OpenAI eyes Disney deal, California defies Trump on AI rules, and Alibaba drops a multimodal powerhouse

Everyone's focused on the AI regulation battle in Washington. But the real action right now is happening in Hollywood, Sacramento, and a lab in Hangzhou - all on the same day.

Quick note before we dive in: if you ever want to dig into past coverage on any of these themes, the Daily Inference archive has you covered. Now, let's get into it.

Alibaba's Qwen team has been on a serious release streak. But which company originally popularized the term 'omnimodal' to describe AI models that natively handle text, audio, image, and video in a single architecture?

The answer is hiding near the bottom of today's newsletter... keep scrolling. 👇

This one caught a lot of people off guard. OpenAI is reportedly closing a deal worth around $1 billion with Disney - a partnership that would make the entertainment giant one of OpenAI's largest enterprise clients.

The details are still emerging, but the scale of the deal signals something important: OpenAI isn't just selling API access to developers anymore. It's going after transformational enterprise contracts with household-name brands.

With an IPO reportedly on the horizon - and serious questions about profitability (more on that below) - locking in a $1B relationship with Disney is exactly the kind of revenue story that makes investors pay attention. The entertainment and media sector is one of the last big industries that hasn't fully committed to AI infrastructure at this scale. This deal could change that calculus fast.

Read the full story →

While the federal government under Trump pushes to keep AI regulation as light as possible, California governor Gavin Newsom signed an executive order yesterday that goes in the opposite direction.

The order requires AI companies seeking state government contracts to meet new standards prioritizing public safety and civil rights. In practice, that means any AI vendor hoping to do business with California - one of the world's largest economies - now has to play by stricter rules than Washington requires.

This isn't just symbolic. California's market size means its regulations often become de facto national standards - companies typically can't afford to build separate products for California and everywhere else. Newsom has been careful to position himself as pro-innovation but anti-recklessness, and this order threads that needle by targeting procurement rather than trying to pass sweeping legislation.

The timing is deliberate. It drops just as the federal AI regulatory debate is heating up, giving California a stake in the ground before Washington decides anything.

Read the full story →

The Alibaba Qwen team just released Qwen3.5-Omni, and the key word here is 'native.' This isn't a text model with bolt-on vision or audio modules - it's an end-to-end omnimodal architecture that handles text, audio, video, and images in a single unified system.

The model is positioned as a direct competitor to Gemini 3.1 Pro and other frontier multimodal systems. Alibaba says it supports real-time interaction across all modalities, which puts it squarely in the territory of conversational AI assistants that can see, hear, and respond fluidly.

The distinction between 'native' and 'stitched-together' multimodal models matters more than it sounds. When separate encoders are bolted onto a text backbone, there are latency costs and information bottlenecks at the handoffs. A native omnimodal architecture eliminates those seams, which should translate into faster and more coherent responses when juggling mixed input types.

For developers building voice AI or video-aware applications, this is worth watching closely. Alibaba has a habit of releasing competitive models quickly - and making them accessible.

Read the full story →

Here's the uncomfortable truth sitting behind all of OpenAI's big announcements: the company is valued at $850 billion and reportedly planning a stock market float this year, but it still isn't turning a meaningful profit.

A new analysis from The Guardian lays out the tension clearly. OpenAI has been criticized for 'casting its net too wide' - pursuing too many product lines simultaneously, from video generation to search to consumer apps, without nailing down a core profitable business.

For a public offering to work at that valuation, OpenAI needs to show investors a credible path to profitability - not just revenue growth. That means strategic discipline around which bets to double down on (enterprise contracts like Disney) and which to cut (possibly why Sora got axed last week).

The irony is that OpenAI's biggest competitor right now might be its own ambition. The company that changed the world with ChatGPT now has to prove it can run like a real business, not just a research lab with a product team attached.

Read the full story →

Earlier this month, Microsoft launched Copilot Health, letting users connect their medical records and ask health questions directly. Amazon expanded its Health AI tool beyond its One Medical service to a broader audience. The healthcare AI space is suddenly very crowded.

MIT Technology Review digs into the critical question nobody wants to answer: are these tools actually safe and effective? The honest answer right now is 'it depends, and we don't fully know yet.'

Consumer health AI tools are proliferating faster than clinical validation studies can keep up. That creates a real risk: millions of people using AI to interpret symptoms, review lab results, or make decisions about medications, with limited oversight and uneven accuracy.

The tools may be genuinely useful in low-stakes scenarios like understanding a diagnosis or prepping questions for a doctor. But the line between helpful and harmful gets blurry fast when users start relying on them for things that need clinical judgment.

If you're building in this space, our healthcare innovation coverage is worth a browse for context on where the regulatory and clinical lines are being drawn.

Read the full story →

South Korean AI hardware startup Rebellions just closed a $400 million pre-IPO round at a $2.3 billion valuation. The company designs chips specifically for AI inference - the part of the AI pipeline where trained models actually respond to queries - rather than training.

That's a strategic bet worth noting. Training chips get all the headlines, but inference is where the volume is. As AI gets deployed at scale across enterprise and consumer applications, the demand for efficient, cost-effective inference hardware is enormous.

Rebellions is one of a growing list of semiconductor startups positioning itself as a Nvidia alternative, specifically targeting inference workloads where custom silicon can outperform general-purpose GPUs. With a public offering planned later this year, the company is clearly moving fast.

The broader chip competition story is one of the most consequential in AI right now. Whoever wins inference hardware wins the economics of deployed AI - and that's not a settled question yet.

Read the full story →

By the way - if you're working on a project and need a fast, professional web presence, check out 60sec.site - an AI-powered website builder that gets you from idea to live site in under a minute. Worth bookmarking. And for more daily AI coverage, dailyinference.com is where we keep everything.

The answer is OpenAI! OpenAI popularized the term 'omnimodal' when describing GPT-4o (the 'o' stands for omni) in 2024, positioning it as a model that natively processes text, audio, and vision in a single end-to-end architecture rather than through stitched-together components. Alibaba's Qwen3.5-Omni follows directly in that architectural tradition.

With AI health tools from Microsoft, Amazon, and others now reaching consumers - have you used an AI tool for anything health-related, and did you trust the output? Hit reply and let me know - I read every response and I'm genuinely curious whether people are actually relying on these things or just experimenting.

That's it for today - a lot moving across regulation, enterprise deals, and the hardware race all at once. See you tomorrow with more.