FuriosaAI Turns Down Meta’s $800M Offer Following Breakthroughs in LLaMA 3.1 and AI Inference

• Meta reportedly bid $800 million to acquire FuriosaAI.
• Furiosa turned the offer down and will continue developing its AI chips independently.
• Furiosa’s chips are positioned as a faster and more efficient alternative to GPUs.

Amid huge investment in AI infrastructure, Big Tech firms like Meta are increasingly interested in developing their own hardware.

However, the Facebook owner’s latest attempt to gain a foothold in the AI chip space was struck down after FuriosaAI turned down an $800 million acquisition offer.

FuriosaAI Rejects Meta Takeover Attempt

Founded in 2017, Seoul-based FuriosaAI is among a crop of new startups developing AI inference solutions that could reduce data centers’ reliance on Nvidia chips.

Amid surging demand for AI compute, the company became an acquisition target for Meta, which reportedly bid $800 million for Furiosa.

However, according to several media reports, Furiosa rejected the offer and will instead continue to grow its business independently.

Meta’s AI Hardware Ambitions

With its growing cloud business and ambitious AI plans, securing a hardware advantage is strategically important for Meta.

Following in the footsteps of the Big Three cloud giants, the company has even started to develop its own AI chip . But such initiatives inevitably take time and acquiring a hot chip startup like Furiosa would give Meta a major boost.

GPU Alternatives

To date, the AI infrastructure race has mostly been characterized by battles over who can build the biggest GPU cluster.

But as AI adoption increases demand for inference, other factors are coming into play.

Like similar solutions from Groq and Cerebras, Furiosa’s chips are positioned as a faster and more efficient alternative to GPUs.

For example, the company claims its RNGD platform offers “unmatched performance,” when running Meta’s LLaMA 3.1-8B model, achieving a throughput of 3,200–3,300 tokens per second while using just 181 watts of power per card.

In contrast, high-end GPUs like Nvidia’s Blackwell can consume up to 1200 watts each.