George Colony Says AI Investment Will Continue, Despite a Market Selloff That He Calls an Overreaction

Asked what is driving the sharp selloff, George Colony urged viewers to keep the situation in perspective. He said Forrester believes tech expenditure will reach about $6.1 trillion next year, up around 8%. In his words, tech expenditure is not slowing down; it is increasing.

He added that AI will account for about 30% of all tech expenditure by 2030. Calling AI the most important technology invented in his lifetime, he said the market should calm down.

The host asked whether a post-election political shift could slow the data center build-out, pointing to Texas, where many data centers were expected to come online. He noted that the Texas governor partly campaigned on data centers and that local legislative bodies appeared to be shifting toward the Democrats.

Colony said he expects the data centers to be built, just in different states. He described a migration of capital: definitely not in Texas now, probably not in Ohio, and not in Georgia. He called the technology too powerful and too useful to not happen.

Colony distinguished public models, the large frontier models such as those from Gemini and Anthropic, from private models. Private models are built by banks, insurance companies and car companies, and Colony said Forrester believes 70% of AI revenue will come from them.

He argued that capital is currently misallocated toward public models and that much more will flow to private models, where most of the activity will be in the future.

Asked about the bull and bear cases for a slowdown, Colony gave a nuanced answer. He said private models built inside companies such as JPMorgan or Bank of America would lead to a repopulation of data centers, as computing moves back into them to run those models.

He said this could take some pressure off the build-out by hyperscalers and outside data centers, and could amount to a moderation. The host noted the market had been rallying on the view that hyperscalers were well placed, because they already have data centers and have dealt with local communities.

Colony agreed in part but added a caveat. Forrester's view is that cloud computing data centers built over the last ten years are being converted to what it calls AI computing.

When that happens, demand for the old systems running in those data centers will fall, which means some moderation in demand for data centers.

The discussion turned to which chips will run the data centers. Colony said powerful companies do not like to be held hostage, and that NVIDIA is holding them hostage now. He said Forrester sees training systems being built by AWS and by Microsoft, and that other chips besides NVIDIA will appear in those data centers.

He said this does not mean NVIDIA will crash, only that its growth may not be as optimistic as people imagine.

The host noted that NVIDIA once seemed to be the stock market itself, with earnings events drawing watch parties, and that its next report is due November 17. Colony said moderation is likely because of the new chips other companies will produce.

The host said that the S&P 500 was down significantly, about 3.1%, and that the Nasdaq was down more than 5%, with individual names hit hard.

The host asked at what point the selloff might become more than a stock story, for example if the cost of capital keeps rising and spreads widen. Banks, Apollo and other financial firms were also being hit hard, and a pullback in lending could make the problem more fundamental.

Colony said the industry had over-capitalized the large-model part of the world, with too much capital flowing there. He said private models will also need more capital, so funding is rebalancing.

He said this will force models not to get bigger. The industry will move toward distilled models in the billion-parameter range rather than the trillion-parameter range, and some Chinese models are already following this trend.

The host asked whether the AI trade is entering a phase where investors are more selective and NVIDIA matters less to the ecosystem. Colony agreed, saying there is too much capital in one direction, that the capital needs closer examination, and that some of it is moving to private models.

The host cited a story by the Bloomberg team reporting that Microsoft is souring on neo clouds, citing delays, regulatory snags and high costs. Colony said he was with Microsoft on this and that there are cracks in the structure, which he sees as another indicator.

The host corrected the word shattering to cracks and said the columns are not falling yet, asking whether there is a risk of contagion. Colony answered that the technology is simply too powerful.

Colony framed the debate as a difference in time horizon. He said he is thinking of a five-year timeframe while others are thinking of a shorter one, calling AI an extraordinary technology that will be very good for society, businesses and people.

He said the adoption will happen, but it will not happen along a straight line.

The host noted that Forrester's research says the benefits of AI have not yet come through in company bottom lines. Companies are using it, but converting that use into revenue has not happened yet. The host asked when that needs to happen, and Colony said five to ten years out, calling the technology revolutionary.

The host asked whether there will be a moment when companies lose patience and pull back on spending. Colony answered that a large language model is not a business model, and that OpenAI and Anthropic have not yet worked out their business models.

The host asked who captures the most value in a new phase of the AI trade: hyperscalers, chip makers, neo clouds or infrastructure providers. Colony recalled that during the early web build-out, Cisco was the highest market cap company in the world because it built the pipes.

He said NVIDIA is today's highest market cap company because it builds the chips, but that this is not sustainable. He said these are early days of the wave: NVIDIA will ultimately make money, but not at these levels, and value is shifting to software.

The host noted that software stocks are also selling off and asked which software businesses will survive. Colony said software companies are under pressure and that there is roughly a 50/50 chance that many will not survive the change, because the industry is moving to what Forrester calls AI computing, a new way to build and run software.

He said survival depends on a company's ability to shift from structured to unstructured data, and from deterministic to probabilistic data. He added that the next three years will be critical for that transformation.

The host asked where investors can hide in a market where almost everything is tied to AI, citing Caterpillar, which is now also traded as an AI trade. Colony answered that there will be AI in every device, including generative AI and what he called genetic AI.

Colony said AI will be embedded in every device within ten years, and that the timeframe could be shortened to five. He said AI will be everywhere and should be everywhere, because it is a technology that can improve everything people use.