Meta’s $2B AI chip bet faces setbacks as Rivos acquisition stumbles six months later

Meta spent roughly $2 billion to acquire AI chip startup Rivos with a clear goal: build more of its own AI hardware and rely less on Nvidia. Six months after the deal closed, that plan appears to be running into serious challenges.

According to a report from The Information, the integration of Rivos into Meta has been far from smooth. More than a quarter of the startup’s employees who joined Meta have reportedly been laid off, and development of a chip intended to train Meta’s largest AI models has been paused.

“Meta Platforms bought semiconductor startup Rivos last year to accelerate development of in-house chips and reduce its reliance on Nvidia as it pours cash into data centers for its AI ambitions. Now six months since the acquisition closed, Meta is struggling to make it work,” The Information reported.

The developments offer a rare look inside one of Meta’s most important AI infrastructure bets. They suggest that buying chip talent may be easier than turning that talent into products capable of competing with Nvidia’s dominance in AI computing.

A Strategic Acquisition in the AI Arms Race

News of the acquisition first surfaced in September 2025, when Bloomberg and Reuters reported that Meta was moving to acquire Rivos, a Santa Clara-based semiconductor startup focused on AI accelerators built around the open-source RISC-V architecture.

At the time, the deal made strategic sense.

Rivos had attracted attention across the semiconductor industry after raising roughly $370 million from investors and reportedly exploring a financing round that valued the company at about $2 billion. The startup was developing high-performance AI chips, including a CUDA-compatible RISC-V processor that had already been taped out and sent to Taiwan Semiconductor Manufacturing Company for trial production.

Meta was reportedly one of Rivo’s largest customers before the acquisition.

Following the deal announcement, Meta Vice President of Engineering Yee Jiun (YJ) Song said the acquisition would “help us accelerate our vision for scalable compute to power our AI ambitions” and expand work on the company’s Meta Training and Inference Accelerator, or MTIA, program.

The acquisition fit squarely into Meta’s larger effort to build a custom AI hardware stack capable of supporting models such as Llama across its growing network of AI data centers.

Why Meta Wanted Rivos

The acquisition came as major technology companies raced to gain greater control over the hardware powering artificial intelligence. Unlike many chip startups, Rivos began with the software stack before moving into hardware, a strategy its founders argue gives them an edge in meeting the demanding computational requirements of large language models and advanced analytics.

Rivos is also working on its own graphics processing unit, or GPU—the type of chip that drives most AI workloads. That makes it a particularly strategic fit for Meta, which already has a team building in-house AI chips under its Meta Training and Inference Accelerator (MTIA) program but still spends billions each year buying GPUs from external partners, including Nvidia, the market leader.

Meanwhile, Nvidia remains the dominant supplier of AI chips, but demand has exploded so quickly that many large technology companies have spent years developing alternatives. The appeal is obvious. Custom chips can lower costs, improve efficiency, and reduce dependence on external suppliers.

For Meta, the financial stakes are enormous.

The company is spending tens of billions of dollars each year on AI infrastructure, data centers, and computing capacity. Any reduction in spending on third-party GPUs could translate into significant long-term savings.

Rivos brought expertise that aligned closely with those goals. The startup specialized in RISC-V, an open-source instruction set architecture that offers companies more flexibility than proprietary alternatives. Its engineers had experience building full-stack AI systems and custom silicon, two areas Meta considers central to its long-term AI strategy.

On paper, the combination looked like a natural fit.

The Reality Has Been More Difficult

Six months after the acquisition closed, the picture looks far less straightforward.

The Information reports that Meta has faced integration challenges and strategic shifts inside its hardware organization. More than 25% of former Rivos employees have reportedly been laid off, and at least one important chip project focused on training large AI models has been halted.

The setbacks highlight a problem that has followed many technology acquisitions. Bringing a startup into a company the size of Meta can create friction around priorities, development processes, and decision-making.

Chip development adds another layer of difficulty.

Building advanced AI processors requires close coordination between hardware, software, manufacturing partners, and internal infrastructure teams. Integrating Rivo’s technology into Meta’s existing MTIA roadmap appears to have proven more difficult than anticipated.

Public discussions on workplace forums such as Blind have echoed reports of significant layoffs among former Rivos employees, though those accounts remain anecdotal.

A Reminder of How Hard AI Chip Development Really Is

Meta is hardly the first company to discover that building competitive AI chips is one of the hardest challenges in technology.

Google spent years developing its Tensor Processing Units before they became a major part of its AI infrastructure. Amazon invested heavily in Inferentia and Trainium before deploying them at scale across Amazon Web Services. Both companies continue to use Nvidia hardware extensively.

The Rivos situation serves as a reminder that acquiring talent and intellectual property is only the beginning. Turning those assets into production-ready silicon capable of supporting frontier AI models can take years and billions of dollars.

The challenge is particularly significant for RISC-V. The architecture has attracted growing interest across the semiconductor industry, but its ecosystem remains less mature than alternatives that have benefited from decades of software optimization and developer support.

What Comes Next

Meta has not publicly addressed the reported layoffs or the specific challenges described by The Information.

The company continues to present custom silicon as a key part of its long-term AI strategy, and the MTIA program remains a major area of investment. That suggests Meta is unlikely to abandon its chip ambitions.

Still, the early struggles surrounding Rivos raise questions about how quickly Meta can reduce its dependence on Nvidia and whether its internal chip roadmap will take longer to mature than investors expected.

The strategic logic behind the acquisition remains intact. The execution has proven much harder.

For a company investing tens of billions of dollars to build AI infrastructure, the success or failure of that effort could shape Meta’s competitive position for years to come.