FieldAI has raised $700 million at a $10 billion valuation, more than quadrupling its value in just over a year as investors pour capital into artificial intelligence systems designed to give robots greater autonomy in the physical world.

The latest financing values the California robotics company at five times the $2 billion valuation it reached after raising more than $400 million last year from investors including Jeff Bezos’ family office, Laurene Powell Jobs’ Emerson Collective and Khosla Ventures, according to a person familiar with the deal.

The investor leading the new round was not immediately clear.

Founded in 2023, FieldAI is pursuing a model of robotics that differs from the traditional approach of building software for a specific machine or narrowly defined task. The company describes its technology as a “universal general-purpose brain” designed to operate across robots, tasks and environments.

Its software is intended to control a broad range of machines, including humanoid robots, robot dogs, drones and industrial rovers. The underlying proposition is that the same AI system can allow different forms of physical machines to navigate and perform work in environments where conditions are unpredictable and cannot be fully programmed in advance.

That ambition has become one of the most heavily funded areas of the AI industry. Investors are increasingly targeting “physical AI,” referring to systems that can perceive their surroundings, make decisions, and act in the physical world rather than simply generate text, images, or computer code.

FieldAI’s new valuation puts it in the same broad group as some of the most highly valued private companies pursuing general-purpose robotics intelligence. Physical Intelligence is valued at roughly $11 billion, while Skild AI has been valued at more than $14 billion.

The large valuations reflect expectations that a general-purpose intelligence layer could eventually become more valuable than individual robot designs. If the software can be deployed across different types of hardware, robotics companies and industrial customers would not necessarily need to develop a separate AI system for every machine, environment or application.

But the technology faces a considerably harder test than conventional AI software.

A robot operating in a factory, construction site or outdoor environment has to deal with constantly changing physical conditions. Objects can move unexpectedly, surfaces can change, weather can interfere with sensors, and seemingly simple tasks can require a machine to understand its surroundings before acting. Errors that are tolerable in a software application can have physical consequences when an autonomous machine is operating around people, equipment, or valuable infrastructure.

The exigencies make the ability to generalize across environments one of the central challenges in physical AI. A system that performs reliably in a controlled demonstration still has to prove that it can maintain that performance when deployed at scale in unfamiliar locations.

FieldAI appears to be gaining traction with commercial customers as it attempts to demonstrate that its approach can move beyond research laboratories.

Since June, the company has added at least $35 million in revenue and customer contracts, taking its total to more than $135 million, according to the person familiar with the company. Business Insider previously reported that FieldAI had surpassed $100 million in revenue and customer contracts across more than 30 customers.

“We have seen very, very fast growth in the last several months,” Chief Executive Ali Agha told Business Insider in June.

The company’s customer base includes construction companies, data center operators and defense businesses. Those sectors are particularly relevant to autonomous robotics because they contain large amounts of physical work carried out in environments that can be difficult, dangerous, or expensive for humans to operate in continuously.

Construction sites, for example, change as projects progress and rarely resemble the controlled environments in which industrial robots traditionally operate. Data centers contain tightly organized but complex infrastructure, while defense applications can involve highly variable terrain and operating conditions.

FieldAI has also assembled a team from some of the technology industry’s most prominent AI and robotics organizations, hiring talent from Google DeepMind, Tesla, Nvidia and Boston Dynamics. The company is headquartered in Irvine and the Bay Area in California.

The fundraising comes as the robotics industry enters a period in which capital is now concentrated around companies claiming to develop general-purpose physical intelligence rather than single-purpose automation.

Traditional robotics businesses can generate revenue by selling machines designed to perform defined tasks, but their markets are often constrained by the economics and capabilities of each application. A general-purpose robotics platform potentially has a much larger addressable market if its software can be transferred across hardware and industries.

The challenge is proving that transferability.

However, FieldAI’s $10 billion valuation represents more than a bet on demand for robots. It is a bet that the company can solve one of robotics’ most difficult problems: creating an AI system capable of functioning reliably when the environment, task, and physical platform change.

The company’s growing customer contracts provide an early commercial signal, but revenue alone will not establish whether its technology can become a broadly deployable robotics platform. The more important test will be whether customers continue expanding deployments and whether FieldAI can maintain performance across increasingly complex environments without requiring extensive customization for every application.

That is also where competition among physical-AI companies is likely to intensify. Physical Intelligence, Skild AI and other well-funded startups are pursuing their own approaches to general-purpose robotic intelligence, while major technology and automotive companies continue investing in robotics, autonomous systems and AI models capable of interacting with the physical world.

FieldAI’s latest funding gives it substantially more capital to compete in that race. It also raises the expectations attached to the company.

At $10 billion, investors are no longer just financing an early-stage robotics experiment. They are placing a sizeable bet that a general-purpose AI “brain” can become a foundational layer for a future industry in which robots operate across factories, construction sites, warehouses, data centers, and other real-world environments.

Bitget has frozen about $1.1 million of the nearly $388 million in cryptocurrency stolen during last week’s cyberattack, but the exchange’s chief executive said the amount ultimately recovered is likely to remain limited as investigators trace the funds across the crypto ecosystem.

Gracy Chen, Bitget’s CEO, told CNBC that frozen assets had not necessarily been returned to the exchange and declined to disclose how much of the stolen cryptocurrency had actually been recovered.

The limited recovery outlook highlights the difficulty of retrieving digital assets once they have moved through a network of wallets and exchanges. Speaking on CNBC’s “Squawk Box Europe” on Wednesday, Chen said she was “not expecting to recover a lot of funds,” pointing to the historically limited recovery rates following major cryptocurrency exchange hacks.

But she said the incident also placed a responsibility on exchanges to demonstrate that they can protect customers and respond effectively when security failures occur.

“Exchanges have a responsibility to demonstrate how they protect users, particularly when something goes wrong,” Chen said.

Bitget has maintained that customer account balances were not affected by the attack. The exchange has instead committed its own capital to rebuilding a protection fund that was sharply depleted during the incident.

The scale of the theft initially made the incident one of the more significant recent attacks on a cryptocurrency trading platform, but subsequent investigations have revealed a more complicated attack path than a conventional theft of private keys. Investigation reports published on September 30 by Mandiant, part of Google Cloud, and blockchain security firm SlowMist concluded that attackers compromised two third-party security products before obtaining access to Bitget’s production wallet systems.

SlowMist traced the earliest malicious activity identified in the available logs to August 31, when the attackers exploited a previously unknown, or zero-day, vulnerability in one of the products.

Mandiant said the attackers subsequently obtained privileged internal access and were able to bypass Bitget’s normal customer-facing withdrawal process without stealing private keys. The incident did not depend on obtaining customers’ private keys and then authorizing conventional withdrawals. Instead, the attackers reportedly used compromised infrastructure and privileged access to manipulate the exchange’s production wallet environment.

“The method, I would say, is quite sophisticated,” Chen said.

She added that the attackers deleted traces of their activity after transferring the funds, complicating the investigation and the effort to follow the stolen assets.

Neither Mandiant nor SlowMist identified the affected security products in their public reports. Chen also declined to identify the vendors or products, saying that releasing information beyond the published findings could create additional security risks.

However, the incident illustrates a broader problem facing cryptocurrency exchanges as their security architecture becomes increasingly dependent on external software and infrastructure. A vulnerability in a third-party component can potentially provide attackers with a route into systems that otherwise have controls designed to prevent unauthorized withdrawals.

That creates a security challenge for exchanges that extends beyond safeguarding private keys and customer accounts. Vendor access, privileged credentials, monitoring systems, and production infrastructure can all become potential attack surfaces.

Bitget Rebuilds Protection Fund With Its Own Capital

Bitget’s response has also focused on reassuring customers that the financial consequences of the theft will not be transferred to account holders. Before the attack, Bitget valued its protection fund at more than $464 million. The fund fell below $200 million following the theft, according to Bloomberg’s calculation based on the wallet addresses Bitget has disclosed publicly.

The exchange subsequently rebuilt the fund to more than $300 million.

“We restored the Fund using Bitget’s own capital,” Chen said. “The financial impact is being absorbed by Bitget rather than passed on to our users.”

Chen said the replenished fund remains publicly verifiable on-chain and is separate from the reserves supporting customer balances.

Bitget’s latest Proof of Reserves, based on a September 29 snapshot, reported an overall reserve ratio of 131%, with all 19 covered assets showing reserves above 100%. The figures are self-reported by the exchange, meaning they provide an indication of the assets Bitget says it holds rather than independently resolving every question surrounding its financial position.

The distinction between the protection fund and customer reserves is important for users assessing whether the exchange can withstand the financial impact of the hack. Bitget’s stated approach is to use corporate capital to absorb the loss rather than draw directly from customer balances.

The investigation has also left unresolved questions about who carried out the attack.

The Mandiant and SlowMist reports did not attribute the incident to North Korea. Chen had previously said that preliminary technical indicators were highly consistent with known North Korean hacking groups, which have been linked to numerous cryptocurrency thefts.

Asked about the attribution after the new reports were released, Chen said the company would wait for more evidence.

“We will have to wait further for further details on this,” she told CNBC.

The uncertainty over attribution reflects the difficulty of identifying sophisticated crypto attackers, particularly when they deliberately erase traces and move stolen assets through multiple addresses.

Meanwhile, Bitget has begun restoring normal operations. Withdrawals of bitcoin, ether, and USDT have resumed, while withdrawals for the remaining cryptocurrencies, as well as fiat and peer-to-peer services, are scheduled to resume on Friday.

The immediate priority is shifting from containing the breach to rebuilding confidence. Bitget has restored much of its protection fund and says customer balances remain intact, but the relatively small amount of frozen funds compared with the nearly $388 million stolen illustrates the fundamental challenge facing the exchange.