NVIDIA's GTC Berlin event will run from October 20 to 22, and with the debate over AI investment returns intensifying, the conference is being given significance by the market that goes well beyond a mere technology launch.
According to the official agenda, October 20 will be dominated by training workshops, while Jensen Huang will deliver his keynote speech at 11 a.m. local time on October 21.
Ahead of the event, the AI industry is undergoing a round of market stress testing. Debt and financing risks at AI cloud computing companies have sparked investor concern, and the market has begun reassessing whether highly leveraged expansion can be sustained; at the same time, disputes over IPO valuations of AI startups are intensifying, with whether massive infrastructure spending can be supported by future commercial revenue becoming a focal point on Wall Street. Last week, AI-related assets were sold off, storage stocks fell sharply, and capital shifted toward defensive and value-oriented sectors, as the market grew increasingly cautious about the sustainability of the AI investment boom.
Against this backdrop, investors' expectations for this GTC go beyond new chips and technical specifications; they are more focused on whether NVIDIA (NVDA) can further clarify the growth space for AI computing demand and how massive capital expenditure will translate into sustainable commercial returns. At the same time, the conference's impact may extend beyond NVIDIA itself: as competition in AI infrastructure expands into areas such as CPUs and industrial AI, chipmakers like Intel could also be affected.
Rubin Deployment Timeline: What the Market Wants Is a Schedule, Not Specs
NVIDIA's strongest fundamental support still comes from its data center business. In the second quarter of fiscal 2027, the company's total revenue reached $96.2 billion, up 106% year over year; data center revenue was $89 billion, up 117% year over year, accounting for more than 92% of total revenue.
Blackwell is still in the phase of large-scale revenue realization, while Vera Rubin takes on the task of the next platform upgrade.
It is worth noting that NVIDIA's strategic focus is undergoing a structural shift — from emphasizing the performance of a single GPU to integrating CPU, GPU, networking, storage, software and data center design into a complete "AI factory." This change is closely tied to the migration of AI workloads from training to inference: large-scale inference requires balancing latency, throughput, energy efficiency and cost per token, and what customers ultimately buy is no longer just a GPU, but a computing system that can lower long-term operating costs.
It is understood that at the San Jose GTC in March this year, NVIDIA disclosed details of the Vera Rubin system, the Kyber rack and the next-generation Feynman architecture; at the Taipei GTC in June, NVIDIA further announced that the Vera Rubin system had entered full mass production and unveiled the standalone CPU Vera and its accompanying rack.
Therefore, the market's core expectation for Rubin at this GTC is not another round of specification upgrades, but the actual procurement expansion timetable of major cloud providers and enterprise customers, and how much the new platform can reduce inference costs.
AI Agents and Physical AI: The Next Argument for Expanding Computing Demand
AI Agents are another core topic of this GTC.
Unlike the one-off question-and-answer format of traditional chatbots, agents need to continuously call models, retrieve information, use tools and handle multi-step tasks. Even if the cost of a single inference declines, continuously running agents may still significantly increase overall computing volume. NVIDIA hopes that by providing chips, interconnect devices, inference software and development tools, it can expand computing demand from a handful of AI labs to scenarios such as customer service, software development, finance, healthcare and enterprise operations.
However, growth in computing consumption does not necessarily equal an increase in commercial value. Speakers on security topics at this conference include CrowdStrike CEO George Kurtz, Hugging Face co-founder Thomas Wolf and Red Hat's Chris Wright. This arrangement itself reveals the real threshold facing large-scale deployment of AI Agents — identity verification, access management and the maturity of enterprise software architecture will determine the pace of Agent adoption just as much as model capabilities.
Physical AI and robotics represent a longer-term growth space. Participants in this conference include Chen Zhaopeng of Agile Robots, Arthur Kaindl of Siemens Healthineers and Ola Källenius, CEO of Mercedes-Benz Group. Germany's vast industrial system makes Berlin a suitable location to promote this strategy, but the commercialization cycle for robotics and autonomous driving is clearly longer than that of data center chips, and product reliability, regulatory approvals and customer validation may all affect the pace of revenue realization.
Unexpected Beneficiary: GTC May Matter More for Intel
There is also a potential thread at this GTC that the market has underestimated — Intel may be an unexpected beneficiary of the conference.
Currently, Intel Xeon processors serve as the host CPUs for Blackwell-based DGX B300 and Rubin-based DGX Rubin NVL8 systems. For the next-generation Feynman architecture, NVIDIA is expected to introduce the Arm-based Rosa CPU, but Intel Xeon may still continue to serve as the host CPU for Feynman-series DGX systems, and any confirmation of this would have a positive impact on the sustainability of Intel's NVIDIA-related revenue.
More critically, there are persistent market rumors that NVIDIA may manufacture Feynman-generation I/O chips on Intel's 18A-P or 14A nodes. At the same time, given that TSMC's CoWoS advanced packaging capacity is currently tight, whether NVIDIA will include Intel's EMIB-T or Foveros advanced packaging solutions in Feynman's packaging supply chain is also highly watched. Once any of the above is positively confirmed, it could trigger a repricing of market expectations for Intel's revenue.
This GTC has as many as 142 external speakers, covering institutions such as Nebius, Mercedes-Benz, Siemens, Schneider Electric and SAP, but core supply chain companies such as Micron, SK Hynix, Samsung, Broadcom, Marvell, Intel and AMD are all absent from the main stage. This reduces the conference's market heat to some extent, but it also gives peripheral players such as Intel room to leverage the occasion.
Wall Street's Real Question: Evidence of Commercialization, Not Technology Demos
Overall, the core value of this GTC to the market lies in whether it can answer three of investors' most pressing questions: whether enterprise AI, autonomous AI and AI factories can generate additional infrastructure demand; whether AI Agents can create measurable commercial value rather than remaining at the pilot stage; and whether robotics and industrial AI can open up new growth paths beyond data centers.
GTC is an important forum for technology and industry development, but that does not mean every advance showcased at the conference will immediately translate into revenue or stock price catalysts. For today's market, the most weighty part of Jensen Huang's speech will not be the performance curve of the next-generation chip, but whether the AI industry can present a clearer commercialization path and evidence of investment returns while continuing to expand.
Comments