Supercomputer Called Stargate in the Works
Microsoft and OpenAI have banded together to embark on an innovative project: creating an artificial intelligence supercomputer named “Stargate.” According to a report by The Information released on Friday, this ambitious endeavor could come with a staggering price tag of up to $100 billion. The estimate, cited by a source who conversed with Sam Altman and another who reviewed initial cost projections, underscores the enormity of this undertaking, potentially surpassing the costs of even the most significant data centers. Stargate is slated to materialize by 2028 if everything goes according to plan.
The vision for Stargate extends beyond a mere supercomputer; it represents the pinnacle of a broader initiative by Microsoft and OpenAI to construct a series of supercomputers over six years. Stargate is a pivotal component of this grand plan’s fifth and final phase, currently positioned within the third development phase. A significant portion of the estimated expenses for these forthcoming phases stems from the procurement of AI chips, which constitute a crucial component of the infrastructure for these sophisticated computing systems.
However, a significant bottleneck in the AI development pipeline persists—a shortage of graphics processing units (GPUs), which play a vital role in processing data for AI models. The exponential growth of generative AI has fueled an unprecedented demand for these chips, exacerbating supply constraints. Compounding this challenge is the near-monopoly held by Nvidia, a leading GPU manufacturer renowned for producing top-tier chips tailored for AI applications. As demand surges, Nvidia faces mounting pressure to fulfill orders while grappling with escalating production costs.
Nvidia’s CEO, Jensen Huang, recently disclosed at the company’s GTC 2024 event that their latest AI chip, the Blackwell B200, commands a hefty price ranging from $30,000 to $40,000 per unit. With over $10 billion invested in its research and development, this cutting-edge chip boasts unparalleled performance, setting a new standard in the market.
In response to Nvidia’s dominance, a consortium of tech giants, including Intel, Qualcomm, Google Cloud, Arm, and Samsung, has united under the “United Acceleration Foundation.” Their collective aim is to challenge Nvidia’s supremacy in both software and hardware within the AI landscape by establishing an open standard accelerator programming model. This concerted effort marks a significant step towards fostering innovation and competition in the AI sector, paving the way for a more diverse and accessible ecosystem.