — now allows small businesses to run video ads with budgets as low as a few hundred dollars a month, with geographic targeting precise enough to reach specific ZIP codes in New York City.
Roku Ads Manager, for instance, lets advertisers set campaigns at the neighborhood level, with full video ad placements in streaming apps that reach cord-cutters who haven't seen a cable commercial in years. The average Roku household watches roughly four hours of streaming content per day. You can reach Queens, Park Slope, or the Bronx specifically — not the entire tri-state area.
The math: if you can generate a production-quality video ad for free (or close to it), and run it on CTV for a few hundred dollars per month targeted to your actual service area, the cost-per-impression becomes competitive with flyers, direct mail, and social media boosted posts — with significantly higher engagement and recall.
Who Wins, Who Loses
Small businesses win — dramatically. The cost barrier to video advertising has effectively been eliminated. What remains is the time investment to learn the tools, which is hours, not weeks.
Freelance video producers lose (partially) — not entirely, because creative direction, scripting, and brand strategy still require human judgment. But the production execution layer is being automated. NYC's large freelance video community will need to move up the value chain.
Ad agencies lose the commodity work — the straightforward product demo, the social media clip, the basic testimonial spot. These will increasingly be done in-house by clients who couldn't afford agency fees anyway.
American AI companies face real competitive pressure — OpenAI shuttered Sora earlier this year and shifted focus. Google's Veo series is strong but expensive for API access. Chinese models are closing or leading in many categories, with more permissive free tiers, which means small businesses and developers will build habits around them by default.
The Data Point That Should Stop You in Your Tracks
TSMC — the Taiwanese company that manufactures chips for Apple, Nvidia, and essentially every major AI company — just reported a 35% jump in quarterly revenue, driven entirely by AI chip demand. That's not a growth number. That's an acceleration number. The infrastructure being built right now to power AI models like Happy Horse, Sora, and Gemini represents hundreds of billions of dollars in global capital deployment.
What does that mean practically? It means the AI video tools that seem impressive today are early versions. The trajectory is pointing at photorealistic, real-time, fully customizable video generation within 18 to 24 months. The businesses that learn to use these tools now will have a significant operational and marketing advantage over competitors who wait.
Three Things NYC Business Owners Should Do This Week
Sign up for one free AI video tool — Kling AI or Runway ML are the easiest starting points. Generate something. Anything. The learning curve is shorter than you think.
Audit your current video spend — if you're paying for video production, social media content creation, or agency-produced ads, price out what AI-assisted production would actually cost you. The gap will be eye-opening.
Look at CTV advertising options for your ZIP codes — connected TV is where your customers are watching, and the floor for testing it has dropped below what most people assume.
The tools exist. The cost is manageable. The competitive window — before every other small business on your block figures this out — is open, but it won't stay that way.
China beat the room on AI video. The room doesn't have to care who made the tool. It just has to use it.
The Metro Intel covers New York City business, real estate, and local life. Published Monday through Friday.