AI compute demand set to exceed supply; US-China policy moves may bifurcate global market: Report
New Delhi, July 29
Demand for computing capacity is expected to significantly outstrip supply for years to come, while the increasing likelihood of AI policy interventions in both the US and China could further deepen the bifurcation of the global AI market, according to a Morgan Stanley report.
Noting the sharp decline in AI infrastructure stocks in recent weeks, Morgan Stanley believes much of the weakness is due to market and technical factors rather than a deterioration in fundamentals.
"To be clear, we believe a meaningful driver of weakness has been technical rather than fundamental," it noted.
It further noted, while enterprise limits on token spending are unlikely to materially constrain AI revenue. Morgan Stanley highlighted, companies may impose limits on employees' use of computing resources, as seen in a few high-profile cases. However, the data suggests this is unlikely to be a significant problem.
As per the report, the median monthly token spending by enterprise users is currently very low, below USD 11. At the same time, the economics of AI use remain highly attractive, with AI potentially delivering around USD 55 in labour cost savings for every USD 2-3 spent on tokens.
"Enterprise failure to adopt the most advantageous AI capabilities will lead to large competitive disadvantages, a dynamic we expect to become increasingly clear over time," it noted.
Morgan Stanley however flagged, "Chinese open-weight models may become a greater competitive threat to American frontier LLMs, which in turn could drive lower spend on compute to train LLMs."
It further noted that both large language models (LLMs) and more efficient models generate strong returns on the underlying AI infrastructure.
At the same time, Enterprise AI use cases are also highly cost-effective, with token costs accounting for only a small fraction of the benefits generated.
"We see an increasing probability of both US and Chinese AI policy intervention, which could lead to a greater bifurcation of the global market," it said.
— ANI
Reader Comments
The data on token spending is eye-opening! Under $11 per month on average, but saving $55 in labor costs. That's a 18x return! 🚀 No wonder companies are going all in. But the supply crunch is real—we saw what happened with GPU shortages last year. India needs to aggressively invest in domestic chip manufacturing and cloud infrastructure. Relying on foreign compute is a strategic risk.
I'm a bit skeptical about the 'enterprise failure to adopt' argument. Yes, early movers will have an advantage, but AI is advancing so fast that last year's model is obsolete. The real risk might be over-investing in proprietary LLMs when open-weight models from China are catching up fast. The bifurcation the report mentions could actually benefit open-source ecosystems.
Arre yaar, this is exactly what I was telling my team last month! The compute crunch is coming, and we need to start planning our AI workloads accordingly. But I'm worried about the policy angle—if the US and China start restricting exports of chips and models, countries like India will be squeezed. We should be building our own foundational models using Indian languages and data. Imagine a desi ChatGPT that speaks Hindi, Tamil, Telugu fluently! 😄
The labor cost savings calculation seems too simplistic. It doesn't account for the cost of retraining staff, integrating AI into existing workflows, or the risk of hallucination in critical decisions. And $55 savings for every $2-3 spent? That's assuming perfect implementation. In reality, many companies will struggle with change management. Still, the direction is clear—AI is a net positive for productivity.
R