Industry executives and experts share their predictions for 2025. Read them in this 17th annual VMblog.com series exclusive.
By Lalit Ahuja, CTO, GridGain
2024 was a pivotal year in tech as GenAI transformed
business models and sent engineers in nearly every industry back to the drawing
board on how best to take advantage of it. 2025 may be very different, though.
Enterprises must now contend with nefarious use of GenAI and put checks and
balances in place to secure their information and prevent identity thieves and
scammers from putting their business, confidential information, and reputation
at risk. Here are my three 2025 predictions for how we make real and practical
progress toward these goals.
1. The rise of the
new data ecosystem
The proliferation of AI has only put a bigger spotlight on
data and the speed of data processing.
The thirst for global information, characterized by the ever-greater
amount and scope of data, has created complex problems for enterprises that
must process all this data at tremendous speeds. As a result, these enterprises
must identify and implement new technology solutions.
The challenge, however, is that data today is
multidimensional, requiring multiple data sources, data formats, data lifecycle
stages, and data processing strategies. An easily relatable example of this
complex set of demands is credit card fraud detection, which includes (at
least) streaming data to process every incoming transaction, historical data
processing to understand customer usage patterns, intelligent model execution
based on combining the historical and streaming data, intelligent analysis of
the data to support updating AI/ML models, and low latency to enable execution
within milliseconds.
In 2025, I expect vendors to start delivering complete
solutions capable of rising to this challenge by offering at least a basic
level of support for:
- Combining processing of both data in motion and
data at rest - Virtually holding data from multiple sources
while offering persistent storage – and ensuring data integrity, security,
durability, and availability - Reducing latency caused by movement across
networks - Enabling the execution of advanced AI/ML – or
any complex computation on large datasets – at extremely high speeds
I will also not be surprised if technology providers look at
strategic partnerships or even acquisitions to bring such comprehensive,
tightly integrated solutions to market.
2. Reality reins in
GenAI euphoria
Everyone who predicted 2024 would be the year of GenAI was
right. We saw tremendous technical progress and even more hype. But now what?
2025 will be the year when reality sets in, when the enthusiasm for
implementing GenAI use cases will at least come with a dose of pragmatism.
Borrowing from one of my childhood favorite superheroes – with more
information comes more responsibility. As enterprises build and deploy these
new data ecosystems and use them for GenAI applications, they will need
guidelines in place on the use of the data. This is now more obvious than ever
before. Enterprises are already seeing
challenges with the use of GenAI related to security, governance, and
operationalization. Hallucinations and bias are well documented AI challenges,
but the problem runs deeper. Every time employees ask a question of a public
LLM (e.g. ChatGPT or CoPilot), they may be sharing proprietary or sensitive
information. Even if they aren’t putting financial or product information into
a prompt, they may still be providing contextual information, such as their
intent or competitive concerns, enabling the LLM to triangulate and construct a
revealing or possibly damaging profile that becomes available to investors,
competitors, and potential customers.
Imagine a CEO asking ChatGPT to draft an email to all
employees discussing the impact of the pending sale of the company – weeks
before the planned public announcement. That information is now discoverable
and could easily be publicly revealed.
In 2025, as companies look to operationalize GenAI, they
will have to confront complex AI governance issues, and many will be forced to
slow their GenAI plans. They will put AI governance councils in place, and they
will look to emerging regulations for guidance. In the meantime, they will
choose – or be compelled – to place restrictions on the use of GenAI, or at
least institute corporate training on the use of GenAI for all employees with
access to sensitive information. Private deployment of these LLMs may also see
an uptick.
By 2026, AI governance will be a standard component of
InfoSec practices, and enthusiasm for implementing GenAI use cases will likely
soar again – tempered by the recognition that it must be done right. In the
meantime, it should not surprise anyone if at least one high-profile leak of
sensitive information due to the inappropriate use of a public LLM occurs in
the next few months. Such an incident will slow down the implementation of
planned use cases and further intensify the focus on GenAI governance.
3. Recency in RAG
becomes a thing
Despite all the hype around what GenAI can do (much of it
deserved) and the power of an enterprise data ecosystem behind it, we are
missing a key capability necessary to make some of the most transformative AI
use cases possible: recency. It’s recency that’s required for investment
portfolio management based on up-to-the-second risk assessments. Recency is key
to enabling retailers to analyze in real-time an online customer’s activities,
including their search history, browser tabs, and cookies, to recommend
products that meet the customers unique needs and preference – even if the
customer doesn’t really know what they are.
Retrieval-augmented generation (RAG) is obviously a critical
processing mechanism required for these use cases. It enables businesses – the
investment firm, the retailer – to feed proprietary data to the LLM to make the
model more accurate and deliver personalized responses. But RAG use cases
require accessing and moving data across network silos, preprocessing such
cross-enterprise data, and analyzing it – and all of these steps introduce
latency.
This is where the need for technology to support recency
comes in. If RAG doesn’t have access to up-to-the-second data from diverse
sources in different locations, the benefits are limited. In 2025, demand for
recency in RAG will become a hot topic for businesses, analysts, and technology
vendors.
Conclusion
As excited as I am about new technology, I am even more
excited when that technology begins to deliver real value to businesses and
their customers and partners. That’s why I am cautiously optimistic about the
coming year and am looking forward to the technological advances that 2025 will
bring to the industry and society in general.
##
ABOUT THE AUTHOR
Lalit
Ahuja is the Chief Technology Officer for GridGain, a provider of a leading
unified real-time data platform, where he is responsible for the strategy,
execution and delivery of GridGain’s technology portfolio, while ensuring
customers gain the most value from their GridGain investments.
