Lets Talk

AI started to become more mainstream in 2023; across 2024, it seemed as if everything was now using AI. Generative AI seems ubiquitous in business now. That said, there has been a significant lag in terms of businesses recognising the risks that AI poses to their businesses, and I don’t mean in terms of the “I’ll be back” kind. The truth is that businesses were never prepared for AI, meaning corporate steps were never taken to protect themselves from the risks posed.

Data Leakage is the most basic and ever-present risk posed by generative AI tools. Corporate sensitive data is directly or indirectly shared with the AI tools meant to help. Information used to train an AI can be extracted from that AI, meaning what you share can be the same as sharing it publicly. Subsequent generations of AI are also trained using new and old data, leading to additional data privacy questions.

We know that AI development is on the increase as more providers, including Microsoft, seek to secure nuclear power plants to support their needs. The concern should be around AI’s revenue generation capabilities to “Free” providers, Meta and Google, for example. The message is clear: AI is here, and it can be used to generate revenue for organisations, going beyond historic Data Mining capabilities of the past. We must remember that AI cannot be programmed in the same way that a traditional computer program would be written; rather, it must “learn” from experiences in much the same way as a baby must learn from experiences with a central set of protocols. This means that generative AI is, at its heart, a probability engine tasked by protocols to test what value should come next. So, in the case of writing a paragraph with AI, it has been tasked and then trained to evaluate what word to put next in a sentence and if that should be the end of the sentence/paragraph.

No matter what you think of how the free providers generate their revenue, AI will be incredibly damaging for them as their “unique” position is threatened by organisations being able to deploy their own more targeted AI-driven applications to better enable them to steer their own futures, will their revenue streams be threatened? For those organisations that depend on more traditional revenue streams like Microsoft and their Azure platform, companies are seeing cloud spending increasing; storage is remaining stable and relatively stable; however, the requirements and so the cost of compute is rising. This increase in cloud cost is becoming a concern for an increasing number of businesses. Declouding has been a fringe discussion topic for many years and has no doubt been slightly masked as a topic by the disruptions of 2020 and 2021. Over 50% of companies have reported bringing at least one workflow back on-premise that had previously been deployed.

The maturity of Generative AI, combined with the progress in terms of hardware capabilities that businesses can deploy themselves at lower costs than ever before, makes hosting their own private AI service possible. Mini-compute units or lower-cost rack mounts are replacing the need to deploy costly server solutions. For reference, an RTX 3060 provides an excellent value proposition for a business looking to host their own AI. Its relatively low cost and relatively high RAM mean it can internally process a reasonably large AI model without needing external storage.

Geordie Consulting, as a Cloud Consultancy, has been monitoring the situation across 2024. However, the change from Power BI to Fabric needs to be observed. In the 12 months since Fabric was initially launched. Multiple amazing new features have been added, many of which could be described as too good to be true. It is worth noting that we still cannot recommend Fabric to our customers as the pricing model still proves too vague, and transitioning customers from the O365 flat cost into the Azure variable cost model needs to be crystal clear; the current suite, with all its changes and EA features, makes the formulation of a business case difficult. This brings us to the natural extension of self-hosted AI: an organisation can now use their internally hosted AI solution to help develop its own internal data processing tools using existing tools, or Kubernetes clusters built using low-cost hardware. Geordie Consulting has always retained onsite backups of our devices and also for the raw content from our YouTube channel and business – Geordie Intelligence. In 2025 we are expecting to investigate the capability of self-hosted AI and other data management workloads.

Reference

Why declouding is taking place

Infosys Consulting | Understanding the growing trend of including

Gartner Hype Cycle AI 2024

Hype Cycle for Artificial Intelligence 2024 | Gartner