Jeff Tatarchuk, CEO of TensorWave, hosted an event last year titled "Beyond CUDA." Is NVIDIA's CUDA ecosystem showing vulnerabilities? Jeff Tatarchuk has always been proactive in challenging industry giants, specifically NVIDIA. Tatarchuk has strong motivation to contest this chipmaker's dominance—he runs the startup TensorWave, which specializes in renting out AI chips from NVIDIA's competitor, Advanced Micro Devices (AMD). During last year's NVIDIA GTC conference, Tatarchuk held an event called "Beyond CUDA" in San Jose, California, attracting hundreds of participants who eagerly discussed alternatives to NVIDIA's software stack. This year, the event returns but with a noticeably moderated tone. Tatarchuk has renamed the gathering the "Beyond Summit," scheduled to commence on Wednesday, three weeks after the conclusion of NVIDIA's 2026 GTC event in San Jose. This change, however, was not entirely his preference. He indicated that some potential sponsors and attendees expressed concerns about participating in an event so explicitly positioned against NVIDIA. In an industry where many companies still depend on NVIDIA hardware, even symbolic opposition could pose risks, he noted. Nevertheless, Tatarchuk has mixed feelings about the rebranding. On one hand, he acknowledges the industry pressures and has managed to organize a larger event this year. On the other hand, the original name more directly addressed the core issue: NVIDIA's chips dominate the AI computing sector, with CUDA serving as a key barrier to entry. This software ecosystem provides the comprehensive support needed for its hardware to operate efficiently. Even the event's venue arrangements reflect NVIDIA's industry influence. Tatarchuk mentioned that his initial plan was to host the event again in San Jose, but he discovered that the previous venue had been booked by NVIDIA for several years into the future. Consequently, this year's event was rescheduled to take place weeks after the GTC conference and relocated to San Francisco. Despite these changes, the core purpose of the gathering remains unchanged. Tatarchuk stated, "Outside of NVIDIA's closed ecosystem, the industry is nurturing new technological advancements." He reported that an increasing number of AI labs are beginning to conduct large-scale model training on AMD chips—a topic rarely discussed openly until recently. For instance, both OpenAI and Meta Platforms have recently entered into significant agreements with AMD to procure its chips for AI computing tasks. Tatarchuk is increasingly vocal about the industry reaching a potential inflection point for breaking NVIDIA's monopoly. He believes that, for the first time in years, customers are willing to consider abandoning CUDA in favor of alternative solutions. "AI labs are starting large-scale training on AMD hardware, something that was seldom mentioned before," he said. "There are numerous technically proficient companies that have already moved beyond relying on CUDA." Simultaneously, a wave of startups is emerging, gradually eroding CUDA's dominance. These companies focus on developing compilers, kernels, optimization layers, and other components of the software stack. Some of these firms have been featured in the Global Startup 50 rankings for 2024 and 2025. Tatarchuk confirmed that most of these companies will participate in TensorWave's upcoming event.
The "Coachella of Compute" While the Coachella music festival is scheduled for this weekend, Stanford University recently hosted a "compute version" of Coachella. A new undergraduate course on AI infrastructure at the university has been described as the "Coachella of compute." The course is fully enrolled, and its lineup of guest speakers rivals that of top-tier industry conferences. While attendees won't find Justin Bieber or Sabrina Carpenter, for those interested in chips, computing power, and large models, the event offers substantial value. Scheduled speakers for the semester include NVIDIA's Jensen Huang, AMD's Lisa Su, OpenAI's Sam Altman, Microsoft's Satya Nadella, and OpenAI co-founder Andrej Karpathy. This star-studded roster is partly due to the course instructors: Anjney Midha, an early investor in Anthropic and former partner at Andreessen Horowitz, and Michael Abbott, former engineering lead at Apple who helped drive the company's efforts to build its own cloud services. Both are now involved with the AMP compute investment initiative, which aims to lower the barriers for companies to access computing infrastructure. It is likely that the guest speakers are also keen to engage with the students. The course project requires students to attempt cutting-edge artificial intelligence research over the next 10 weeks, utilizing the limited computing resources provided by AMP.
Comments