Meta did not acquire FuriosaAI. On February 12, 2025, TechCrunch reported, citing Forbes, that Meta was in talks to buy the South Korean AI-chip startup. By March 24, reports said FuriosaAI had rejected a Meta offer worth about $800 million. The reported dispute involved post-acquisition strategy and organizational structure, and FuriosaAI continued as an independent company.
Current status as of August 16, 2026: FuriosaAI remains independent. Its public materials describe RNGD entering mass production and list commercial relationships involving LG AI Research, Broadcom, Samsung SDS and Equinix.
What was actually reported in February 2025?
The original story supported a narrow claim: Meta was reportedly considering an acquisition and could announce its intentions as soon as February 2025. TechCrunch attributed the report to Forbes. It did not establish that Meta and FuriosaAI had signed a definitive agreement.
| Statement | What the evidence supports |
|---|---|
| “Meta is in talks to acquire FuriosaAI” | Reported by TechCrunch, citing Forbes, on February 12, 2025. Read the report. |
| “Meta offered $800 million” | Reported by Korean media and cited by TechCrunch; it was not a publicly confirmed transaction price. |
| “Meta acquired FuriosaAI” | Incorrect. The reported offer was rejected and no acquisition was announced. |
Who is FuriosaAI?
Founded in 2017 by June Paik, a former Samsung Electronics and AMD engineer, FuriosaAI is a South Korean fabless semiconductor company. It designs processors mainly for AI inference—running trained models to produce answers or predictions—rather than general-purpose graphics or every stage of AI training.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →#1 Best Overall
- NVIDIA Volta GV100 Architecture — 4,608 CUDA Cores, 640 1st-Gen Tensor Cores delivering 14 TFLOPS FP32 and 112 TFLOPS deep learning performance for AI training, inference, HPC, and scientific computing workloads
- 32GB HBM2 ECC Memory — 900 GB/s Bandwidth — High-bandwidth memory on a 4096-bit bus with ECC error correction provides the memory capacity and throughput required for the largest AI models, simulations, and datasets
- PCIe 3.0 x16 Interface — 250W TDP — Standard PCIe Gen3 connectivity with passive cooling designed for enterprise rack server deployment in HPE ProLiant, Dell PowerEdge, and Supermicro platforms with adequate chassis airflow
- NVLink — Scale to 96GB Unified Memory — Connect two V100 GPUs via NVLink at 300 GB/s bi-directional bandwidth to scale GPU memory from 32GB to 96GB for larger AI training and HPC workloads
- Multi-Precision Computing — Supports FP64 (7 TFLOPS), FP32 (14 TFLOPS), FP16 (112 TFLOPS) and INT8 precision modes for flexible deployment across training, inference, and scientific simulation workloads
The company’s product history includes the Warboy accelerator and RNGD, pronounced “Renegade.” Its proprietary Tensor Contraction Processor architecture targets tensor operations used by modern models, while the Furiosa software development kit is intended to compile, optimize and deploy those models on its hardware. FuriosaAI’s company background is available at furiosa.ai/about.
What does RNGD do?
RNGD is a specialized data-center inference accelerator. FuriosaAI’s current NXT RNGD Server page lists the following vendor-stated specifications:
Rank #2
- High-Performance AI Processing: The MX3 is designed to handle the most demanding AI computer vision workloads, delivering exceptional performance and efficiency.
- Flexible Integration: The MX3 can be easily integrated into your existing systems via its M.2 M-key form factor and support for Linux operating systems.
- Energy Efficient: The MX3 is designed to provide high performance while minimizing power consumption.
- Comprehensive Software Development Kit (SDK): The MX3 is supported by a comprehensive SDK that simplifies development and deployment.
- Hardware compatability: The MX3 is compatible with the PCI-SIG M.2 M-key 2280 Specification. It can be used with the Raspberry Pi 5 with a M-key 2280 HAT.
| Specification | FuriosaAI’s stated figure | Important qualification |
|---|---|---|
| Server configuration | Eight RNGD cards | Company specification for the NXT RNGD Server |
| Performance | Up to 4 petaflops | Vendor-stated; workload and measurement conditions are not independent test results |
| Memory | 384 GB HBM3 | Server-level figure |
| Memory bandwidth | 12 TB/s | Vendor-stated |
| Power | Approximately 3 kW per listed server configuration | Vendor-stated system figure |
A separate FuriosaAI and Broadcom announcement describes an RNGD PCIe accelerator built on TSMC’s 5-nanometer process with 180 watts of power per card. Those figures describe a specialized inference product, not a universal replacement for Nvidia GPUs used in training, graphics or unrelated workloads. See the announcement at Nasdaq.
Why would Meta want FuriosaAI?
The reported talks fit Meta’s broader effort to expand its AI-computing options. Meta has been developing internal accelerators while relying heavily on Nvidia hardware. Buying FuriosaAI could have provided chip-design talent, an inference-focused architecture and software expertise without building all of those capabilities internally.
Rank #3
- ✅Powered by 26 Tera-Operations Per Second (TOPS) Hailo-8 AI Processor. 2.5W typical power consumption
- ✅Scalable, enabling simultaneous processing of multi-streams & multi-models
- ✅Enabling real-time, low latency and high-efficiency AI inferencing on the edge devices
- ✅Supports TensorFlow, TensorFlow Lite, ONNX, Keras, Pytorch frameworks
- ✅Supports Linux and Windows. Supports the temperature range of -40°C to 85°C
- Less dependence on Nvidia: A second hardware platform could give Meta more negotiating leverage and supply flexibility.
- Better inference economics: Serving large language models at scale makes power, memory and throughput significant operating costs.
- Specialized expertise: FuriosaAI’s focus is narrower than a general-purpose GPU, but that specialization can be useful for suitable inference workloads.
- Alignment with Meta’s spending: TechCrunch reported that Meta planned to spend up to $65 billion in 2025 on AI-related infrastructure and initiatives.
These are strategic reasons the talks made sense, not a statement from Meta that it intended to replace every Nvidia system with RNGD.
Why did the acquisition talks collapse?
On March 24, 2025, TechCrunch reported that FuriosaAI had turned down an offer of approximately $800 million, citing Korean media. Yonhap also reported that FuriosaAI decided not to proceed with takeover negotiations. FuriosaAI and Meta did not announce a completed deal; TechCrunch reported that FuriosaAI declined to comment and that Meta had not immediately responded to a request for comment at the time.
Rank #4
- 48GB AI graphics accelerator
The reported disagreement concerned what would happen after a purchase: FuriosaAI’s business direction, organizational structure and how the startup would operate inside Meta. Available coverage attributed the breakdown to those strategic and organizational issues rather than to a documented demand for a higher price. An offer of $800 million should therefore not be described as FuriosaAI’s valuation or as a completed transaction value. TechCrunch’s follow-up contains the reported details.
What did FuriosaAI do after staying independent?
FuriosaAI pursued customers and infrastructure partners instead of becoming a Meta division.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Best Value
- Professional AI & Creator Workstation: AMD Radeon AI PRO R9700 GPU with 32GB GDDR6 is engineered for AI development, professional content creation, and compute-intensive workloads.
- Massive 32GB Memory Capacity: 32GB of GDDR6 memory on a 256-bit bus provides ample bandwidth for large AI models, 8K video editing, and complex 3D rendering.
- Advanced RDNA 4 with AI Accelerators: 64 Compute Units with 3rd Gen Ray Tracing and dedicated 2nd Gen AI Accelerators for groundbreaking AI performance and visual computing.
- Professional Blower Cooling: Efficient single blower design exhausts heat directly out of the chassis, ideal for multi-GPU workstation and server configurations.
- Enterprise-Grade Thermal Solution: Vapor chamber heatsink with industrial Honeywell PTM7950 thermal interface material ensures reliable cooling under sustained professional loads.
- LG AI Research: In July 2025, FuriosaAI announced that RNGD would support LG’s EXAONE ecosystem. FuriosaAI said its testing on EXAONE workloads produced 2.25-times better inference performance than competing GPUs. That is a company-reported result tied to particular models and test conditions, not a universal benchmark. TechCrunch reported the partnership.
- Broadcom: A May 2026 strategic partnership focused on a next-generation, rack-scale inference platform.
- Samsung SDS: A July 2026 announcement concerned a domestic Korean NPU-as-a-service offering.
- Equinix: FuriosaAI has described RNGD infrastructure deployed at an Equinix data center in Lisbon.
FuriosaAI’s newsroom and company pages list these developments and state that RNGD has entered mass production: furiosa.ai/newsroom.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What this means for the AI-chip market
The episode illustrates several competing routes away from a single dominant accelerator supplier: internal chip design, acquisitions, strategic partnerships and workload-specific inference hardware. Meta could have obtained proprietary technology and talent through an acquisition. FuriosaAI instead retained the ability to sell to multiple customers and build an independent business. The later LG, Broadcom, Samsung SDS and Equinix relationships are consistent with that independent path, although they do not prove that independence was the sole reason FuriosaAI rejected the offer.
For investors and business readers, the important distinction is between technical potential and commercial scale. A chip can offer strong performance per watt yet remain difficult to deploy if its compiler, supported operators, quantization formats, supply chain or customer support are limited. A benchmark advantage on one model may disappear on another, and lower chip power does not eliminate networking, cooling or memory requirements.
What a potential enterprise customer should check
FuriosaAI is aimed at data-center and enterprise deployments, not ordinary PC buyers. Hardware or evaluation pricing was not publicly stated in the reviewed materials as of August 16, 2026. FuriosaAI advertises Furiosa Access evaluation locations including Seoul, the Bay Area, Lisbon and Johor Bahru.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCrashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minute- Whether the target model and required operators run on the current Furiosa SDK.
- Support for the customer’s quantization formats and model architecture.
- Latency and throughput at the actual batch size, sequence length and concurrency.
- Engineering effort required to port from CUDA-based software.
- HBM capacity, memory bandwidth, networking and rack-power limits.
- Availability in the required geography, plus warranty, replacement and support terms.
- Total cost of ownership, including software migration, utilization, cooling and deployment.
- Whether the workload needs training as well as inference.
Customers needing broad CUDA compatibility, extensive training support or consumer graphics are generally evaluating a different category of product. Relevant alternatives include Nvidia data-center GPUs (Nvidia), AMD Instinct with ROCm (AMD), AWS Trainium and Inferentia (AWS) and Google Cloud TPUs (Google Cloud).
Quick Recap
Timeline
- 2017: FuriosaAI is founded.
- February 12, 2025: Meta’s reported acquisition talks are published by TechCrunch, citing Forbes.
- March 24, 2025: Reports say FuriosaAI rejected Meta’s approximately $800 million offer.
- July 2025: FuriosaAI announces the LG AI Research EXAONE partnership.
- May 2026: FuriosaAI announces its Broadcom partnership.
- July 2026: Samsung SDS announces an NPU-as-a-service initiative involving FuriosaAI.
- August 16, 2026: FuriosaAI’s public materials show an independent company pursuing mass production and commercial deployments.
What readers should not conclude
- Meta did not buy FuriosaAI.
- The reported $800 million figure is not a confirmed valuation or completed deal price.
- RNGD is not a universal Nvidia replacement.
- The 2.25-times LG result is a company claim for EXAONE testing, not proof of across-the-board superiority.
- Partnerships with LG, Broadcom, Samsung SDS or Equinix do not represent ownership transactions.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
