NVIDIA

NVIDIA introduced CUDA, a parallel computing platform and API enabling GPUs to run general-purpose compute-intensive applications. This investment, costing over $1 billion, became the foundation for GPU-based AI computing.

AlexNet, trained on NVIDIA GPUs using CUDA, won the ImageNet competition by a wide margin, demonstrating the power of GPU-accelerated deep learning and sparking the deep learning revolution that NVIDIA would dominate.

NVIDIA gifted its first DGX-1 supercomputer to OpenAI, designed to reduce training time for large AI models. This marked NVIDIA's strategic positioning as the hardware backbone of the AI industry.

NVIDIA announces the Volta V100 GPU, the world's first GPU to feature Tensor Cores — dedicated hardware units designed specifically for deep learning matrix operations. Tensor Cores delivered up to 12x the training throughput of previous Pascal-generation GPUs, fundamentally transforming the speed and scale of AI model training. The V100 became the de facto standard for AI research and was deployed in the DGX-1 supercomputer.

NVIDIA released the A100 GPU based on the Ampere architecture, specifically designed for AI training and inference in data centers. It became the workhorse GPU for AI labs worldwide, including DeepSeek's training clusters.

NVIDIA agreed to acquire Arm Limited from SoftBank for $40 billion in stock and cash, in what would have been the largest semiconductor acquisition in history. The deal was intended to create a dominant computing platform combining NVIDIA GPUs with Arm's ubiquitous mobile CPU designs. It triggered an 18-month regulatory battle across the US, UK, EU, and China.

NVIDIA and SoftBank officially terminated the $40 billion Arm deal after significant regulatory challenges, including an FTC lawsuit, UK national security concerns, and EU antitrust probes. The collapse forced NVIDIA to pursue organic CPU development (Grace) instead, while Arm pivoted to a public listing. The failed deal shaped the industry for years, leaving Arm independent.

NVIDIA announced the H100 GPU based on the Hopper architecture, featuring the Transformer Engine optimized for transformer model training. The H100 became the most sought-after hardware in the AI industry, with multi-year waitlists and commanding premium pricing as demand far outstripped supply during the AI boom.

NVIDIA launched the NVIDIA AI Enterprise software platform and NeMo framework for building and deploying custom generative AI models. The platform extended NVIDIA's influence beyond hardware into the AI software stack, providing enterprises with end-to-end tools for LLM training, customization, and deployment.

NVIDIA announced the GH200 Grace Hopper Superchip, combining a Grace CPU with H100 GPU using NVLink-C2C interconnect for unified memory access. The architecture enabled training of trillion-parameter models on a single superchip, becoming the building block for the world's largest AI supercomputers.

NVIDIA became the seventh U.S. company to reach a $1 trillion market valuation, driven by explosive demand for its AI GPUs during the AI boom. The company's data center revenue surged as every major AI lab purchased NVIDIA hardware.

NVIDIA's H100 GPU (Hopper architecture) became the standard for large language model training and inference. Demand far exceeded supply, with the chips selling for $30,000-$40,000 each and driving NVIDIA to record revenues.

NVIDIA's market capitalization surpassed $2 trillion for the first time, driven by insatiable demand for its AI GPUs from cloud providers and enterprises. The milestone cemented NVIDIA as the primary beneficiary of the AI infrastructure buildout and made it one of the most valuable companies in the world.

NVIDIA unveiled the Blackwell B200 GPU, the world's most powerful AI chip with 208 billion transistors delivering up to 20 petaflops of AI performance. Blackwell represented a generational leap in AI compute density, enabling training of models over 10 trillion parameters and reducing AI inference costs by up to 25x.

NVIDIA became the first company in the world to surpass $4 trillion in market capitalization amid unprecedented demand for AI data center hardware. It later surpassed $5 trillion, cementing its position among the 'Magnificent Seven'.

NVIDIA announced a new AI chip designed to handle complex functions including video and software generation, planned for launch by end of 2026. This expanded NVIDIA's roadmap beyond traditional GPU compute into specialized AI workload accelerators.

NVIDIA committed up to $100 billion to support OpenAI's massive data center expansion built around NVIDIA's AI processors. This deal cemented NVIDIA's position as not just a chip supplier but a major capital partner in AI infrastructure.

NVIDIA makes its largest acquisition ever, buying AI chip startup Groq for approximately 0 billion to enhance AI inference capabilities. The deal brings Groq's Language Processing Unit (LPU) technology into NVIDIA's portfolio and is integrated into the Rubin platform.

NVIDIA launches the Rubin platform at CES 2026, comprising six new chips — Vera CPU, Rubin GPU, NVLink 6 Switch, ConnectX-9 SuperNIC, BlueField-4 DPU, and Spectrum-6 Ethernet Switch — designed as an AI supercomputer platform. Built on TSMC 3nm with 336B transistors, 288GB HBM4, and 22TB/s bandwidth, Rubin delivers up to 10x inference cost reduction over Blackwell.

NVIDIA invested $30 billion in OpenAI's record-breaking funding round (alongside Amazon's $50B and SoftBank's $30B), which valued OpenAI at $730 billion and was later extended to $120B in March 2026. This was the largest private technology fundraise in history at the time.

At GTC 2026 in San Jose, Jensen Huang detailed the Vera Rubin platform architecture, announced Dynamo 1.0 open-source inference framework, showcased NemoClaw enterprise AI agent platform, and demonstrated Groq-powered inference optimizations. Microsoft announced Fairwater AI superfactories using Vera Rubin NVL72 systems.

NVIDIA unveiled Cosmos 3 at GTC Taipei/COMPUTEX 2026, an open physical AI foundation model built on a breakthrough mixture-of-transformers architecture. The model unifies vision reasoning and multimodal generation across text, image, video, ambient sound and action. Available in Cosmos 3 Super (64B) and Cosmos 3 Nano (16B) variants, it was released as open-source on Hugging Face.

SK Group and NVIDIA announced a USD 500 billion-plus strategic collaboration spanning AI factories and next-generation memory. SK Telecom will build a 2-gigawatt NVIDIA Vera Rubin DSX AI Factory (online 2027). SK Hynix was locked as long-term HBM supplier, securing roughly 60 percent of global HBM capacity. NVIDIA also invested USD 1 billion in Naver. This is the largest single AI infrastructure commitment announced to date.

NVIDIA Labs open-sourced NOOA (NVIDIA Object-Oriented Agents), a Python framework where an AI agent is defined as a single Python class with methods as actions, fields as state, and docstrings as prompts. The framework achieved 82.2% on SWE-bench Verified and 86.8% on CyberGym L1 benchmarks. NOOA was released under Apache 2.0 license alongside the launch of the Open Secure AI Alliance, a 37-member industry coalition. The release marked NVIDIA's first dedicated open-source agent framework, extending its software ecosystem beyond hardware and infrastructure into the AI agent development space.

NVIDIA announced a $1.5 billion investment in SB Energy, SoftBank's data center developer, and provided up to $105 billion in credit support to build OpenAI's PORTS-Pike data center in Pike County, Ohio. The facility will scale from 4.25 GW to 8 GW and exclusively host NVIDIA compute under a 20-year OpenAI lease. The deal signals NVIDIA's strategic evolution from chip supplier to AI infrastructure financier, with the credit guarantee ensuring its GPUs power the $33 billion natural gas-backed campus on former DOE land.

NVIDIA agreed to pay $6 billion for a non-exclusive license to AI startup Poolside's Model Factory platform (behind its Laguna family of open-weight coding models) and extended job offers to 109 Poolside employees. NVIDIA also invested $1 billion in Poolside at a $12 billion pre-money valuation, without acquiring the company. The deal signals NVIDIA's deepening pivot from chip supplier to AI model developer, securing internal model-building capability as competition in AI training hardware intensifies.

Nvidia announced at Hot Chips 2026 that its Groq 3 LPX, the dedicated AI inference accelerator built from the $20 billion Groq acqui-hire, has entered full production. The chip slots into the Vera Rubin platform with up to 256 LPX accelerators per rack and is designed to deliver ultra-fast token generation for agentic AI workloads. Nebius Group signed on as the first cloud customer.

NVIDIA announced a 3.5 billion USD investment in MediaTek convertible bonds -- about 90 percent of the issue -- alongside an expanded collaboration in which MediaTek adopts the NVIDIA NVLink Fusion platform, letting its custom AI accelerators plug into NVIDIA NVLink-connected rack-scale AI factories. The move extended NVIDIA's ecosystem beyond its own silicon for the first time at scale, and MediaTek shares jumped 10 percent in Taipei trading. NVIDIA framed it as opening its AI infrastructure to third-party XPUs, deepening its edge-to-cloud and automotive ambitions.

NVIDIA announced its largest acquisition ever: buying Hugging Face, the central hub of the open-source AI ecosystem, for approximately 12.9 billion USD in cash. The deal extends NVIDIA's spending spree beyond chips into the software layer where millions of developers build and share models, and raised concerns in the open-source community about the independence of the de facto public library of AI. It bookends NVIDIA's pivot from GPU vendor to full-stack AI platform company.