請同意我們的隱私權規範，才能啟用聽新聞的功能。

（中央社舊金山5日綜合外電報導）路透社引述知情人士報導，微軟針對與提供人工智慧（AI）雲端服務的英國資料中心業者Nscale簽署向其德州資料中心租更多算力的合約，將不會行使合約規模擴大1倍多的選擇權。

英國「金融時報」（Financial Times）昨天率先報導這項消息，並稱微軟（Microsoft）若行使這項選擇權，對Nscale而言價值數百億美元。

Nscale去年與微軟達成擴大合作協議的合約，由Nscale在其歐洲和美國資料中心部署約20萬顆輝達（Nvidia）的AI晶片。

根據消息人士說法，微軟早在數月前便決定不行使上述選擇權，比Nscale在9月遞交首次公開募股（IPO）申請時更早，但這不影響Nscale在IPO文件中揭露約1030億美元的合約總值。

Nscale在IPO文件中還說，與微軟在去年9月至今年4月間已簽署數份工作說明書（statements of work，SOW），總值最高達約438億美元，效期至2033年12月，不含擴大合約的選擇權。

這名知情人士補充道，Nscale對於其德州資料中心的其餘算力，已經正在和數家大客戶洽談中。（編譯：張正芊）1151105

Airbnb has teased its plans to build its own AI travel agent before, but on Thursday the company announced that the new AI feature will launch in the first half of 2027. In the company’s Q3 shareholder letter on Thursday, Airbnb also provided more detail about how the agent will work, what it will actually do, and how it’s different from a typical AI chatbot.

In prior interviews, CEO Brian Chesky sketched out the travel company’s approach to AI, describing a plan to build an AI agent that has a richer interface than a conventional chatbot. He also expressed his opinions about how chatbots aren’t the right interface for travel, noting that people take pleasure in the planning process and in anticipating a trip. Any agent that it built, then, shouldn’t take away from the inspiration and the browsing, he told TechCrunch in an interview last month.

Chesky today reiterated that belief, telling analysts on the earnings call that people won’t use one AI agent for everything.

“Today’s AI agents like Muse and Instinct are good at simple routine tasks like making a restaurant reservation or scheduling an appointment. But planning a trip is different,” he said. “Travel is aspirational. Discovering where to go and where to stay is part of the experience. In fact, 80% of travelers say they enjoy spending time researching trips. People want to look at photos. They want to compare homes. They want to make decisions with people they’re traveling with,” he continued. “And I don’t think the general-purpose agents are the best to do that. So that is why we are building our own AI agent,” Chesky said.

Airbnb described its agent as having a personalized interface that’s designed specifically for travel, where it will “build an interface” around a user’s question instead of simply responding as a chatbot would.

“What we are doing is developing a custom-designed AI-native experience that responds to what guests are looking for,” Chesky said. “So instead of relying on traditional chat alone, which is very text-heavy—which, again, is definitely not the way you should plan travel, our agent draws from a library of hundreds of travel-specific interface components—maps, photos, comparisons, calendars, itineraries—and composes them in real time to match what they’re looking for.”

The exec said the agent will feel “a little bit like generative UI” as it’s designed to offer up “the right interface at the right time” for someone’s specific travel needs.

In its shareholder letter, the company touted that its nearly 20 years of experience in bringing hosts and guests together is informing how the agent is built, as it incorporates insights about how people both book and host on its platform. The agent will also be able to help travelers beyond the booking itself.

“Our agent will be able to help guests throughout their trip—from changing a reservation or resolving a payment issue to getting help with refunds, rebooking, or 24-hour safety support. These are things an outside agent can’t easily do on its own,” the company wrote.

Over time, Airbnb expects the agent will be able to help travelers with other aspects of their trip, including not only homes and hotels, but also services and experiences. Services and experiences are two areas that have become a bigger focus for the company following its redesign last year.

While the company previously said to expect its AI agent sometime in the next year, it’s now more specifically committing to roll out the new technology to consumers in the first half of 2027.

The news of the agent’s release was delivered alongside strong third-quarter earnings, with revenue growing 19% to $4.87 billion, topping Wall Street estimates. It also raised its 2026 revenue growth forecast, which it now predicts will be in the high teens, instead of the mid-teens.

“Airbnb is back in growth mode,” Chesky told analysts on the Q3 earnings call on Thursday. “It’s rare for a marketplace of our size to reaccelerate growth, but that’s exactly what we’re doing. Q3 was our fastest revenue growth in more than three years,” he added.

Dropbox is making a big push into AI infrastructure with a $350 million acquisition of cloud technology company Tigris Data, marking what appears to be its largest acquisition to date.

Announced Thursday, the deal is expected to strengthen Dropbox’s AI infrastructure capabilities and support the development of DBX Agent Cloud, a new storage platform aimed at businesses and developers building AI models and agents.

Dropbox told TechCrunch via email that the transaction is expected to close in the fourth quarter of 2026. The company also said it currently has no plans to reduce headcount as part of the acquisition. The company already cut 9% of its global workforce last month, WSJ reported.

DBX Agent Cloud will initially offer object storage as a stand-alone product, allowing businesses to purchase storage independently and use it with existing tools and applications built for Amazon S3 (Simple Storage Service), which is a cloud storage service that enables businesses to store and retrieve large volumes of data.

Beyond its infrastructure ambitions, Dropbox is also bringing more AI capabilities to its core product. Users will be able to ask AI to perform supported tasks using their files, choose between different AI models, and connect their Dropbox content to third-party tools, including ChatGPT, Claude, and Muse.

The company also plans to expand DBX Agent Cloud with additional capabilities designed to help AI systems work with different types of content, including documents, images, audio, and video. Other planned additions include filesystem features that will allow AI agents to manage files and maintain continuity across tasks and sessions.

Dropbox says the updated experience has already reached more than 70,000 teams. The initial rollout is available to eligible Standard and Advanced customers in the U.S., U.K., and Australia, with broader availability planned for 2027.

Tigris Data was founded by former Uber engineers who helped develop the ride-hailing company’s storage infrastructure. The company provides cloud storage technology that helps businesses store and access data across different regions. Its platform automatically positions data closer to the users and computing resources that need it, helping reduce access times and improve performance.

Tigris Data raised $25 million in a Series A funding round last year.

While the Tigris deal appears to be Dropbox’s largest acquisition, it is not the company’s first move to expand its AI capabilities through M&A. Dropbox has previously acquired several AI-focused companies, including Reclaim.ai, Promoted.ai, and Mobius Labs.

Serve Robotics, a maker of self-driving sidewalk delivery robots, is expanding beyond its autonomous delivery operations in a bid to become a technology supplier to other companies building robots.

The company, which spun out of Uber’s acquisition of Postmates in 2021 and went public in 2024, said Thursday it will launch a new business line called Physical AI Services.

Specifically, Serve Robotics will offer customers its connectivity and data tools. Bedrock Robotics, a startup developing self-driving systems for construction equipment, is a customer, according to Serve. It’s also working with T-Mobile to bring AI computing power closer to robots, Serve announced alongside its third-quarter earnings on Thursday.

“We are solving the problem of robots in human spaces,” Serve Robotics CEO and co-founder Ali Kashani told TechCrunch.

Serve Robotics already makes two kinds of robots, the cute googly-eyed sidewalk delivery robot and one designed for hospitals that it got through its acquisition of Diligent Robotics.

Those businesses are not going away, Kashani emphasized in a recent interview. Instead, he said there is opportunity to sell the technology it developed for its own robots to other companies developing robots for farms, warehouses, and other uncontrolled environments.

“I kind of think of it as the AWS moment where we are taking the technology we’ve built for us, and we’re opening it up for more people to build on,” he said, adding this wouldn’t have been possible without developing its own robot fleet.

The expansion comes as Serve’s sidewalk delivery business expands, but is shifting away from Uber and toward other partners, including DoorDash and Wonder.

Its new Physical AI Services business line will offer three products. Serve Link helps connect robots in the field to AI systems running in data centers, which would allow companies to potentially tap into more powerful AI models without having to put all that computing power onboard.

“Any robotic application generally needs to connect the robots to some offline intelligence, whether it’s to collect data or do exception handling or whatever,” said Kashani. “It’s the connectivity piece that has to be really high performance. You need to be able to manage really unreliable networks and still reliably connect, and that’s what we solve.”

It’s also launching a data curator tool that helps companies find edge cases in the reams of real-world data a robot collects, which can then be fed back into a training model to improve performance. Its third product is a real-world, foundation-model data catalog built from Serve’s own sidewalk robots that can teach AI models how to understand complex human environments.

Kashani said he’s been thinking about this idea — a playbook he learned from Amazon — for years.

“When the technology is so nascent, you have got to build the end-to-end application because that’s the only way you know what to build, but then over time, as the technology gets more advanced, a lot more people are going to be wanting to build similar things,” he said. “And what you want to do is to open your technology so others build with you and on your platform.”

Muse Voice, a feature that lets users call and video chat with Meta’s consumer AI agent, Muse, is now available to everyone in the U.S., the company announced on Thursday.

First introduced at Meta’s Connect developer conference in September, the video-chatting feature involves new AI models, Muse Realtime Avatar and Muse Realtime Voice technology, which help power the digital avatar that chats with the user.

While other AI competitors offer voice chat features, Meta’s Muse offers a cute, animated avatar (nicknamed Jolly), which users interact with as they ask questions or request tasks to be handled on their behalf. Users can also customize the avatar as they choose, changing it to something that looks more like a person, animal, or some other character.

The company believes that anthropomorphizing Muse in this way could make consumers feel more connected to their AI agent — a connection that could grow even stronger as they begin to engage in live conversations with it.

As the Muse avatar speaks, Meta says it will use “subtle expressions and gestures” that match the tone of the conversation. (Muse doesn’t actually “see” the user, but it hears them and reacts based on their conversations.)

The news of the rollout was announced via blog post and on X by Meta’s chief AI officer, Alexandr Wang, who shared a demo of his Muse in action. The video saw Wang joking around, referencing his love of posting memes on X, which he told his Muse may have gotten him in trouble with his boss. His Muse, a fluffy white dog, responds with a suggestion to add disclaimers to those memes that could be misinterpreted as Meta endorsing anything risqué. (Ha!)

The added support for video chats and calls comes just before Meta’s plan to launch a Tamagotchi-like wearable device for its Muse AI agent before year-end. When the device arrives, the digital avatar will appear on its small screen, letting users talk to the agent, which can then reply or carry out tasks for them.

Meta says the real-time conversations are supported on mobile, web, and Mac and will later come to its AI Glasses and the upcoming “Muse Charm” hardware device.

Meta president Dina Powell McCormick said that the Democratic wins in the midterm elections are going to make it a lot harder for companies to build data centers, adding that the industry has “not done as good a job as we needed to explain the benefits of these investments.”

Speaking at the WSJ Tech conference on Wednesday, the Meta exec pointed out that the backlash against data centers — which was very much on the ballot during this election season — was also tied to people’s feelings about AI technology overall.

“The visible sort of manifestation of AI is a data center, and it’s something that people can say, ‘I don’t want this in my community,’” she explained. “I think there’s an underlying bigger issue, which is how people feel about AI in general.”

Meta believes it could turn around the negative consumer sentiment around AI if it’s able to demonstrate the value, not only of the data centers coming to local communities, but also in the technology itself.

“The stakes of who builds the technology that cures cancer, that gives access to education, that is also a very fundamental difference … Mark [Zuckerberg, Meta CEO] is very ‘long human,’ and he’s betting on humans more than machines,” Powell McCormick explained. “And that’s what we mean when we say the future is for everyone. That people should not be left behind. But the way to ensure that is that this technology is democratized, that people have access to it in a very fair and transparent way,” she said.

The exec also touted Meta’s experience in building data centers, saying that the company has been doing so for 20 years, as it takes a lot of compute power to offer the products it does in the social networking space. Today, the company has data centers in 23 states, she said, and five of them, including one in Texas, are planned as projects of 1 gigawatt or more.

But Meta is now worried that it could lose some of its investments after politicians ran on anti-data-center platforms during the midterms.

“I think it’s going to be harder,” Powell McCormick said, describing what the future of building data centers looks like. “It has been hard, I should say, but I also think that it’s becoming clear what models work, and the states want those models that work.”

Specifically, communities only want data centers when it benefits them in some way, she explained, laying out the approach that Meta today uses to close its data center deals in local communities.

Successful projects see Meta paying for its own electricity, directing any tax surpluses as a result of the data centers toward teachers and first responders, keeping the labor force local, and getting bipartisan support for the project, among other things.

Going forward, she expects data center projects to come under more review and be weighed against whether the constituents actually want them.

Meta has pinned its hopes on its AI investments, particularly its consumer AI agent Muse scaling far beyond the 10 million users it has today, to become a business that one day serves billions. But even if that’s not the case, it believes that its investments in data centers will not be for nothing, as the compute itself will be valuable.

“We’re getting calls every day to lease that compute [we’ve built], and so one way or another, we’re seeing a path to monetization,” Powell McCormick said.

Max Justicz, the chief technology officer and co-founder of the startup Charge Robotics, used an OpenAI model to find a severe vulnerability that — if exploited — could allow hackers to take control of servers running Ubuntu, a widely used open source Linux-based operating system.

Justicz, who has a hobbyist’s interest in cybersecurity and previously found a vulnerability in the same codebase, told TechCrunch that he found this issue using GPT 5.6 Sol, released in July, through the AI maker’s Daybreak Blue, a program that gives cybersecurity defenders access to powerful frontier models.

“I literally just aimed ChatGPT at the same body of code and said ‘I found this bug previously that was fixed several years ago, please find bugs as bad or worse,’ and it came back in two hours with this,” Justicz said. “It’s extremely convoluted and hard for a human to find, but it works!”

He reported the issue to the Debian Project, which maintains the software in question and issued a patch to fix the vulnerability. There has been no known evidence of exploitation. Still, the ease with which he found it suggests that we’ve only seen the tip of the iceberg when it comes to potential exploits that even novice hackers can utilize with the help of AI models.

“I can only imagine what one could do with more time and access to less neutered models. We should expect some extremely crazy bugs to be found and exploited over the next few months,” Justicz wrote in a social media thread discussing his finding. “This might be obvious to most folks here, but there is currently huge alpha in taking some code or binary someone has not audited yet and just aiming a powerful AI model at it.”

OpenAI appears to have noticed. Today, the AI giant announced a new model focused on cybersecurity, ChatGPT Cyber, would be made available to the highest tier of users in the cybersecurity program, Daybreak Red, which allows offensive cybersecurity researchers tools for vulnerability hunting and development. Yesterday, OpenAI said that the same model had been used to scan and prepare patches for vulnerabilities in the Linux kernel, finding 948 potential issues. OpenAI also found 81 “novel” vulnerabilities in Firefox, the web browser.

“Our goal is to dramatically expand Daybreak, bringing advanced security research to organizations of every size, independent researchers, and open-source maintainers,” the frontier lab said in a statement.

OpenAI has insisted that there is a narrow window of time for the people who safeguard digital infrastructure to take action before AI models become even better at hacking and easier to access. For cyber defenders, however, it can feel like the frontier labs have unleashed a mountain of new work that outstrips their resources. Justicz said that he personally spent about $5,000 on subscriptions and tokens to find several exploits.

Besides Daybreak’s model access, OpenAI has also announced a $1 billion initiative to support cybersecurity professionals in the program who work on critical infrastructure like the electrical grid or water systems.

The Daybreak program has its own critiques. Justicz notes that the frontier lab appears to have recently “nerfed” Daybreak Blue to make it less useful at finding vulnerabilities. On one hand, it may be good to limit access from potential bad actors — Justicz said he just uploaded a photo of his driver’s license to join the program. On the other, cybersecurity professionals say they have one hand tied if they can’t use frontier technology while bad actors can deploy open-weight models with abliterated guardrails.

Still, OpenAI says just “thousands” of cybersecurity professionals are participating in the Daybreak program, versus a far larger number of potential hackers.

It’s unclear if frontier labs are devoting enough of their resources to this issue. Eric Wallace, a researcher on OpenAI’s technical staff, said that “we are now allocating a significant amount of our frontier compute towards discovering and patching vulnerabilities in open source codebases, fully pro bono.”

But the Linux bug hunt cost $1.2 million, which is significantly less than the approximate $6 million token spend the company invested in finding a potential solution to a famous problem of fluid dynamics.

If there is a narrow window of time to patch the digital infrastructure the world depends on, more urgency may be required.

The U.S. government is warning of an “ongoing cyber threat” facing industrial refrigeration units, after reports of cyberattacks targeting freezers and fridges in supermarkets around the world, U.S. military bases, and at least one blood bank.

U.S. cybersecurity agency CISA, the FBI, and several other federal government departments warned in a joint statement on Thursday that hackers are targeting and compromising internet-connected refrigeration and cold chain monitoring systems in grocery stores and cold storage facilities. The statement said that the hacks are hitting critical infrastructure, such as the food, agriculture, transportation, and medicine-shipping industries.

“If successfully exploited, the actors could change the temperature and possibly cause physical impacts such as fire hazards, loss of inventory and changes to the service life of the system,” the agencies warn.

The warning comes after several waves of cyberattacks targeting industrial refrigeration units across the United Kingdom and Europe in recent weeks. Two separate waves of attacks allowed hackers to crank up the temperature on freezers and chilling units in at least a dozen countries, spoiling defrosted food and destroying produce. Some of the affected systems included remote-controlled refrigeration units made by Danish company Danfoss, which had stopped providing security updates to its customers’ freezers.

Hackers also targeted a blood bank in France, causing its refrigerated units to boost temperatures in its cold units containing donated blood supplies — but were stopped before the attack could compromise blood stores. The U.S. Department of Defense is investigating a spate of hacks targeting freezers run by the military, according to Reuters. The hacks were first reported by Signal and Silence.

The U.S. has not attributed the hacks to a particular group or government, but security researchers have broadly attributed the attacks to Iran; U.S. officials also reportedly suspect Iran is the likely culprit.

CISA, the FBI, and the Departments of Agriculture, Health and Human Services, and Transportation urged affected industries to remove remote control systems from the internet, change and strengthen passwords, apply patches where possible, and retire equipment that’s no longer supported by the vendor.

In a separate fact sheet, the departments said that hackers may attempt to “force units into continuous heating and defrost cycles, which can cause widespread inventory loss and equipment hazards.”

“Every organization that depends on refrigeration and monitoring systems should assume it could be a target,” said CISA’s acting cybersecurity chief Chris Butera.

OpenAI co-founder and president Greg Brockman said the company would be ready for an IPO in 2027, noting that its enterprise revenue has been growing. Speaking at a Wall Street Journal tech conference on Wednesday, the executive touted how the company had been gaining ground in the enterprise market, where rival Anthropic had already successfully made inroads.

In response to the interviewer’s question about being IPO-ready next year, Brockman replied, “I think so. Yes.”

“We’ve been very clear this year that we want to go public when we feel is the right time for us, and I feel extremely good about the business. I feel like we’ve made so much progress this year, and I’ve honestly never felt more confident in what we’re doing,” Brockman said.

He detailed how OpenAI’s trajectory as an enterprise had changed over the course of 2026, noting that the first and second quarters were “pretty rough.” He also admitted that OpenAI “wasn’t even in the conversation” when it came to agentic coding — an area where Anthropic’s Claude Code has gained market share.

“We had Codex, but no one was talking about it,” Brockman continued, “and I think that we just kept our heads down and executed.”

By the third quarter, activity picked up, as OpenAI saw 70% quarterly growth in its annualized revenue — something Brockman attributed to the enterprise side of the business doubling. He explained that the growth wasn’t simply the result of something that happened in the quarter itself, but rather the culmination of investments the company had been making up to that point.

Now the majority of OpenAI’s revenue is coming from the enterprise business, which previously accounted for only around 30% of its total revenue.

“And it’s not that consumer [revenue has shrunk] … It’s been growing, but it’s the enterprise segment has really taken off,” he said, pointing to a recent partnership with JPMorgan Chase as one example of the type of client it’s now winning deals with.

Brockman also spoke more broadly about the lessons OpenAI learned from selling into the enterprise market as well. He said the company had come to understand that its job was to help businesses identify the most important milestones they could achieve with its AI technology, build trust with clients, and demonstrate that it was a reliable partner for the long haul.

“A year ago, we weren’t even really in the enterprise game, right? We were really perceived as a consumer company, and a lot of the companies we’d start talking to would say, ‘Is OpenAI really committed to the enterprise?’” Brockman said. “But it’s so different today.”

Every time venture capitalists write a check, they’re secretly running numbers in their heads about different exit scenarios. It’s their job as fiduciaries. But nearly all will tell you (publicly, anyway) that every portfolio company is headed for an IPO.

Stephen Ward, the co-founder of Brightmind Partners, a two-year-old cybersecurity fund in New York, says otherwise outright. “You have to sell on the way up,” he says.

Ward’s unvarnished pragmatism is one reason he stands out. Another is the size of his firm’s debut fund late last year: a whopping $250 million fund, which he closed with co-founder Gur Talpaz. A third reason is his resume, which includes a stint with the Coast Guard, nine years in the Secret Service, and CISO jobs at TIAA and Home Depot before he ever thought about investing.

One proof point that suggests the combination is working is SGNL, the identity security startup CrowdStrike agreed in January to buy for $740 million, 11 months after Brightmind led its Series A.

Now, Ward is raising a second fund, targeting roughly the same size as the first one. This time, though, he’ll be doing it without Talpaz, who left in April along with another early colleague, leaving Ward as the firm’s sole general partner. He’s betting on himself at a moment when early-stage cybersecurity startups are raising more money than later-stage rivals for the first time in more than a decade.

His unemotional approach to exits traces back to an unlikely career. Ward says he grew up in northwest New Jersey between two farms, in a family that couldn’t afford college for his older brothers. After high school, he joined the Coast Guard, inspired by a commercial of a man jumping out of a helicopter, though he says he barely knew how to swim at the time. He spent four years in search and rescue.

Afterward, he studied criminal justice at Seton Hall in New Jersey and thought half-heartedly about law school. “I didn’t know what I wanted to do,” Ward told me last week over Zoom, wearing a pale hooded sweatshirt and looking more like a basketball coach than a venture capitalist. “I was kind of lost. I missed the military and the camaraderie and all of that.”

After graduating in 1998, he says, he mailed out some 300 resumes but no one responded. A Seton Hall professor who was also a chaplain to Secret Service personnel, as well as to members of the FBI and CIA, began introducing him around, and the Secret Service called first.

The Secret Service has two primary missions: protecting people and investigating financial crimes. Ward gravitated to investigations, then to “cyber” after the September 11 attacks in 2001. That year, he married a former college classmate, an American-born Syrian Muslim. Colleagues warned him the background check would be “miserable.” (They’re still happily married.)

What came next took Ward’s life down an unexpected path. Stationed in New York, he arrested a hacker who’d pulled off the largest credit card data theft of that era. Instead of charging him, however, Ward flipped him, turning him into an informant and running him for about three years. But the hacker kept hacking behind Ward’s back, and after he was caught again, he told the agency, as a diversionary tactic, that Ward was in on it.

By that time, Ward had left for the private sector, and he ultimately spent roughly 18 months clearing his name. “It was hard to get out of bed,” he said of that period, describing the investigation as “almost like a spouse cheating on you,” given his love of the agency. The hacker ultimately received a 20-year sentence, and Audible bought the rights to the story, turning it into an eight-episode podcast called “Operation Firewall.”

Ward’s corporate career has been a success by most measures, though he describes it as a series of jumps he really wasn’t qualified to make. His first private-sector job at J.P. Morgan, where he was recruited in 2007 to build a cybersecurity unit, was the hardest transition of his career, he says. He felt like “the dumbest person in the room,” missed his Secret Service crew, and twice thought about going back, unaware he was still under investigation. He stuck it out and now calls the investment bank the “best management school.”

Six years as CISO at the financial services company TIAA followed, and then two years at Home Depot starting in 2019. But in 2021, feeling burned out, he quit without anything lined up for the first time since age 16.

That chapter didn’t last long. The founders of Wiz, whose security software Ward had used at Home Depot, put him on a short list for a job at their company. Jeff Horing, a venture capitalist at Insight Partners, which had co-led Wiz’s $100 million Series A round in late 2020, interviewed Ward for the role, then offered him one at Insight instead.

It’s a good-news, bad-news story, given that Wiz was acquired last year by Google for $32 billion. When I asked him about it last week, Ward observed that the friend who took the Wiz job “made life-changing money,” but said it doesn’t bother him.

Regardless, he spent just three years at Insight, where he says he worked on Wiz, Abnormal Security, and Elisity, which IonQ agreed in October to buy for up to $1.3 billion. Then, at age 49, he thought about striking out on his own for the first time.

He says people told him it was a young man’s game. He left anyway and began cold-emailing limited partners. Ultimately, fewer than 20 LPs filled the fund, including Horing.

Ward started Brightmind in 2024 with Talpaz, a CrowdStrike corporate development executive whose deals overlapped with Ward’s during his time at Insight. Tayler Sipperly, another CrowdStrike corporate development alum, was the firm’s third employee.

The partnership ended this past spring. Asked what happened, Ward, using carefully worded legalese, cited “differences in opinions regarding operations and the future of Brightmind.” In August, Talpaz and Sipperly co-founded a new venture firm with Elia Zaitsev, CrowdStrike’s global CTO for the past three years. Called Cognition, it’s now reportedly targeting a $170 million first fund to back cybersecurity startups.

Brightmind’s first fund has made 14 investments to date, with one exit: that SNGL sale.

Some investors asked why Brightmind didn’t ride out that bet even longer. But Ward says he’s seen enough across his career that he feels confident in warning founders when they’re “going to race past the turn.”

In cybersecurity, he says, the best sales come between a startup’s third and seventh years. Indeed, he says his team tracks every security acquisition and what buyers are building versus buying. If Palo Alto Networks is likely to buy something, he says, CrowdStrike will experience “FOMO, then Zscaler is going to look over and go, ‘What did they just do? I should probably do the same thing.’”

Understanding how the pieces fit together is the trick, Ward explained, offering some hypotheticals. If CrowdStrike is building a product, he said, it’s unlikely to buy a portfolio company that makes something similar. Palo Alto Networks might love a company, but not if buying it would force its customers to switch away from something it bought five years ago.

“It’s very easy to go, ‘Good match,’ ‘terrible match,’” he said. “And that’s really what we spend a ton of time figuring out.”

He has timing on his side, at least. Early-stage cybersecurity startups are out-raising their later-stage rivals for the first time in more than a decade, according to PitchBook, which counts $4.7 billion worth of seed-through-Series B deals through September.

That total is already ahead of the entirety of 2025, even though the number of deals fell to 248 from 358 last year. Seed rounds in particular are ballooning, per PitchBook. The average cybersecurity seed round jumped to $8.4 million in the first nine months of this year, from $5.54 million a year earlier.

Ward questions how much of that money is really going to AI security. By PitchBook’s count, 278 deals in the first half of 2026 were bucketed as such, but his team’s manual review found only 30 true ones. He doesn’t think Anthropic or OpenAI will “take over the security world,” but instead will do “just enough.”

For real-time protection, which he terms “zero second,” he favors cyber-specific models that run locally on a customer’s own systems, like that of Empirical Security. He also looks for startups that could trounce established companies that he considers vulnerable, name-checking 27-year-old Qualys and 24-year-old Tenable.

Ward says Brightmind’s second fund won’t be entirely cybersecurity, focusing about 20% on AI infrastructure deals. He expects a meaningful number of teams to come from Israel, too. (Roughly 40% of the debut fund’s startups are based in Tel Aviv.)

Asked about the founders he’s most comfortable backing, the response is immediate: second-time founders who stick together.

Brightmind is comprised of five people right now: COO Iva Messy, also a Seton Hall graduate, came from venture firms Felicis and Andreessen Horowitz (a16z) on the recommendation of Ward’s cousin, Carl Eschenbach, a former Workday CEO who is now back at Sequoia. Sven Wollschlaeger, a Midwesterner whom Ward calls one of the hardest-working people he’s known, spent three years alongside Ward at Insight, and is a principal at Brightmind.

CTO Jamie Mneimneh, whom Ward first hired at Home Depot as a distinguished engineer, is the most technical person of the team. He built the cloud computing environment on AWS where the firm tests startups’ products. Mneimneh also built the AI agents that now source most of Brightmind’s deal flow, which is how the firm avoids hiring a bench of analysts and keeps management fees down, says Ward.

Taylor Cleveland, Ward’s chief of staff at Home Depot, joined the firm a few months ago.

For all the career changes, Ward says the real force driving him hasn’t changed. He grew up without money and still worries it could happen again, saying he’s being paid to produce a “3x net fund,” and he’s “not going to celebrate until we get that.”

Said Ward: “I don’t know what’s wrong with me. I have a level of being very naive to like, not being qualified to do something. I’ve just always looked at people and been like, ‘Okay, you know more than me today, but tomorrow, it’s going to be a little less, and the day after that, it’s going to be less, and eventually, I’m going to catch you and surpass you, and I’m going to know more than you.’ That’s how competitive I am.”

Anthropic announced Thursday that Claude is getting a new tool called “Personal Finance,” designed to help users keep track of their money. With it, users can connect their bank and credit card accounts through Plaid and ask Claude questions about their finances. Anthropic says access is read-only, meaning Personal Finance can’t move, manage, or invest users’ money.

By letting Claude connect directly to bank and credit card accounts, Anthropic is positioning its chatbot as a financial companion that can help people understand their spending and make sense of their money, rather than simply offering them generic advice.

With the launch, Claude is playing catch-up to OpenAI’s ChatGPT, which introduced similar personal finance tools in May that also let users connect their financial accounts and ask questions about everything from spending patterns to long-term financial planning.

“Claude gives you answers in everyday language, from your balances and transactions, and explains any term that’s new to you,” Anthropic wrote in a blog post. “Like the rest of Claude, Personal Finance has no ads or sponsored recommendations, and it doesn’t offer credit cards, loans, or savings accounts of its own, so your money questions get a straight answer.”

Image Credits:Anthropic

For instance, at the end of the month, a user could ask Claude, “Where did my money go this month?” The chatbot will then add up spending across the connected accounts and sort it by category, such as rent and utilities, food and drink, and shopping. Users can choose to go deeper by following up with questions like, “Is that more than usual?” or “Anything else I’m missing?”

Anthropic says the tool goes beyond basic spending summaries by analyzing transactions across connected accounts to flag potential issues, such as duplicate charges, unexpected fees, and steadily increasing bills. Users can also ask broader questions about their finances, such as whether they’re spending more than they earn. The tool can also explain financial concepts like APR (annual percentage rate), while providing calculations that show users how much they’ve paid in interest over the past year.

Of course, some people will have reservations about connecting their financial accounts to an AI chatbot. Anthropic says Claude’s access is limited to each connected bank’s name, account name and type, the last four digits of the account number, balances, credit limits, and up to two years of transaction history.

Additionally, the company notes that when users chat with Claude in the Personal Finance hub, those conversations aren’t used to train its models.

Personal Finance starts rolling out in beta this week to users on the Pro, Max, and Ultra plans in the U.S.

The launch of the new tool comes as Anthropic has been expanding Claude into new verticals. For instance, Anthropic launched Claude for Legal and Claude for Financial Services earlier this year.

In Brief

Electric truck startup Harbinger has raised a $240 million Series D funding round as it expands its portfolio of products, including battery storage and defense applications.

The funding comes almost one year to the day after the Los Angeles-based company announced a $160 million Series C, which included an initial truck order from FedEx. Harbinger has now raised more than $650 million to date, according to Bloomberg News, which first reported the round.

The Series D was co-led by Tiger Global, Capricorn’s “technology impact fund,” and defense-focused firm Ridgeline. FedEx invested again after leading last year’s round, and other repeat investors included Maniv, Greycroft, and Leitmotif, a VC firm backed by Volkswagen. Also investing this time was institutional giant Fidelity Management, which tends to invest in pre-IPO startups. (Harbinger is reportedly weighing going public.)

While it didn’t lead this round, FedEx’s relationship with Harbinger continues to deepen. The shipping giant’s senior vice president of safety and transportation, Paul Melander, has joined Harbinger’s board of directors. FedEx also recently placed its first large order with Harbinger — a deal for 2,000 trucks valued at more than $300 million.

Founded in 2022 by former employees of now-defunct EV startup Canoo and battery company QuantumScape, Harbinger started by focusing on making one electric truck chassis. The company has steadily expanded its efforts since. It bought autonomy startup Phantom AI, has begun developing hybrid emergency vehicles, and revealed a driverless hybrid electric platform for the military.

Food delivery startup Wonder is moving closer to becoming a publicly traded company after confidentially filing for an initial public offering (IPO). The company announced the filing on Thursday, marking a major milestone following years of fundraising, acquisitions, and expansion.

Wonder has yet to reveal how many shares it plans to make available or the expected price range.

Founded in 2018 by entrepreneur Marc Lore, Wonder has evolved from a food delivery concept into a broader food technology business. Its model goes beyond traditional delivery apps, with the company operating food hall-style locations where customers can order meals from multiple restaurant brands through a single platform. The company handles everything from meal preparation to delivery, pickup, and dine-in service.

Lore, who previously founded Jet.com and Diapers.com, has made no secret of his plans to take Wonder public. Those ambitions gained further attention in July, when the company announced it had raised more than $650 million in a funding round that valued it at around $9 billion. Investors include Accel, GV, AllianceBernstein, and ARK Invest.

Meanwhile, Wonder has been busy expanding its footprint and strengthening its business through acquisitions. The company has outlined ambitious growth plans, including a major push into Texas, where it aims to open more than 100 locations by the end of 2027.

Additionally, Wonder has acquired meal-kit company Blue Apron, restaurant delivery platform Grubhub, media company Tastemade, and kitchen robotics company Spyce. The latter acquisition gives Wonder access to automation technology that could help streamline food preparation and improve how food is served.

Wonder is also branching out with Wonder Create, its platform focused on letting food entrepreneurs and social media influencers develop new food concepts, restaurant brands, and menu items. The platform leverages AI to design and launch restaurant brands in less than a minute.

Another notable move is its $425 million partnership with DoorDash to expand its campus dining business, which operates at 450 colleges and universities and other locations.

Despite its growth, a move into the public markets will bring new challenges. Wonder is operating in an increasingly competitive food delivery landscape, going up against established names such as DoorDash and Uber Eats.

It feels like AI is cracking a new math problem every week. Last month, OpenAI released 719 manuscriptsclaiming solutions to some of the field’s hardest problems. But even the computer checks meant to confirm those results can slip. After years of AI slop, it’s fair to wonder whether math is next.

Axiom Math is betting math can avoid that fate, not by slowing down, but by standing on firmer ground. The $1.6 billion startup wants to build a verified foundation that AI agents can rely on to write software and, eventually, do science.

On Wednesday, the company announced it had formally verified the Stacks Project, one of the largest bodies of mathematics ever written by humans. Mathematicians often hold Stacks in rare regard — Columbia mathematician Peter Woit has claimed there’s “nothing else really comparable,” comparing the project to “the famous Bourbaki books,” a series that rebuilt math from its foundations in the 20th century.

Woit’s Columbia colleague Johan de Jong started the Stacks Project in 2005 as an open source reference. Mathematicians have contributed to it ever since, building out results from college-level algebra all the way up to the field’s most advanced ideas across algebra, geometry, and number theory.

The company’s fleet of AI agents translates the library’s roughly 7,654 pages into about 16 million lines of computer-checked code.

Formalizing math is a way of making sure the structure underneath it is sound. Mathematicians rewrite the proof as code in a programming language called Lean, and a program called a proof assistant checks the code line by line. If the proof passes, mathematicians can trust that it’s logically sound.

Engineers use the same idea, called formal verification, to confirm their chip designs work, while researchers in France have used it to catch bugs in software that calculates family benefits.

For a library like the Stacks Project, checking each theorem is only part of the job. The theorems and definitions in Stacks are interconnected, and each one builds on the ones before it. Like a family tree, many results in Stacks trace back to the ones that came before it.

That interconnectedness made Axiom unsure the project would work at first. The team had to keep every connection intact across about 16 million lines of Lean code.

“Things that are big and things that are delicate in life just don’t go together,” Axiom CEO Carina Hong told TechCrunch. “[Stacks] is a collection of mathematical results that is the foundation of mathematics. You want [theorems] to be consistent and not, like, paradoxical. You want the definitions to not fight with each other.”

The 25-year-old CEO dropped out of a Stanford math PhD program and founded the company in March 2025. In March, Axiom raised a $200 million Series A led by Menlo Ventures.

“We were not sure if vibe coding would be able to preserve the beauty of Stacks,” Hong said.

To pull off the project, the team gave its agents a loose structure and let them run in parallel. “With Stacks, we imposed a hierarchy … [but] it was not, like, a complete hierarchy of dictating what every single agent was doing,” said Ashvin Swaminathan, a member of Axiom’s technical staff. “The agents were finding tasks for themselves and identified things that they could work on in parallel.”

Axiom also trained its own models to be good at math, an approach Hong said most other verification startups haven’t taken. At first, the team relied on one of its in-house Lean experts to teach the agents a human sense of taste. Over time, Hong said, the agents have picked up taste on their own.

The result fills one of the biggest gaps in formalized math. “When you look at what has been formalized in mathematics, there’s this huge black hole. And that black hole covers algebraic geometry,” Swaminathan said.

Until now, researchers who wanted to check new work in algebraic geometry with Lean had little to build on, Swaminathan said. “Well, now they do.”

Axiom argues that each verified result becomes a building block, so future proofs can lean on checked math instead of starting from scratch. Terence Tao, the UCLA mathematician and Fields Medal winner, has made the general case more simply: “The more formalized proofs you have, the more formalized proofs you can make.”

Mathematicians aren’t Axiom’s only intended audience. Developers have long struggled to guarantee that their code is free of bugs and security holes. That problem grows as AI agents write more of the world’s code.

Axiom is betting that the software of the future will need to rest on verified foundations, the way Stacks now does.

“We believe that mathematics is the domain where you start building these things and that it’s a transferable skill,” Swaminathan said. “Agents can read our code, understand it, and build new infrastructure on top of it. Our Lean formalization is now the substrate that agents can build upon.”

The way Axiom’s agents built Stacks, working in parallel and picking their own tasks, is how Hong expects much of AI’s future work to happen. AI agents, she argues, will need a well-built, reusable body of knowledge to keep from producing slop.

“You want [agents] to have a reusable library for them to build on and keep pushing the frontier,” Hong said. “If you slop drop in a way that’s bad engineering choices in the beginning, then it’s very hard for them to grow that boundary.”

Hong is confident enough in that approach to issue a dare. “I openly challenge all the coding companies to work on formalization at this scale.”

In Brief

The blue wave swept Bluesky on Tuesday night as the microblogging platform recorded its highest daily number of posts, replies, and likes in 2026.

“We’re seeing a wave of new people joining and old friends coming back,” Bluesky CEO Toni Schneider posted.

According to Schneider, posts and likes doubled during the U.S. midterm elections on Tuesday.

COO Rose Wang told TechCrunch that usage began to peak between 8 p.m. to 10 p.m. PT, when many election results were rolling in. Daily posts and likes increased 30% and 70%, respectively, compared to the previous 30-day averages; clicks on links also spiked 20%.

are we about to break the all-time record for likes in a single day?? — alex benzer (@alexbenzer.com) 2026-11-05T00:24:42.561Z

According to open source data, Bluesky reached 44 million likes, 4.8 million posts, and 2 million new follows on November 4. This nearly eclipsed Bluesky’s all-time record 45 million likes in a single day, set in November 2024.

This jump was especially visible in Bluesky’s data traffic, which was 200% higher than usual, reaching up to 35.8 GB/s.

Bluesky is — sometimes snidely — referred to as “Twitter for the left,” given that its user base seems to skew that way politically. This isn’t a coincidence; Bluesky’s most concentrated period of user growth occurred after Donald Trump won the 2024 presidential election, so it makes sense that those users would post actively during a night of overwhelming victories for the Democratic party.

Zipline has raised $1.5 billion and ramped up its partnership with Walmart, a pair of deals that have raised the autonomous drone delivery and logistics startup’s profile and valuation.

Zipline and Walmart reached a new multi-year agreement to add 300 drone delivery sites across the United States. The two companies have had a drone delivery agreement since 2021, but this deal is a significant jump from the 21 active drone delivery sites they have together today.

On the fundraising side, Zipline’s $1.5 billion raise was led by Paradigm, WCM Investment Management, and Layer Global, with participation from Valor Equity Partners, Fidelity Management & Research Company, Uber, Tiger Global, J.P. Morgan, ARK Invest, and Salesforce Ventures. Zipline is now valued at $21.5 billion.

The company, which has developed a drone delivery ecosystem that includes aircraft, launch and landing systems, and logistics software, said it plans to use the funds to expand its U.S. manufacturing facilities and hire more workers to build out a nationwide drone delivery network.

Founded in 2014, Zipline got its start in Africa, where it used its autonomous drones to deliver blood in Rwanda, and later expanded to Ghana and Nigeria. The California-based company began its first operations in the U.S. in 2020, ferrying medical equipment and supplies to Novant Health in North Carolina. The following year, it struck a deal with Walmart.

But it was in 2022 that Zipline’s business really started to take off in the United States. Zipline received its Part 135 air carrier certification from the Federal Aviation Administration that year, a critical regulatory hurdle that authorized the company to complete long-range commercial drone deliveries. It has locked in a number of partnerships since then, including with Uber earlier this year.

Zipline now uses its drones to deliver food, retail, agricultural, and health products in five African countries, several cities in the United States, and Japan. Zipline’s home delivery service uses Zipline’s Platform 2 drones, which are designed to carry up to eight pounds and travel to customers within a 10-mile radius. Its larger Platform 1 drones are used for long-range deliveries for enterprise, business, and government that can cover 120 miles round trip.

The company said Thursday that its share of delivery revenue grew from 10.8% to 34.9% between March and October of this year. This latest funding aims to push that metric even higher.

TikTok has quietly launched a stand-alone TikTok Shop app in select markets as it continues to expand its e-commerce ambitions. The app will offer personalized product recommendations, the ability to shop through livestreams and short-form videos, app-exclusive deals, in-app checkout, order tracking, and customer support.

The app, which was first spotted by social media consultant Matt Navarra, is available in Singapore, Malaysia, Thailand, and the Philippines on iOS and Android.

The launch of a stand-alone app gives TikTok Shop a chance to reach shoppers beyond the main TikTok platform and compete more directly with established e-commerce apps like Amazon and Temu. It also suggests that TikTok is prepared to invest more heavily in social commerce by giving Shop its own dedicated destination rather than simply keeping it within the main app.

TikTok did not respond to TechCrunch’s request for comment regarding the rollout.

It’s unknown whether TikTok plans to release the app more widely and bring it to the United States. If TikTok does bring the app to the U.S., it would be competing not only with Amazon and Temu but also with big-box retailers like Walmart and Target, whose customers have grown accustomed to routinely ordering through the stores’ apps.

According to the app’s description, users will be able to access free shipping offers and exclusive first-order benefits, along with “secure payments for safe checkout.” By offering users app-exclusive deals, the company is incentivizing them to shop in the app rather than the main TikTok app.

Interestingly, the launch creates something of a full-circle moment in the rivalry between TikTok and Amazon. While Amazon previously introduced a TikTok-like short-form video feed called Inspire to draw consumers’ attention away from apps like TikTok and drive sales on its platform instead, TikTok is now making a similar play in reverse. With its new stand-alone commerce app, TikTok is looking to draw consumers away from platforms like Amazon and fuel more purchases within its own ecosystem.

The launch marks another step in TikTok’s efforts to branch out beyond social media. Although the company was once primarily focused on social networking and entertainment, TikTok has gradually ventured into new areas.

Beyond TikTok Shop, the company has added games, restaurant guides, sports hubs, hotel booking, and microdramas, and is also exploring payments.

Instagram announced on Thursday that it’s bringing Edits, its short-form video editing app, to desktop on Mac. The desktop version is designed for more complex video-editing projects that benefit from a larger screen, mouse, and keyboard, the company says.

The move puts Edits in more direct competition with ByteDance’s CapCut, the video-editing app Meta sought to challenge when it launched Edits last year. CapCut, which is a massively popular tool for creators, has been available on desktop for Windows and Mac since 2021.

Instagram notes that the desktop app includes licensed and royalty-free music, sound and video effects, and automatic captions in multiple languages. Audio and text styles that creators saved on the mobile app sync over to the desktop app, and users can customize keyboard shortcuts or import them from other editing software.

It also features a full multi-track timeline, keyframes for transitions, audio waveforms for accurate cuts, and more. Additionally, creators can import their own font files.

Image Credits:Instagram

“A lot of creators have described the same moment to us: a project gets too big for your phone,” Brett Westervelt, head of Edits, said in a blog post. “You’ve got a dozen clips going, you’re trying to land a cut on the exact beat, and your thumb covers the part of the timeline you need to see. Your work is growing, and your tools should grow with it. We’re expanding Edits to take on your most ambitious projects on desktop.”

Creators can publish from the desktop editor directly to Instagram with native features like tagging and location built in. The company says this gets rid of the need for creators to edit on desktop, download the file, move it to mobile, and then make more edits before posting it.

Edits will also support horizontal and vertical video formats, with watermark-free exports in 4K resolution for sharing across platforms or personal use, Instagram says.

The company plans to ship weekly updates to improve the desktop editor, noting that it’s currently working to make it possible for creators to edit one project across both desktop and mobile seamlessly.

Instagram also said it’s going to continue to grow the mobile Edits app for the more than 50 million creators who use it.

Audio AI company ElevenLabs on Thursday said users will be able to publish AI-generated music created using its ElevenMusic platform directly on SoundCloud.

ElevenLabs is also offering SoundCloud Artist Pro users three months of ElevenMusic Pro, which lets you generate up to 1,000 songs, split songs’ stem files 500 times, and five custom styles or vocals per month.

The company first released a music-generation model in 2025, and followed it up in April 2026 with a consumer app that lets you generate songs using prompts. Then a month later, it released a new music model that could switch genres mid-track and handle more complex vocals. ElevenLabs said that users have so far created over 25 million tracks using its models.

SoundCloud has also been giving AI tools to artists, though it’s taking a different approach. Last month, it partnered with AI music tooling company Moises to let artists export and import songs to and from SoundCloud. The company also has integrations with other platforms like Google Flow Music and the music remix platform, Hook.

“SoundCloud has always been a place where artists experiment, push boundaries, and turn creative ideas into something bigger. By becoming the first platform to bring ElevenMusic creations directly to fans beyond its own ecosystem, we’re opening up new possibilities for artists to experiment, share their work, and build an audience,” SoundCloud CEO Eliah Seton said in a statement.

ElevenLabs also has a partnership with Spotify in which authors can use its audio models to generate narrations for their audiobooks and publish them on the music streaming service. While Spotify has signed partnerships with labels for products that let people use AI for remixing tracks, it has not yet announced any model partners like ElevenLabs.

Overall, though, music streaming platforms today seem more inclined toward labeling and helping users filter out AI-generated music, especially as the tech slowly grows popular with musicians. Deezer, for example, is introducing a toggle that lets you filter out AI music completely, while Bandcamp has put a blanket ban on AI-generated tracks. Spotify and YouTube have taken a more cautious approach to labeling tracks.

Hardware startup Aina released a smart ring today that features a screen, a joystick, and a mic so that you can talk to your agents and take notes. The $399 ring has a speaker so you can take calls using it as well.

The startup, which raised $5.5 million in July, is positioning the ring as a controller of sorts for both AI and non-AI apps.

The company previously made Mirage, a three-key controller that let you invoke buttons based on the app on your Mac’s screen. For instance, if you were on a Google Meet call, you could map keys to camera on/off, mic mute/unmute, or to leave the meeting.

Aina has baked in lessons from that device to create the new ring, which it’s simply calling “AI ring.” Using its mic, you can take notes, send a prompt to agents, set reminders, and dictate notes on your Mac using Wispr Flow or the startup’s own dictation engine.

Image Credits:Aina

Now, there are already smart rings in the market that do similar things, like the Pebble Index 01, Sandbar Stream, Oasis, and Natura, which is probably why Aina has added more controls and functionality to stand out. Its ring comes with a speaker so that you can take calls, and there is a screen, which lets you see notifications from your phone or which apps you are controlling.

For navigation, there is a joystick, and it also supports gestures, so that it can trigger different actions based on what’s on your Mac or phone’s screen. The ring also lets you control music or even smart home devices like Apple TV.

I have used the ring for a few days, and it feels like a smartwatch in a ring format, with more controls. I can get notifications and reply to them. It surfaces my meetings; I can call Uber if I want to get a ride home.

It’s easier to use Apple TV’s remote itself, but the Aina ring proved useful the couple times I couldn’t find the remote. Granted, a lot of the software and features are still being fleshed out, so it’s hard to figure out where such a device ring might sit in my daily usage, but it is squarely aimed at folks who are surrounded by digital devices and AI applications.

Image Credits:Aina

“Phones and laptop screens were designed for an age of browsing where you think about opening apps. For interacting with proactive agents, we need a better interface, and that is what we had in mind while designing Aina,” Apoorv Shankar, Aina’s founder, told TechCrunch over a call.

Companies are trying to cater to programmers and AI enthusiasts with specialized hardware that could help them harness different agents. This includes OpenAI’s experimental keyboard and Rabbit trying to position itself as a controller. The hope with such hardware is that enough people will have multiple agents running to invest in specific gadgets to keep tabs on them.

Aina’s ring goes on preorder starting Thursday at early-bird pricing of $399, and the company plans to start shipping the device in January 2027. Once the early-bird period ends, the ring will cost $499.