2026年TMT行业预测：差距缩小，但依然存在

Deloitte predicts 2026 will see the gap between the promise and reality of AI narrow, asfurther movements towards getting it to scale are made ARTICLE9-MIN READ17 NOVEMBER 2025Deloitte Center for Technology Media & Telecommunications••• In 2026, Deloitte predicts the roar around artificial intelligence will be getting quieter—and smarter—asthe sometimes unglamorous, high-impact work of making AI usable at scale continues to get underway.The gap between promise and reality will narrow but not disappear: Progress will come less fromheadline-grabbing new models and more from fundamentals. That more practical focus matters becausetech, media, and telecom’s growing importance is not just about chips and code—it’s about how everyother industry uses those TMT capabilities for its own growth, efficiencies, and innovation. AI helps drive cross-industry transformation In 2026 and beyond, it looks like we have moved from “software is eating the world” to “TMT is eatingthe world,” led by AI—especially agentic AI. In the United States, spending on AI data centers currently accounts for almost all gross domestic product growth in the first half of the year.1In 2008, about 19% ofthe S&P 500’s market value was in tech stocks, and TMT now makes up almost 53% of market capitalization.2Things could change, but at this rate, TMT is poised to not merely become larger thanany other industry, but larger than all other industriescombined—both in terms of value andcontribution to economic growth. One reason for that is that other industries use TMT—tech andtelecom specifically—to power their own AI innovations, and TMT happens to be the hardware,software, and services provider in the AI gold rush. That said, other industries play critical roles. In both TMT Predictions 2025 and again in 2026, we havepulled in specialists from other Deloitte research centers and industries: energy, mining, and chemicals,manufacturing and construction, defense and aerospace, government and public services, and life sciencesand health care. It takes some serious cross-industry collaboration to properly predict generative AI andagentic AI trends and implications. Of our 13 topics for 2026, over half follow an AI theme. At a high level, we’re seeing a narrative aroundmaking AI scale. New foundational models, or even shiny new enterprise agentic applications, continue toimpress—but translating those beyond pilots and trials requires work that’s typically considered lessexciting, like data hygiene, integration into existing workflows, governance, new pricing models, andregulatory compliance. Those may be less glamorous than press releases about AI beating humans on ascience test, but they will likely be more useful in the near term. Gen AI and agentic AI are driving a lot of things that are very much here and now, but we also have aneye on the future. Although Deloitte predicts that robotics and drones will be slow but steady growersover the next year or two, the emergence of “physical AI” models is poised to transform both industrieswith massive acceleration in growth and usefulness. Meanwhile, newer forms of media, like short-form vertical video series, appear to be crossing over from Asia to the rest of the world. And while the spreadof gen AI–created images on social media may be exciting, it may also stimulate regulation. A quick look at our 13 topics for 2026Gen AI inside existing search engines overtakes standalonegen AI Gen AI, possibly one of the most consequential technologies of our decade, may see its user base widenfaster through its incorporation into existing mainstream digital applications than through its usage on astandalone basis Deloitte predicts that in 2026 and beyond, more people will use gen AI when it’s embedded within anexisting application—such as a search engine—than those using a standalone gen AI tool. In terms ofdaily use, accessing gen AI within a search engine (when a search yields a synthesis of results) will be300% more common than using any standalone gen AI tool. Standalone gen AI may require skill inprompt engineering and persistence, whereas passive gen AI is less overt and the experience more familiar;as such, demand is greater because it’s more accessible. Going forward, standalone gen AI app ownerswill likely face a choice between embedding their tools’ capabilities within another application orremaining a standalone interface. Why AI’s next phase will likely demand more computationalpower, not less The world is moving from just training gen AI models to using them at scale. Many believe this meansmore consumer edge computing and less data center computing. Neither is likely to happen in 2026. Deloitte predicts that “inference”—running AI models—will account for two-thirds of all AI computingpower by 2026. Despite forecasts to the contrary, most inference will still take place in new data centersworth nearly half a trillion dollars and in on-premises enterprise servers using costly, power-intensive