Disney highlights links between movies, parks, and tech

Disney’s SXSW session on “The Future of World-Building” lived up to its promise of special guests, with appearances from “The Mandalorian” creator Jon Favreau, Pixar chief creative officer Pete Docter, Marvel Entertainment president Kevin Feige, and “Iron Man” star Robert Downey Jr.

They appeared alongside the Imagineers who create the rides at Disney theme parks, using their stage time to highlight upcoming attractions drawn from Star Wars, Pixar, and Marvel movies and shows. (Sometimes, the inspiration can go in the other direction, with droids built for the parks also appearing in the upcoming film “The Mandalorian & Grogu.”)

Downey compared the Imagineers to his superhero character Tony Stark/Iron Man (a role he’ll be reprising in both “Avengers: Infinity Defense” and “Stark Flight Lab”), praising their “drive to put something good into the world” and make it “at a minimum, more fun.”

In a grandmaster-level chess move, OpenAI has signed a five-year, $11.9 billion agreement with the GPU-heavy cloud service provider CoreWeave, according to Reuters, which cites people close to the deal.

The deal involves OpenAI receiving $350 million worth of equity in CoreWeave, the sources told Reuters. The private placement is said to be separate from CoreWeave’s planned IPO.

CoreWeave filed to become a public company last week, but it has not yet priced or scheduled its debut.

It’s a win for both companies. One reason this agreement is so eye-popping (besides the billions involved) is that before this deal, CoreWeave’s biggest customer was Microsoft. In fact, in 2024, Microsoft accounted for 62% of CoreWeave’s revenue, which grew to a stunning $1.9 billion — nearly an eightfold increase from just $228.9 million in 2023.

Backed by Nvidia, which holds a 6% stake, CoreWeave runs an AI-specific cloud service with a network of 32 data centers that operated more than 250,000 Nvidia GPUs as of the end of 2024, according to the company. Since then, CoreWeave has added more GPUs, including Nvidia’s latest product, Blackwell, which supports AI reasoning, the company said.

Such dependence on one customer is usually worrisome for IPO investors, and could have added “hair” as they say, to CoreWeave’s hopes of raising $4 billion or more in its IPO. Landing OpenAI as a direct customer in a multi-billion-dollar deal should help CoreWeave appease investors.

Microsoft and OpenAI’s relationship

What makes this move equally interesting is that it’s another step in the deteriorating, frenemies relationship between Microsoft and OpenAI. 

It’s as if OpenAI CEO Sam Altman saw Microsoft’s usage of CoreWeave and said, “Hold my beer.”

Not only will OpenAI have access to the same cloud, but it will also have an ownership stake in the company that runs it.

Microsoft is, of course, a big backer of OpenAI in a deal that entitles Microsoft to collect a portion of OpenAI’s revenues. But tensions between two companies have been rising for years, as OpenAI’s fortunes have soared. OpenAI competes with Microsoft for enterprise customers and is even reportedly working on rolling out pricey AI agents. 

In January, as part of the massive Stargate AI infrastructure deal with SoftBank, Oracle, and others, Microsoft ceased being OpenAI’s sole cloud provider. OpenAI needs more compute resources. Just last week, Altman complained that OpenAI is “out of GPUs.”

For its part, Microsoft is working on its own AI “reasoning” models comparable to OpenAI’s o1 and o3-mini. It’s developing a whole family of its own models called MAI that are competitive with OpenAI. It also hired Altman’s rival, Mustafa Suleyman to lead Microsoft AI.

But CoreWeave is a surprising chess piece.

CoreWeave began its life as a crypto mining operation, founded by former hedge fund guys, it said. The three co-founders have already cashed out of $488 million worth of shares — over $150 million apiece. CoreWeave also has a stunning $7.9 billion of debt on the books.

If the IPO generates the billions of new capital they hope it will, the company says it will use at least some that to pay down the debt.

While these founders were once literally attempting to use GPUs to mint money, they are figuratively apparently accomplishing it. 

CoreWeave and OpenAI did not respond to our request for comment.

Poolside co-founder and CEO Jason Warner didn’t mince words: He thinks that most companies looking to build foundation AI models should instead focus on building applications. Poolside is an AI-powered software development platform.

Warner told the audience at the HumanX AI conference in Las Vegas on Monday that he thinks intelligence is the most important commodity in the world — on par with electricity — and anyone who doesn’t believe this should not be building a foundation model.

“If you’re one of those people, if you want to take one side of the fence, you’re a printing press for cash unlike anything we’ve ever seen in the world,” Warner said. “Or if the other side of the fence, you’re basically changing and bending the arc of humanity in a way that we’ve not done before. And I believe that to be true.”

Warner added that his company is “literally” going after AGI through software. If someone looks at foundational models as more of a “nice to have” as a way to raise VC cash, the company should just build a wrapper on an existing foundational model instead, he added.

With all that being said, however, Warner said he thinks that companies building foundation models can’t just have a foundation model as their product. Instead, that should be a part of their product — especially as the landscape gets more competitive.

“In my view, if I’m going to go build this type of business, I’m building on one side, going after intelligence on compute, I need to go after the hardest environment,” Warner said. “You can’t do simple on one side and hard on the other. It doesn’t really make sense, because if you’re going to go for everything, go for everything.”

He added that this is why Poolide is going after tough fields like defense and working with the government. But Warner said the company plans to launch a consumer application at some point, too.

San Francisco-based Poolside was founded in 2023 by Warner, the former CTO of GitHub and VC at managing director at Redpoint, and Eiso Kant, a serial founder. The company has raised more than $620 million in venture funding and is currently valued at $3 billion.

DeepSeek’s founder Liang Wenfeng is in no hurry to get investment from outsiders, the WSJ reported Monday.

DeepSeek is one of the hottest AI startups in the world right now after the Chinese AI company took Silicon Valley by storm with its latest model earlier this year.

Unlike DeepSeek’s AI model provider counterparts, who regularly announce mega-rounds filled with prominent investors, Liang hasn’t announced any fundraises, despite lots of VC interest. Rumors about its supposed investors have even fueled (baseless) rallies in some Chinese stocks.

DeepSeek’s founder doesn’t want to lose control 

An analysis of Chinese corporate records done by TechCrunch shows that DeepSeek is 84% owned by Liang. The rest of the startup is owned by individuals affiliated with Liang’s hedge fund, High-Flyer. 

That means that unlike most startups, which require outside capital and are thus used to at least some external influence, DeepSeek is basically a one-man show. And Liang doesn’t have the highest regard for VCs’ opinions.

When Liang was trying to raise capital in the past, he was put off by VCs’ focus on rapidly monetizing AI as opposed to fundamental research, he said in a 2023 interview with Chinese media. 

So one big reason why Liang hasn’t said yes to the investors pounding down his door is that he doesn’t want to share control of his company, the WSJ reported.

DeepSeek hasn’t required outside funding — yet

Most startups need capital from investors from the start. But DeepSeek is a unique beast. Liang has been able to fund DeepSeek through High-Flyer’s profits, reducing his need for outside investment.

“Money has never been the problem for us; bans on shipments of advanced chips are the problem,” Liang said in 2023.

Investors could deepen trust and privacy concerns

As a Chinese company, DeepSeek operates under strict Chinese laws that grant its government broad data access.

Concerns over this have prompted DeepSeek bans from a rising number of governments and even some private companies.

Those bans could get even worse if DeepSeek accepts funding from a Chinese investor, who face similar issues.

The U.S. government has a history of sanctioning Chinese tech companies it says are close to the Chinese government, like telecom giant Huawei and popular drone maker DJI. 

That hasn’t stopped some Chinese state entities from approaching DeepSeek for investment, The Information reported, although there’s no indication DeepSeek has accepted any.

Why this could all change

This doesn’t mean DeepSeek will never raise outside capital, though.

Earlier this month, DeepSeek announced a (largely theoretical) profit margin for the first time, signaling a shift toward monetization — something VCs value but that Liang previously dismissed.

To keep up with other AI heavyweights, DeepSeek will also likely need access to more and better AI chips — the biggest bottleneck on its development, Liang said in 2023. Those chips are expensive and heavily restricted in China due to U.S. export controls. 

DeepSeek’s ability to be self-funding may also be fading. While High-Flyer has done well in the past, some of its flagship funds have underperformed since 2022, the WSJ reported.

It also doesn’t help that the Chinese government has been cracking down on quant funds like High-Flyer since 2024.

While few concrete names are circulating, DeepSeek has already drawn interest from Tencent and Alibaba, according to multiple news reports.

DeepSeek didn’t immediately respond to a request for comment.

A nifty new Gmail capability powered by Google’s Gemini AI has arrived for Google Workspace customers. Starting Monday, users can add events to a Google Calendar directly from an email.

Gemini will automatically detect calendar-related content in an email and present an “Add to calendar” button. After clicking the button, the side panel in Gmail will open to confirm the event has been added to the calendar. 

Google notes in a blog post that the feature is only available in English and on the web for now. A calendar event created via the “Add to calendar” button won’t include other guests, and it also won’t appear for emails with already-extract events, like restaurant and flight reservations.

Users on Google Workspace Business and Enterprise tiers, as well as customers with a Gemini Education, Gemini Education Premium, or Google One AI Premium plan, are eligible for the new feature. (Users who previously purchased the now-deprecated Gemini Business or Gemini Enterprise add-ons are also eligible.) Admins can enable “Add to calendar” by switching on smart features and personalization from the Workspace Admin console.

“Add to calendar” is only the latest Gemini-powered tool to reach Gmail inboxes. In June 2024, Google added new capabilities to Gmail on the web to help users write emails and summarize email threads, plus ask questions and find specific information from emails within an inbox. Some of those capabilities came to the Gmail apps for iOS and Android toward the end of last year.

A few years ago, several startups with a specific focus on digital avatars appeared because of all the metaverse buzz. While that buzz died down, generative AI has given a new life to avatars as it is easier to spin up different virtual identities. Companies are trying out different use cases for avatars, including D-ID and Synthesia in the enterprise space, Zoom for meetings, Glance for fashion, Praktika for learning, and TikTok and Captions in the creator space.

However, Isaac Bratzel, who created popular virtual influencers such as Lil Miquela and Amelia 2.0, thinks there is a lack of high-quality avatars that not just look great but have personalized traits. And that thought process led him to build AvatarOS.

Bratzel previously worked in design roles at IPsoft (where he created Amelia 2.0), virtual influencer company Brud (where he created Lil Miquela), and Dapper Labs after the company acquired Brud. He started AvatarOS after he left Dapper Labs in 2022.

The company said it closed a seed funding round of $7 million led by M13’s Latif Peracha with participation from Andreessen Horowitz Games Fund, HF0, Valia Ventures, and Mento VC.

AvatarOS is in an exploratory phase to find the right product-market fit. Bratzel noted that the company is aware that customers don’t always want or need what you can do as a company, technologically, or what is cool.

As for M13, Peracha said that this is an exploratory round and the opportunity to back a founder who has a robust track record in the avatar space.

“We are going to look at the right business model through this round of exploration and have a bit more clarity on the way forward. We think that because of Isaac’s history in IPsoft to Brud, he is clearly the right person to build the business,” he said.

He also added that he did part of the due diligence by talking to an avatar of Bratzel to know more about the founder.

The founder said that AvatarOS is geared toward making high-end avatars in 3D space rather than catering to a world of click-to-generate content.

“One obvious parallel is spam emails. When it is easy to create content, it proliferates everywhere, and you want to have that differentiation from the saturation of content. That’s where we want to be in the avatar space,” Bratzel told TechCrunch over a call.

“While there are existing products that have tech for avatar generation, we want to focus on the avatar itself. If you look at Lil Miquela … That is a permanent entity beyond one single project and was able to accrue value over time,” he added.

The company is currently onboarding beta users and giving them access to a few existing avatars. The startup is also releasing a simple API that clients can use to integrate avatars with their sites. Bratzel said these organizations can power these avatars with large language models (LLMs) to provide info, and also change things like camera angles and views.

AvatarOS currently creates premium and customized avatars for clients themselves. But down the line, it wants to provide more tools for creation and adjustment to clients. Bratzel said the company’s main differentiation would be the way avatars move in their space.

“The main thing that is important to us is the humans move in a unique way. Pretty much every avatar solution can create something that might look like you but moves generically. Our view is that humans don’t move in the same way, and we want to recreate that,” he said.

The company will use the funding to grow its team and also build out a machine learning-based deformer that is responsible for creating lifelike movements in avatars.