{"text":[[{"start":6.64,"text":"That 2025 would be another banner year for artificial intelligence was not in doubt. But few realised just how much it would belong to the builders of AI rather than the users — or how that imbalance would start to perturb Wall Street. "}],[{"start":24.59,"text":"Investors were transfixed all year by the steady escalation in the amount of capital being thrown at AI data centres. Morgan Stanley had predicted spending would grow 20-25 per cent. By last month, that forecast had ratcheted up to 68 per cent, with the total projected to hit $470bn before jumping again to $620bn in 2026. Even that pales compared with the giant deals that were being signed in the second half of this year, led by the $1.4tn of new data centres that OpenAI has lined up."}],[{"start":63.230000000000004,"text":"For now, the companies leading the charge report that demand for all that new capacity is running comfortably ahead of supply. That makes this fundamentally different from the internet mania of the late 1990s, when all the new telecoms networks far exceeded requirements."}],[{"start":81.72,"text":"When supply does finally catch up, watch out below. But that may not happen soon, as bottlenecks — such as in the availability of electric power — threaten to slow progress. This should hand the advantage next year to companies that have already developed expertise in building and managing at huge scale."}],[{"start":103.06,"text":"A second notable part of this year’s boom for AI builders has been in the valuations of the companies that train the leading models. As the year dawned, this didn’t seem likely. They appeared to have hit a wall, with diminishing returns from throwing ever-larger amounts of data and computing power at the problem. At the same time, the ultra-low training costs claimed by Chinese company DeepSeek threatened a disruptive new wave of open-source AI."}],[{"start":134.46,"text":"In the event, the leading-edge models kept getting better, as the focus shifted to post-training techniques, and the world hasn’t swung to open source. The results have been electrifying. A little more than a year ago, OpenAI had just completed a funding round that valued it at $157bn. Now, it is reported to be stretching for a valuation of more than $800bn. Others have been swept along: Anthropic is targeting a valuation of more than $300bn (its final round in 2024 was at $18bn), while x. AI believes it is worth $230bn (up from $50bn)."}],[{"start":182.16000000000003,"text":"How many of these companies will ultimately be needed, or what their long-term economics will look like, is still a matter of conjecture. Depending on where you stand, these are either set to become the dominant platforms of the AI era, or they will struggle for differentiation and face commoditisation. The answer to that question is no clearer now than it was a year ago."}],[{"start":206.14000000000001,"text":"One important consideration in all this is the price of the tokens that are the model companies’ main output. Prices continued to collapse all year. For just $1.10, users of Google’s Gemini 3 Flash, launched last week, can purchase the same amount of machine intelligence that would have cost $65 two and a half years ago with the then-new GPT-4, calculates Tomasz Tunguz, a venture capitalist. That fits with the annual 10-fold improvement in price/performance that Andreessen Horowitz says has become standard for the leading models."}],[{"start":248.33,"text":"Such severe price deflation could end up favouring companies that control more parts of the technology “stack”, from chips to data centres and applications — one reason why Alphabet has swung back into stock market favour. It should also drive the technology into the mainstream more quickly."}],[{"start":268.86,"text":"There were signs of usage taking off; revenue, less so. A year ago, 300mn people turned to ChatGPT at least once a week. The number has swelled to more than 800m. Coding assistants have become a mainstay in the software world. But 2025 didn’t bring any new consumer “killer apps” to power greater use of the technology, or suggest when business spending on generative AI would take off."}],[{"start":298.35,"text":"Against that backdrop, it wasn’t surprising that bubble talk became widespread. That has left the AI boom resting heavily on the shoulders of a handful of big tech companies with solid balance sheets and cash flow. Eight US tech companies, including the leading cloud and internet companies as well as chipmakers Nvidia and Broadcom, are now valued at $1tn or more; These companies gained a collective $4.7tn, extending a stock market run that has seen their value roughly triple since ChatGPT was launched, to $23tn."}],[{"start":339.18,"text":"So far, these tech mainstays have been up to the demands of the AI build-out. They generated combined free cash flow of about $300bn in the first nine months of this year — the same as the year before — despite a $100bn increase in their combined capital spending. But if the AI boom continues its blistering pace through 2026, the stresses could start to show."}],[{"start":374.97,"text":""}]],"url":"https://audio.ftcn.net.cn/album/a_1766729027_3861.mp3"}