The world of artificial intelligence (AI) is abuzz with activity, and the demand for its infrastructure is a hot topic. While some may question the sustainability of this demand, industry leaders are painting a different picture.
The Unstoppable Demand for AI
Pat Gelsinger, a prominent figure in the tech industry, describes AI demand as "almost unlimited." He believes the potential economic value of increased intelligence is infinite, cutting across all imaginable industries. This perspective is shared by several AI executives, who assert that the demand for AI compute and infrastructure remains strong, despite recent market volatility.
Supply Constraints and Excess Capacity
The market has seen some intriguing moves, such as Meta's decision to sell its excess AI computing capacity, raising questions about overcapacity. Similarly, Elon Musk's xAI rented out its excess capacity, adding to the debate. However, these moves seem to be unique cases, as Marc Boroditsky, Chief Revenue Officer at Nebius, emphasizes that the demand for compute far exceeds available capacity. Andrew Feldman, CEO of Cerebras Systems, echoes this sentiment, stating that the industry is short on data centers and other compute inputs.
The AI Infrastructure Buildout
The buildout of AI data centers is a key focus for investors, with chip stocks experiencing a blistering rally over the past year. Companies like Nebius, Cerebras, and Lumentum are at the forefront of this infrastructure development, with Lumentum's stock up an impressive 600% in the last 12 months. Michael Hurlston, CEO of Lumentum, highlights the company's efforts to meet the demand they see out five years, a testament to the sustained interest in AI infrastructure.
Enterprise Spending and 'Valuemaxxing'
A shift is occurring in enterprise spending on AI, moving away from 'tokenmaxxing' towards a more rationalized approach. Companies are now focusing on the return on investment (ROI) from AI, especially as frontier models from companies like OpenAI and Anthropic remain expensive compared to open-source offerings. Nebius' Boroditsky suggests that 'tokenmaxxing' is only worthwhile if it leads to ROI, and enterprises should aim for 'valuemaxxing' by applying AI to create value that justifies the spending. This shift towards more rational spending is a familiar pattern seen in previous tech cycles, according to Boroditsky.
The Future of AI Models
The AI landscape is diverse, with a range of models offering different capabilities. Andrew Feldman predicts that in the future, specific models will be used for particular situations. He compares this to choosing the right vehicle for a task, suggesting that certain workloads will migrate to different types of compute based on their complexity. This evolution in AI deployment is an exciting prospect, as it indicates a more sophisticated and efficient use of AI resources.
Conclusion
The AI industry is experiencing a period of intense growth and development, with demand for its infrastructure showing no signs of slowing down. While market volatility and unique cases like Meta and xAI may raise questions, the overall picture is one of sustained interest and investment. As enterprises adopt a more rationalized approach to AI spending, the focus on ROI and valuemaxxing will likely drive further innovation and efficiency in the use of AI models and infrastructure. The future of AI is bright, and its potential economic value remains almost limitless.
