Nvidia Posts $81.6B Quarter, Doubles Startup Stakes to $43B Amid Agentic Expansion

Nvidia reported $81.6 billion in revenue for the quarter ending April 26, representing a 20% increase from the prior quarter. [3] The company's data center revenue reached a record $75.2 billion. [3] Looking ahead, Nvidia projected $91 billion in revenue for the next quarter, signaling a growth deceleration to 12%. [4] Despite the projected growth deceleration, the company authorized $80 billion in share repurchases. [4]

In a notable shift, Nvidia revealed its stakes in privately held companies nearly doubled to $43 billion by April 26, up from $22 billion at the start of the quarter. [4] This jump was driven by $18.5 billion in purchases during the quarter, compared to just $649 million in the previous quarter. [4] The company also committed to investing $30 billion in OpenAI in February, while CEO Jensen Huang noted a pending, significant capacity buildout for Anthropic. [3]

Beyond hardware, Nvidia unveiled trusted 24/7 AI agents for telecom operators at the DTW Ignite 2026 conference in Copenhagen. [14] The company stated that the telecom industry is moving toward autonomous operations where AI agents can reason, make decisions, and execute actions with minimal human intervention. [14] Because 54% of operators identify data-related barriers as their biggest obstacle, SoftBank Corp. is utilizing technologies like Nvidia NeMo Safe Synthesizer and NeMo Anonymizer to generate privacy-preserving synthetic datasets for fine-tuning. [13]

In the robotics domain, researchers at the Nvidia GEAR lab developed an agentic harness called ENPIRE that enables AI coding agents to autonomously direct robot training. [9] Using this framework, AI agents figured out a training regimen that successfully taught robots to cut zip ties and insert GPUs into motherboard sockets. [9] The ENPIRE harness was tested with models including OpenAI’s Codex with GPT-5.5, Anthropic’s Claude Code with Opus 4.7, and Moonshot AI’s Kimi Code with Kimi K2.6. [9]

Nvidia's recent research emphasizes that the software wrapper around an AI model is more critical than the model itself for long-horizon tasks. [20] According to Adel El Hallak, vice president of product in Nvidia’s AI unit, an agent consists of the model, the tools it utilizes, the runtime, and the associated skills and libraries. [20]

To demonstrate this, researchers used a custom harness tweaked to handle memory well and include a supervisor component. [20] With the AVO harness, Claude Opus 5 achieved a 100% score on the interactive reasoning benchmark ARC-AGI-3, a set of 2D games with no instructions. [20] Without the harness, the same Opus 5 model scored 30%, which was previously the top result among tested models. [20]

On the infrastructure side, Nvidia announced that its latest AI system can be cooled with a recirculated liquid mixture operating at 113 degrees Fahrenheit. [11] Nvidia's chief sustainability officer, Josh Parker, claimed that the water consumption challenge for data centers is largely solved by this next generation of infrastructure. [11]

Meanwhile, US export controls have squeezed supply routes to China, causing the price of Nvidia's flagship DGX B300 server to double on the Chinese black market to more than 8 million renminbi, or $1.1 million. [15] The price of the RTX 6000 Pro workstation chip also rose from approximately 50,000 renminbi to as much as 130,000 renminbi. [15] To comply with US export control rules, Nvidia plans to begin small-batch shipments of a tailored LPU for Chinese customers by the end of 2026. [19]

Nvidia’s massive $18.5 billion quarterly injection into private startup stakes marks a shift from pure hardware supplier to aggressive ecosystem kingmaker, directly mirroring its $30 billion commitment to OpenAI. By explicitly linking hardware capacity buildouts to investments in frontier labs like Anthropic, Nvidia is cementing its dominance across the entire AI stack. The deployment of the ENPIRE harness across rival models—including Claude Code and Codex—shows Nvidia positioning its scaffolding as the indispensable layer for autonomous reasoning, regardless of which foundational model wins out. While the liquid cooling advancements attempt to blunt environmental criticism of this massive data center expansion, the soaring black market prices for DGX B300 servers highlight the persistent geopolitical friction in deploying this infrastructure globally. What the sources don't address: How regulators will respond to Nvidia leveraging its hardware monopoly to rapidly acquire $43 billion in captive software and startup equity.