Global Stock Selloff: AI Investment Will Continue, George Colony Says
AI investment is likely to continue, but shift from public-model infrastructure toward company-specific models, alternative chips, and software built for AI computing.
- Expect data-center investment to migrate across states and moderate as companies repurpose existing facilities for AI workloads.
- Watch for more private company models, smaller distilled models, and non-NVIDIA chips as AI investment becomes more selective.
- Software firms must adapt to unstructured, probabilistic data and AI computing; the speaker warns many may not survive the transition.
The interview gives a market thesis and forecasts, but no detailed demonstrations or practical walkthroughs. · Best for: Technology and investment professionals tracking AI infrastructure, models, and software.
George Colony argues that AI investment will persist despite market volatility, but capital may shift away from large public models and concentrated infrastructure spending. He expects more private models inside companies, some data-center demand moderation, and a longer-term shift in value from chips toward software.
AI spending and data-center siting0:00
- Colony cites Forrester’s view that overall technology spending will keep growing and AI will take a larger share over time.
- If local opposition slows data-center projects in some states, he expects investment to move elsewhere rather than stop.
- He says private models built by companies could account for much of future AI revenue.
Private models and infrastructure demand1:40
- Banks, insurers, and car companies are building models for their own use.
- Running those models inside companies could bring computing back into corporate data centers and ease pressure on hyperscaler data-center expansion.
- Converting existing systems to AI computing may reduce demand for older workloads.
Chips, capital, and model size3:05
- Large companies are developing chips such as AWS Trainium and Microsoft Maia to reduce reliance on NVIDIA.
- Colony expects this competition to moderate NVIDIA’s growth, not cause a crash.
- He expects capital to shift toward smaller, distilled models rather than ever-larger models.
Cloud risks and long-term outlook5:19
- Colony sees signs of strain in the AI infrastructure market, including concerns about delays, regulation, and costs at neo-cloud providers.
- He frames the transition as uneven rather than a straight-line expansion, while maintaining that AI adoption will continue over a five-year horizon.
Business models and who captures value6:39
- Colony says a large language model is not itself a business model; he sees OpenAI and Anthropic potentially supplying technology that underlies companies’ private models.
- He expects value to shift over time from chip makers toward software, comparing the current stage to the early web infrastructure build-out.
Software’s adaptation challenge9:00
- Software companies face pressure as computing and software development change.
- Colony says survivors will need to move from structured, deterministic data toward unstructured, probabilistic data.
- He expects AI to become embedded broadly across devices and industries.
Tools & resources mentioned
- Forrester Research — forecasts and analysis cited by Colony.
- Gemini and Anthropic — examples of public, frontier models.
- AWS Trainium and Microsoft Maia — examples of companies’ alternative AI chips.
- NVIDIA — chip provider discussed in relation to AI infrastructure concentration.
- OpenAI and Anthropic — model providers discussed as potential suppliers for private-model deployments.
- Oracle and SAP — comparison for the possible future role of AI model providers in enterprise systems.
Notes are AI-generated from the video's transcript and may miss nuance.