In its intriguing report, consultancy firm Deloitte predicts that the tech, media and telecoms (TMT) sector “is on the verge of a significant leap forward” which will be “largely powered by rapid generative AI adoption.”
However, its research suggests that 2025 is unlikely to be the year of “accelerated development”, because of the discrepancies between AI aspirations and the reality on the ground.
Some of the conclusions in the report are applicable to this area of technology more broadly and can be used to interpret current and short-term developments in areas away from media and telecoms.
In particular the following hurdles facing more rapid and wider AI adoption can be cited as more broadly relevant:
| Infrastructure and monetization | Investment in AI currently 10x or even higher than the return, with this gap widening |
| Data center electricity and sustainability | Lack of grid capacity is being accentuated by sustainability targets |
| Gender | Women are less likely to use gen AI tools, which is hampering adoption and growth |
| Deepfakes | The rapid proliferation of deep fakes is reducing trust in the technology |
| Large organization adoption (studio in the report) | Expectations and reality around adoption by large corporate entities different as result of these exercising caution, particularly around potential intellectual property issues |
| Autonomy | The pace of improvement is accelerating, but the report makes clear that most autonomous systems are at the POC stage, implying that consistent and reliable task completion continuous to be a challenge |
| Cloud spending | Promised cost savings of using cloud systems are being diluted by the decentralized and poorly controlled spend |
Another interesting prediction in the report is connected with on-prem AI hosting. According to the report 61% of companies are “investing more in their own hardware” because “owning their own servers gives companies a more private, secure, flexible, and possibly cheaper IT environment for AI.”
Data centers
The report predicts that “about half of the enterprises worldwide will add AI data center infrastructure on premises, primarily to protect their IP and sensitive data, comply with data sovereignty or other regulations, and save costs.”
Coupled with some of the enterprise concerns above, including those around practical implementation challenges and cost, it seems possible that enterprise AI adoption may well look very different and be a lot more fragmented than that based on the valuations of the mainstream AI models out there.
In the short term some of the challenges faced by the AI ecosystem potentially have significant consequences for cyber security in particular because AI is already being widely used by cyber criminals and other nefarious actors.
According to a McKinsey report published in November, “AI-enhanced advances also make it easier to exploit a growing attack surface, in turn introducing new risk exposure” as the annual number of cyberattacks using generative AI accelerates rapidly.
The report points out that the number of phishing sites detected in 2024 stands at over five million and this figure has grown 138% (from just over two million) since the launch of Chat GPT in 2022.
So while an AI gap year in the TMT sector does not sound ominous, a gap year in terms of AI developments in cybersecurity may well be a different thing entirely.
