News
IndustryTSMC’s grip on advanced chipmaking will face competitionIndustryCadence joins European Chip Design Platform to support semiconductor startups and SMEsMaking chipsMarc Hijink on ASML’s Culture, Supply Chain, and GeopoliticsProcessorsIntel's post-Nova Lake CPUs take shape as Razor Lake desktop and laptop variants surfaceAI chipsHow Google’s New AI Chip Software Helps It Battle NvidiaIndiaMore IC Packaging Plants Emerge In IndiaIndustryUS Tightens China AI Chip Access as Nvidia, TSMC and Semiconductor Stocks SlideDesignAxelera AI moves into HPC with TitaniaMaking chipsChinese envoy says chip equipment export curbs threaten EU-China trade thawIndiaSEMICON India 2026 Day 2 Sees 25 MoUs, Strategic Collaborations Across Semiconductor Value ChainIndustryTSMC’s grip on advanced chipmaking will face competitionIndustryCadence joins European Chip Design Platform to support semiconductor startups and SMEsMaking chipsMarc Hijink on ASML’s Culture, Supply Chain, and GeopoliticsProcessorsIntel's post-Nova Lake CPUs take shape as Razor Lake desktop and laptop variants surfaceAI chipsHow Google’s New AI Chip Software Helps It Battle NvidiaIndiaMore IC Packaging Plants Emerge In IndiaIndustryUS Tightens China AI Chip Access as Nvidia, TSMC and Semiconductor Stocks SlideDesignAxelera AI moves into HPC with TitaniaMaking chipsChinese envoy says chip equipment export curbs threaten EU-China trade thawIndiaSEMICON India 2026 Day 2 Sees 25 MoUs, Strategic Collaborations Across Semiconductor Value Chain
EveryChip
B200

B200

AI GPU with 208 billion transistors for trillion-parameter models

The B200 is a graphics processing unit designed by NVIDIA for artificial intelligence workloads. It features 208 billion transistors and supports models with up to ten trillion parameters. The chip uses a custom TSMC process and connects via high-speed links to form larger systems.

Share on WhatsApp

  • Contains 208 billion transistors on two connected dies
  • Supports AI models with up to 10 trillion parameters
  • Manufactured using custom TSMC 4NP process node
  • Delivers 1.8 TB/s throughput via fifth-gen NVLink
  • Reduces LLM inference energy use by up to 25x

What it is

The B200 is a graphics processing unit built for accelerated computing and generative artificial intelligence. NVIDIA designed the chip to handle real-time inference for large language models containing up to ten trillion parameters. It serves as the core component of the Blackwell platform, which aims to reduce energy use and operating costs for AI tasks. The architecture includes dedicated engines for reliability, security and data decompression to support complex industrial and scientific workloads.

How it came about

NVIDIA named the architecture after David Harold Blackwell, a mathematician known for his work in game theory and statistics. He was the first Black scholar inducted into the National Academy of Sciences. The B200 succeeds the Hopper architecture, which launched two years prior. Jensen Huang, the chief executive of NVIDIA, described the chip as the engine for a new industrial revolution in computing. The company announced the platform alongside major cloud providers and server makers who plan to adopt the technology.

Inside the chip

The chip contains 208 billion transistors and uses two reticle-limit dies connected by a ten terabyte per second link. This design allows the pair of dies to function as a single unified graphics processor. It features a second-generation Transformer Engine that supports four-bit floating point inference. The fifth-generation NVLink interface provides 1.8 terabytes per second of bidirectional throughput. A dedicated RAS engine handles reliability and preventative maintenance, while a separate decompression engine accelerates database queries. Secure AI capabilities protect data through native interface encryption protocols.

How it was made

TSMC manufactures the B200 using a custom-built 4NP process node. The construction involves connecting two large GPU dies to create a single logical unit. This method allows for a higher transistor count than a single die could typically support within standard manufacturing limits. The resulting chip packs 208 billion transistors into the package. The production process supports the high bandwidth and power efficiency required for trillion-parameter AI models.

Where it went

Major cloud providers such as Amazon Web Services, Google, Microsoft and Oracle intend to use the B200 in their data centres. Companies like Meta, Tesla and OpenAI also plan to adopt the hardware for training and running large language models. The chip supports workloads in engineering simulation, electronic design automation and computer-aided drug design. It is a key part of the GB200 Grace Blackwell Superchip, which combines the GPU with a Grace central processing unit. Systems based on this superchip target hyperscale AI clouds and high-performance computing tasks.

Versions and family

The B200 is the primary graphics processing unit in the Blackwell family. It pairs with the NVIDIA Grace CPU to form the GB200 Superchip. Two of these GPUs connect to one Grace CPU over a high-speed interconnect. Larger configurations include the GB200 NVL72, a rack-scale system that combines 36 superchips. This setup links 72 B200 GPUs and 36 Grace CPUs. The architecture replaces the previous Hopper generation, which served as the foundation for earlier AI accelerators from the company.

Why it matters

The platform claims to reduce energy consumption and operating costs by up to 25 times compared to its predecessor. It enables the training and inference of models with up to ten trillion parameters. Industry leaders state that the chip will accelerate discoveries in science and improve the efficiency of cloud services. The widespread adoption by major technology firms suggests a significant shift in infrastructure for generative AI. The design focuses on maximizing uptime for massive-scale deployments through improved resiliency features.

Questions people ask about B200

What is the NVIDIA B200 used for?

The B200 is a graphics processing unit designed for artificial intelligence workloads. It handles real-time inference for large language models and supports tasks in scientific simulation and data analytics. Major cloud providers use it to power generative AI services.

How many transistors does the B200 have?

The chip contains 208 billion transistors. NVIDIA achieves this count by connecting two reticle-limit GPU dies with a high-speed link, allowing them to function as a single unified processor for complex computing tasks.

Who manufactures the B200 chip?

TSMC manufactures the B200 using a custom-built 4NP process. This advanced node allows for the high density and performance required by the Blackwell architecture, which targets trillion-parameter AI models.

What is the predecessor to the B200?

The B200 succeeds the NVIDIA Hopper architecture. Hopper was launched two years before the announcement of Blackwell. The new design aims to provide significant improvements in cost and energy efficiency over the previous generation.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on B200 yet, so the facts on this page come from the sources listed above: nvidianews.nvidia.com/news/nvidia-blackwell-platform-arrives-to-power-, tomshardware.com/pc-components/gpus/nvidia-introduces-a-new-merged-cpu. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryChip wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    Photo credits on this page

    The small pictures above come from Wikimedia Commons under free licences. Some are cropped to fit.

    The weekly digest

    Every Monday: the chip of the day, new pages on EveryChip, and the conferences and anniversaries of the week ahead.