Nvidia's GTC Evolution: From Chip Maker to AI Infrastructure Giant

Every March, when Nvidia CEO Jensen Huang takes the stage in his signature black leather jacket, the global IT industry watches intently. The event is the GPU Technology Conference (GTC), where he delivers his keynote address.

GTC originally started as a technical conference for graphics developers. With the advent of the artificial intelligence era, it has become the most important event shaping the global technology paradigm for years to come. Examining the key announcements from the past three years of GTC—from 2024 to 2026—reveals Nvidia's evolution from a simple semiconductor manufacturer to a massive infrastructure company.

2024: Blackwell's Debut and the Dawn of the 'AI Revolution'

The centerpiece of GTC 2024 was the introduction of Blackwell, the next-generation AI accelerator. It dramatically improved computing and inference performance compared to the previous generation, setting the hardware standard for the generative AI era.

Nvidia announced Nvidia Inference Microservices (NIM), software that helps deploy AI models easily, marking the full-scale expansion of its ecosystem. The company also emphasized its Omniverse vision to support physical AI including humanoid robots, signaling that AI would extend beyond the digital world into the physical realm.

2025: Beyond Chip Maker to 'Infrastructure Company'

In 2025, Nvidia moved beyond focusing on single-chip performance to present a blueprint encompassing entire data center systems. The company recognized diversifying demand across various industry segments, including hyperscalers, sovereign AI, and enterprise AI-native clouds.

From this point, Nvidia solidified its identity not as a mere chip company but as a "platform and infrastructure provider." CEO Huang began emphasizing a virtuous cycle where securing computing capacity directly translates to corporate revenue. With the emergence of agentic AI, this soon became reality.

2026: Completion of the 'AI Factory' and Space Data Centers

The most recent GTC in 2026 signaled that the AI industry's center of gravity had completely shifted from training to inference. To handle token processing demand that surged 10,000-fold since ChatGPT's launch, Nvidia unveiled a thoroughly specialized architecture.

This GTC drew assessments that Nvidia's business scope has rapidly expanded. While previously known only for graphics cards and AI accelerators, the company is now ready to supply "AI factories" themselves as turnkey solutions, combining silicon, optical communications, and software.

Nvidia revealed the "Vera Rubin" system, comprising seven different chips and five dedicated racks, unifying entire data centers into a single AI factory. Notably, by integrating Groq's Language Processing Units (LPU), the company achieved up to 35-fold computational performance improvement in premium inference segments compared to the previous generation—thanks to deterministic design where the compiler statically controls data flow.

To drastically reduce power consumption, Nvidia commercialized co-packaged optics (CPO) technology in switch chips. The company also announced Project Space One, a plan to build data centers in space orbit to overcome physical power grid and cooling limitations.