2024 Trends To Watch Generative AI GenAI PDF v2

2024 Trends to Watch:
Generative AI (GenAI)
Eden Zoller, Chief Analyst, Applied Intelligence
Natalia Modjeska, Research Director, AI and Intelligent Automation
Lian Jye Su, Chief Analyst, Applied Intelligence
Alexander Harrowell Principal Analyst, Advanced Computing for AI
Andrew Brosnan, Principal Analyst, AI Applications in Life Sciences
askananalyst@omdia.com
Copyright © 2023. All rights reserved. Informa Tech, a trading division of Informa PLC
“GenAI can be transformative, but only if
built on solid data foundations, executed
with care, and applied to use cases where it
provides genuine value. Do
not treat GenAI as a quick fix or expect it to
make bad things good.”
Eden Zoller
Chief Analyst, Applied AI
Copyright © 2023. All rights reserved. Informa Tech, a trading division of Informa PLC
Trends to Watch 2024: Generative AI (GenAI) | December 2023
Contents
• Omdia view
• Key messages
• Key recommendations
• From scale to diversity: The revolution in open-source AI
• Input/output trumps compute for GenAI inference
• GenAI will trigger a fresh wave of investment in enterprise data quality, management, and governance
• An explosion in mass-market GenAI chatbots and turbo-charged voice assistants
• Implementation of GenAI in healthcare will ramp up in 2024
• GenAI developments in Asia & Oceania will be driven by the need for localization
• Appendix
Page 3 Copyright © 2023. All rights reserved. Informa Tech, a trading division of Informa PLC
Omdia view: brace for new waves of Global GenAI software revenue by industry: Top 15
innovation, disruption and anxiety 2028 2024 2023
11,992
Consumer 2,206
• GenAI has created a new paradigm. Foundation models (e.g., LLaMA, 8,324
Media & entertainment 2,067
PaLM, Cohere, GPT-4, Stable Diffusion, Anthropic’s Claude, BLOOM,
6,720
Baidu's ERNIE-ViLG 2) are proliferating and becoming more capable (e.g., Healthcare 1,251
multi-modal inputs and outputs) while acting as an adaptive base on which Business services 1,120
6,165
more specialized downstream models and tools can be built. This adaptability, Information technology 1,157
5,394
coupled with better availability of open-source models and the intuitive 4,921
Automotive 1,548
prompt interface, makes GenAI highly accessible and is firing innovation.
3,856
Financial services 844
• Enterprises are enthusiastic about GenAI. According to Omdia’s IT Enterprise 3,505
Insights survey (run in 2023), 13% of total respondents (5,185) have already Retail 618
implemented text-based GenAI solutions, 25% are in the process of Telecommunications 560
2,765
implementing it while just over a quarter are in test mode. Advertising 559
2,727
• Consumer usage of GenAI apps is strong and set to increase. Omdia’s 2023 Manufacturing 458
1,740
Consumer AI survey shows that among respondents that have adopted GenAI, Education 309
1,537
a quarter use applications once a day (10%) or more (15%), another quarter 1,145
Defense
use GenAI at least once a week, while 29% use it a couple of times per week. 184
Existing users anticipate their consumption will increase over the next 12 Legal 204
885
months, and to a significant degree for 37% of respondents. Energy 380

66
• Global GenAI software revenues are ramping up. Omdia estimates that 0 2,000 4,000 6,000 8,000 10,000 12,000 14,000
revenues will grow from $6.7bn in 2023 to $62.9bn in 2028. The consumer
domain will see the sharpest increase ($12bn by 2028) followed by media and Revenue ($m)
entertainment ($8.3bn by 2028) then healthcare ($6.7bn by 2028).
Source: Omdia, Artificial Intelligence Software Market Forecasts – 2H23 © 2023 Omdia
Omdia view (cont.) What worries or concerns you most about GenAI? (multiple choice)
It will displace job functions, put people out of

52%
work
• GenAI will continue to push boundaries. Certain GenAI players (e.g.,
OpenAI) are committed to developing ever more powerful AI models that It could be could be used to commit fraud 50%
have capabilities beyond any existing model and will push the boundaries
of AI to create what is being termed frontier models. These big, largely It will amplify misinformation/manipulation 48%
experimental, cutting-edge models will be able to do things existing models
cannot, and they will be highly prone to emergent capabilities and It threatens data privacy 44%
unpredictable behavior, which means frontier models can pose extreme
long-term risks. This makes them the subject of increasing concern and Negative impacts on education 39%
regulatory scrutiny. Over reliance on chatbots/virtual companions
37%
• Issues and harms posed by GenAI persist—and consumers are aware of disconnect us from the world
them. This is a long list and includes the way that GenAI outputs can It makes mistakes and/or makes things up 32%
infringe copyright, displace jobs, and disrupt education. GenAI models can
be prone to bias and discrimination, can make mistakes, be inaccurate, or Determining who is responsible when generative
31%
“hallucinate” (i.e., make things up). The increasingly convincing outputs AI causes harm
produced by GenAI open a new chapter in manipulation and It infringes copyright 26%
disinformation.
The visual and written outputs of generative AI
• A complex regulatory framework is built around GenAI. GenAI is evolving threaten the creative arts
20%
rapidly, and lawmakers are struggling to keep pace. They must contend
with known harms and issues from existing GenAI systems, as noted above, Negative impacts on the environment 18%
and consider mechanisms for addressing longer-term risks and uncertainty,
particularly those posed by frontier models. 0% 10% 20% 30% 40% 50% 60%
Note: n=2,302
Source: Omdia’s Consumer AI Survey: 2023 © 2023 Omdia
Key messages
1 2 3 4
Despite progress, the From scale to diversity: the Input/output trumps compute New wave of investment in data
governance framework for GenAI revolution in open-source AI for GenAI inference quality, governance, and
will be highly uneven management
The European Union’s (EU) AI Act 2023 saw the launch of huge, Serving up AI inference in As the initial exuberance around
will not be implemented before powerful GenAI models (e.g., production is becoming more GenAI subsides and enterprises
2025, leaving a governance void GPT-4), but these have been important, leading to greater dig deeper into the tech, many
while GenAI is developing so fast eclipsed by an explosion of interest in which processors to will realize just how foundational
that new issues will emerge. The innovation in the 5–50bn use for inference. AI training data is for AI. Those not
US presidential elections in 2024 parameter model size unleashed tends to be about sheer previously invested in scalable,
could see a new head amend or by the leak of LLaMa into open computing power and on-chip secure, reliable, and trusted data
revoke the current executive source. This also brought new memory; inference tends to be infrastructure and governance
order on AI. Voluntary AI training and fine-tuning about memory bandwidth, will lag, which should serve as a
principles will move to fill technology, so expect a dense-packing containers, and wake-up call and spur much-
regulatory gaps, but some will spectacular proliferation of north-south networking. needed investments in this area.
amount to ethics washing. domain-specific AI systems.
.
Key messages
5 6 7
An explosion in mass-market Implementation of GenAI in GenAI developments in Asia &
GenAI chatbots and turbo- healthcare will ramp up in 2024 Oceania will be driven by the
charged virtual voice assistants need for localization
There will be a new wave of The rubber will hit the road for Vendors in China, South Korea,
multimodal GenAI chatbots GenAI in healthcare in 2024, with and Japan have launched
offering low-friction investment in GenAI increasing localized GenAI solutions
customization. But some will be 10 times from 2023 to 2028. optimized for local culture,
poorly executed and elude Enterprises should prepare languages, and value. The
responsible AI guard rails. GenAI through the development of Singapore government is
will give general-purpose voice robust implementation and risk providing incentives to
assistants a much-needed shot in mitigation strategies, while domestic developers. More
the arm, enhancing vendors should invest in ease-of- localized LLMs will appear in
personalization and the range of use and advisory services. Southeast Asian and South Asian
use cases they can support. markets, focusing on regional
languages and even dialects.
Key recommendations
Enterprises Enterprises (continued)
• Seize the moment to define AI that works for you. Technology such as • Beware of the Fear of Missing Out (FOMO). Rather than asking your
QLORA is dramatically reducing the barriers to entry to shaping AI, not organization “How do we leverage this technology and make money
just to using it. Enterprises should take the opportunity to use their from it?” focus on business challenges and whether GenAI can help with
unique data assets and expertise in their own business and avoid those. Start with use cases and enterprise strategy to realize sustainable
becoming mere consumers of undifferentiated AI services that business value beyond proofs-of-concept.
competitors can use with equal effect.
• Invest in data quality, management, and governance. Third-party GenAI
• Moving from AI development to production will change your models need to be grounded in enterprise data to deliver business
technology requirements. AI in production is less about core training value. And it is still “garbage in, garbage out,” only more so with GenAI,
and more about inference, the phase where a trained model is used to given how fluid and human-like the output of LLMs are and their
make predictions or generate outputs on new, unseen data. The propensity to "hallucinate."
emphasis is north-south rather than east-west in networking terms,
more about latency than about compute.
• In the face of accelerated innovation in transformational tech,
enterprises must focus on the fundamentals. This includes people,
processes, and technology while evolving their governance policies and
risk mitigation strategies. A strong foundation will enable organizations
to be agile in adoption and intervene quickly to mitigate risks.
Key recommendations
Vendors Vendors (continued)
• In Asian markets, the priority should be on localized GenAI capabilities. • Do not assume flagship GPU growth will go on forever. Growth in the
Vendors operating across Asia and Oceania will witness a rise in demand size of state-of-the-art AI models was already slowing when open-source
for domain-specific GenAI models that can understand the local context. models hit, and the latter have dramatically shifted the emphasis from
Region specificity is critical as these markets have unique linguistic the race for trillion-parameter models to the 5B-50B weight class. Think
preferences, customs, and value systems. Customers will only be open to scalability, and make sure your support for the new ones is excellent.
GenAI models when they can assimilate well with local culture.
• Plan for a shift from flagships to dual-purpose, scalable processors
• Provide a strong technology roadmap with clear messaging. Now that after the AI training boom peaks. The growing importance of both AI
the industry has witnessed the first wave of GenAI, it is important for inference and training or tuning smaller models, is pointing the way
vendors to support the next wave of developments with a well- here.
articulated technology roadmap. This will enable customers to have
• Help your customers build foundational underpinnings for GenAI. Make
clarity around future services and to better plan their GenAI investments
data dependency clear and the need for robust, secure, governed,
and rollouts.
scalable, and trustworthy data infrastructure. Be clear in your
• Prepare early for new GenAI regulations and brace for further change. contractual language about where/how client data used to fine-tune
GenAI governance is complex, at times controversial, and still evolving. your models will be processed and stored, including prompts. Data and
Vendors must give themselves ample time to understand what is AI governance increasingly look like vendor due diligence and
required, which could entail further investment and modifications to procurement teams may lack critical skills and knowledge while facing
existing solutions. pressure to move fast as enterprises experiment.
Key recommendations
Telcos
• Although a handful of Asian telcos are building their own GenAI models
in-house, for most telcos, this will not be an option. Constructing a
model from the ground up provides maximum control but is a massive
undertaking that requires deep AI expertise and pockets. Fine-tuning an
existing foundation model will be more viable for the majority. This
process allows telcos to adapt pre-built, trained models to be domain-
and task-specific
• Look to off-the-shelf solutions. This includes those that are telco-specific
to horizontal solutions supporting use cases that are applicable to most
verticals. This is a lower-friction, lower-cost, and faster route to deploying
GenAI. Moreover, telcos could combine in-house efforts and off-the-shelf
GenAI solutions.
• Asian telcos should choose a solution provider that truly
appreciates regional complexity. Select only solution providers with
proven track records operating in multiple markets in Asia and Oceania
and a clear understanding of local regulations and requirements. GenAI is
more than just data and computing—its deployment needs
complete alignment with local market practices.
Despite progress in 2023,
the governance
framework for GenAI will
be highly uneven
Governance frameworks struggle to

keep pace with GenAI A growing body of consumers think AI is in need of urgent regulation
• The EU’s AI Act was finally passed in December 2023 and will no doubt Do you think AI is in need of regulation?
influence other legislative regimes. But the law is not expected to come
into effect until 2025. This leaves a governance void while GenAI is
developing so quickly that new challenges could emerge over the next year 60%
Yes, urgently
that create tension with the existing AI Act or present issues it is not 41%
scoped to cover. We also expect to see legal challenges and pushback
against the AI Act on the grounds that restrictions on highly advanced
Yes, in the medium to longer-term (next 27%
proprietary AI models stifle innovation.
2-5 years) 42%
• The US presidential elections in 2024 could see a new, incoming president
amend, rescind, or revoke the Biden-Harris Administration’s Executive
Order on AI. But even as it stands, the executive order is sprawling and will Don’t know
9%
be hard to implement. 11%
• Voluntary AI Principles will come to the fore and proliferate, attempting to

fill regulatory gaps and foster responsible AI. Some will be robust in their No
4%
own right and/or will back up regulatory regimes (e.g., in the US and UK). 6%
But others will be disingenuous (i.e., ethics washing) and used to deflect
regulations and scrutiny. 0% 20% 40% 60% 80%
• Despite efforts to make GenAI content identifiable, 2024 will see a

landmark ruling on copyright infringement caused by GenAI in favor of the 2023 2021
plaintiff. No sooner has a technique been introduced to identify content Source: Omdia 2023 Consumer AI survey.
produced by GenAI than ways are found to circumvent it. 2023 n=3,118, 2021 n=3,137
© 2023 Omdia
GenAI governance extends beyond advanced models and high-risk systems:

Everyone needs to step up
Player type What will the impact be? Impact How should players respond?
rating
Model The current rounds of AI legislation that have passed or are Start preparing now for forthcoming AI regulations, and do not
builders and in the pipeline will have a far-reaching impact on GenAI. +2 underestimate how long it could take to align with new rules.
solution This may be painful for some at first but it will ultimately be +1 GenAI regulations are complex and can impose strict
vendors of benefit to all AI stakeholders and society. The regulatory − requirements that may entail further investment and
burden is writ large with highly advanced AI models used to -1 modifications to existing AI systems. Falling foul of regulations
power GenAI and/or those AI systems that are considered -2 will damage your reputation, erode customer trust and hurt your
high risk, which is as it should be. For example, under the bottom line. AI principles are now a common tool in AI
EU AI Act, highly advanced AI models must meet strict governance and can be an effective way to supplement
measures to ensure greater transparency. But developers of regulations or provide standards of behavior and operational
lower-risk systems should not be complacent as such guidance in the absence of legal direction. But those that
systems can still have negative outcomes if poorly executed subscribe to AI principles must do so with integrity and not view
or if inherent vulnerabilities of GenAI are not mitigated AI principles as a marketing tool. Regulators and customers will
(e.g., mistakes, hallucinations). inevitably see this for what it is.
From scale to diversity:
The revolution in open-
source AI
Open-source projects will slash the

cost of GenAI innovation
Open source LLMs have unleashed enormous creativity and will change the
• The leak of LLaMA’s weights in February 2023, followed by Meta’s decision industry in 2024
to accept this and release the entire family of models as open-source
software, is having a dramatic impact on the field of AI research. Although
Meta retains certain rights over LLaMA, within three months of the leak,
the RedPajama project succeeded in replicating the model and its training
dataset from scratch as a pure open-source project.
• These events have ushered in an unprecedented wave of innovation as
55,762 99.3%
LLaMA and LLaMA-2 serve as a basis for open-source LLMs, mostly Asia & Oceania
targeting the so-called “missing middle” in terms of model size, between 5- OF CHATGPT’S ACCURACY
50bn parameters. APACHE-LICENSED MODELS
On the Vicuna benchmark. Guanaco-
• Many of these have proven capable of delivering state-of-the-art On HuggingFace as of November 30,
65B after eight hours of QLORA fine-
performance in their specific domains, rivaling huge proprietary models 2023. Another 26,291 are released
such as the GPT series from OpenAI. OpenChat-7B matches or beats tuning on a single GPU
under MIT Public License
ChatGPT on 6 out of 8 tasks despite being 4% its size.
• Much of this innovation centers on better methods of model development,
training, and tuning—methods that themselves help to accelerate
fundamental AI innovation. Perhaps the most important is QLORA, which
Source: QLORA: Efficient Finetuning of Quantized LLMs
demonstrated fine-tuning a model based on LLaMA with 50,000 tokens of (https://arxiv.org/pdf/2305.14314.pdf)
new data in eight hours with a single NVIDIA A100 GPU or in a day on an
enthusiast-grade PC GPU. This dramatically lowers the cost of building your
own AI. Source: Omdia © 2023 Omdia
GenAI models will become smaller, specialized, and possibly yours
Player type What will the impact be? Impact rating How should players respond?
Hardware The race for model size is slowing down as smaller +2 Do not plan for infinite flagship GPU demand growth. Do
vendors specialized models proliferate and fine-tuning +1 think in terms of scalability—both scaling up into big
techniques advance. Customers for AI training − training clusters and scaling down into manageable
hardware will become smaller and more diverse -1 inference or double-duty systems. Include features of major
in nature—not just giant corporates, research labs -2 open-source families in the models you design for—
and hyperscalers. Mediatek is doing this for LLaMA-derived models in its next
APUs.
Enterprises This is your moment to define AI models that +2 Get experimenting, ideally with something that is not just a
work for you, built on your own unique data +1 chatbot project. Some of the most exciting current
assets, developed for your application. It will get − technology can run on an Apple Silicon Mac, and there are
easier to adapt AI to your purposes, but this is -1 some compelling startups in this space, e.g., Lamini.
also true for your competitors. -2
Application AI applications developers will have to decide It is crucial to understand whether you are more interested
+2
developers between pursuing differentiation through their in rapid product development, or in serving the same model
+1
own AI developments or sticking with generic “big cheaply. Whether or not smaller open-source LLMs pan out
−
AI”. The impact on unit economics will be big but -1
cheaper in inference is open, as hyperscalers have
difficult to model in advance. significant scale economics. You may have to find out by
-2
trial and error.
Input/output trumps
compute for GenAI
inference
Serving GenAI models will demand a AI cloud and enterprise data center processor forecast for compute capacity, world
new hardware approach markets: 2022–28
80,000
• GenAI applications are trained in much the same way as others—although
often with even more data. However, inference operations are 70,000
substantially different.
– GenAI applications are very often interactive, with the app offering the 60,000
user a succession of intermediate results they respond to with further
prompts. Latency is crucial for user experience. 50,000
Revenues ($m)
– A key technique in current GenAI applications is retrieval-augmented
generation (RAG), in which the AI model formulates its response to a 40,000
prompt using an external data source like a search engine.

30,000
– GenAI models are also often used to retrieve information from a vector
database.
20,000
• These use cases are especially demanding in terms of input/output (I/O)
and interconnect. Generating more floating-point operations is relatively 10,000
easy, but I/O Asia & Oceania is harder.
• As a rough guide, I/O costs one or two orders of magnitude more energy 0
than compute. 2022 2023 2024 2025 2026 2027 2028
• Omdia expects to see more interest in middle-weight AI accelerators with

Low (<150 TOPS) Medium (150–250 TOPS) High (>250 TOPS)
dedicated passive optical I/O, inspired by Google’s approach with its
Tensor Processing Unit (TPUs) chipsets. Source: Omdia AI Processors for Cloud and the Data Center Forecast 2023
© 2023 Omdia
© 2023 Omdia
GenAI drives demand for I/O across AI hardware
rating
Hardware By the end of 2024, inference and dual-purpose The passive optical interconnect in Google’s TPUs is likely to
vendors inference/training will grow compared to full-up LLM +2 be something the whole industry will try to imitate, and a
training. Customers with more inference workloads will +1 standards-based solution is optimum. Be skeptical of
need greater north-south interconnect at every level in − chasing performance numbers, e.g., tera-operations per
the data center, from the Internet to the system-on-chip. -1 second (TOPS) or even on-chip high bandwidth memory (as
-2 it must be filled up).
Enterprises As you move from AI development to AI production, you’ll +2 If you have inference workloads that will need to scale out,
tend to move from AI model training to fine-tuning and +1 you need to think in terms of the fastest path from the
inference. − network to the model. But not everyone will as some AI use
-1 cases (e.g., financial services) are high value but low
-2 volume.
Application Investments in flagship graphical processing units (GPUs), +2 Buying decisions need to consider the mix of capabilities
developers or for that matter in lower-tier GPUs, might lose their +1 your workload needs holistically, including not just memory,
value faster than anticipated. − compute, and I/O, but also software support and the
-1 consequences for power consumption of how they interact.
-2
GenAI will trigger a fresh
wave of investment in
enterprise data quality,
management, and
governance
GenAI will accelerate spend on enterprise data quality, management and

governance
Enterprises have enthusiastically embraced GenAI
• Enterprise adoption of GenAI will continue at a steady clip over the next 12 months,
enabled by an expanding smorgasbord of pre-trained models, tools, and applications. All How applicable is text-based GenAI (e.g., ChatGPT) both currently
major software vendors are actively integrating GenAI into their enterprise and line-of- and within the next 24 months?
business offerings. Hyperscalers, integrators, and start-ups—and let’s not forget open
source—are constantly offering new tools, platforms, pre-trained foundations, and fine-
tuned models and are pushing the boundaries of compute, with much of the innovation Fully adopted 12.83%
happening at break-neck speed.
• This means there is a lot to choose from, and no one will leave the table hungry,
Implementing 24.82%
regardless of their organization’s size, business model, industry sector, or maturity with AI
and appetite for risk.
• However, the utility and business value of these experiments will depend on
Pilot testing 25.67%
the availability of high-quality data that is needed to fine-tune and ground foundation
models to the specifics of an enterprise to make them truly usable. Hence, we expect
heightened awareness of how critical data is for AI and new investments in data quality,
data management, and governance to address some of the perennial challenges Currently considering 27.16%
enterprises have struggled with, notably, poor quality, questionably governed data.
• You can build a beautiful GenAI house and maybe even host a housewarming party, but
No interest 9.53%
you won’t be able to live there unless the house has a proper foundation and utilities in
the ground (water and gas mains, telecommunication, etc.). Data is the foundational
infrastructure for AI. AI vendors would do well to make this dependency clear, educate 0% 5% 10% 15% 20% 25% 30%
buyers, and help build foundational underpinnings for enterprise AI. Note: n= 5,185
Source: Omdia IT Enterprise Insights: IoT, Cloud, AI, 5G, and Sustainability – 2024 © 2023 Omdia
•
The success of GenAI in the enterprise hinges on data quality

rating
Enterprises Enterprises will continue to explore the benefits of GenAI Enterprises need to realize that data is foundational for AI.
+2
while facing challenges with insufficient knowledge and ChatGPT works for consumers, but enterprise value comes from
+1
experience, lack of skilled AI personnel, low data and AI grounding foundation models in enterprise data and context.
−
literacy and maturity, and, for many, chronic underinvestment Organizations not previously invested in scalable, secure,
-1
in data infrastructure and quality. The ever-growing GenAI reliable, and trusted data infrastructure and governance will
-2
ecosystem and market instability will contribute to the pains. continue to lag and will need to accelerate funding for quality
and management tools, processes, training, and governance.
Vendors Data quality is the perennial challenge for enterprises and one +2 Vendors must provide thought-leadership and act as a partner to
of the biggest barriers to AI adoption, and GenAI is no +1 enterprises eager to deploy GenAI, especially those of low
exception. The good news is that many plan strategic − maturity and lacking resources. Such partnership must involve
investments in data quality, security and privacy (35%) and -1 education and transparency and focus on mutual long-term
data integration, preparation and orchestration (29%) as part -2 value. Using clear contractual language and being clear about
of their 2024 AI plans. But they still need help. prerequisites (e.g., around data dependency) are also a must.
Integrators Omdia anticipates that GenAI added more than $6.7 bn to the This is why integrators play a critical role, especially since many
+2
AI software market in 2023 alone and $63 bn in 2028. In other of them have deep vertical knowledge, know their clients well
+1
words, the potential is enormous, and hyperscalers, software −
and have long-standing relationships. Leverage that bond and
vendors and GenAI start-ups cannot address the need all by -1 trust. You are an integral player in the market providing industry-
themselves. -2 specific GenAI solutions, as well as much needed guidance,
education and best practices.
An explosion in mass-
market GenAI chatbots
and turbo-charged voice
assistants
The new wave of GenAI chatbots and voice assistants must be handled
with care
• Expect a new wave of GenAI chatbots that are more capable and flexible than previous Global software revenues for GenAI enabled chatbots & virtual
iterations. OpenAI’s ChatGPT will once again lead the charge with multimodal assistants
capabilities that will become the norm (e.g., image and text image prompts, voice and
text outputs). OpenAI’s fully customizable GenAI GPTs will be popular: they are open to 5,000
all and, unlike Microsoft Co-Pilot, will not impose steep usage fees. 4,645
4,500
• There will be fresh attempts to monetize GenAI chatbots. The arrival of OpenAI’s GPT
app store will give developers new routes to monetization. Meta will explore 4,000
monetization opportunities for its Meta AI GenAI chatbot as it rolls out across the 3,500 3,379
Revenue ($m)
company’s key platforms (WhatsApp, Messenger, Instagram).
3,000
• GenAI will boost horizontal voice assistants with multimodal capabilities, support for
2,500 2,342
new use cases, and greater personalization. There will be a wider release of the 2,223
integrated Google Assistant and Bard (Google’s GenAI chatbot). We expect Amazon to 2,000
1,642
use its investment in Anthropic (developer of Claude) to supercharge its Alexa voice 1,526
1,500
assistant and for Apple to unveil its own version of ChatGPT, which will benefit Siri. 1,167
904
1,000 759
• Powerful on-device AI will improve the performance of GenAI chatbots and 461 437
voice assistants. The number of AI-enabled end user devices, especially smartphones, 500
126 58
216
will increase significantly, helped by developments such as Qualcomm Snapdragon 8 0
Gen 3 and MediaTek’s Dimensity 9300. 2022 2023 2024 2025 2026 2027 2028
• Many GenAI chatbots and GPTs will be useful, fun tools, but others will be poorly Chatbots Voice assistants
executed and could elude responsible AI guard rails, which will cause controversy and
attract regulatory scrutiny. Source: Omdia Artificial Intelligence Software Market Forecasts – 2H23 © 2023 Omdia
The full benefits of GenAI chatbots and voice assistants can only come with
careful execution
rating
Enterprises GenAI chatbots can enhance internal enterprise apps (e.g., Execution needs to improve—stop treating GenAI as a quick fix.
IT service desk Q&A, data analysis) and external use cases +2 Enterprises are enthusiastic about GenAI chatbots and see
(e.g., customer care, personalized marketing). This can +1 customer care/experience scenarios as low-hanging fruit. In
improve enterprise productivity and cost efficiencies. But − contrast, consumers take a dim view of AI-enabled chatbots and
GenAI chatbots can make mistakes that, if relied on, could -1 virtual assistants, as evidenced by Omdia’s 2023 Consumer AI
be damaging, while a poorly executed chatbot can impair -2 survey where respondents ranked these applications second
rather than enhance the customer experience. from last in terms of perceived utility.
Service There has been no major game changing innovation in voice Explore how GenAI’s multimodal capabilities can give users new,
+2
providers assistants for some time but rather incremental flexible ways to engage with horizontal voice assistants. For
+1
enhancements or improved reach via device integrations. example, assistants embedded in devices with displays can make
−
GenAI can give horizontal voice assistants a new lease of life, use of GenAI image and video. Only release GenAI enhanced
-1
with improved personalisation and nuanced conversational -2
voice assistants with firm guardrails in place to ensure they
abilities being conspicuous benefits. behave in accordance responsible AI principles.
Vendors The market for conversational platforms powering Vendors must provide well-thought-through technology road
+2
enterprise chatbots and virtual assistants is crowded. GenAI maps, for example, how your chatbot solution can support multi-
+1
can improve performance and bring new capabilities, which modal capabilities and the benefits that will bring to enterprises.
−
can in turn provide differentiation. However, the flip side is Vendors must have a meaningful USP, given the competitive
-1
that no sooner has one vendor innovated than others nature of this market (e.g., stand-out tools, ease of integration,
-2
quickly follow. domain expertise, scalability, and customization features).
Implementation of GenAI
in healthcare will ramp
up in 2024
In healthcare GenAI is moving from experimentation to deployment
• The US spends approximately $4tn ($9tn globally) on healthcare every year, Global revenue for AI software in healthcare
of which approximately $360bn is spent on administration annually.
• In 2024, the sector will move from experimentation with GenAI to 20.000
implementation, beginning with low-risk, high-return on investment (ROI) use 18.000

cases. GenAI software spending in the sector will rise from $620m in 2023 to
$6.7bn in 2028, representing a CAGR of 69.4%. 16.000
• The top use cases in 2028 in terms of software revenue are medical image 14.000
analysis, computational drug discovery, virtual assistants, medical treatment
Revenue ($m)
recommendation, and cybersecurity for medical devices. 12.000
• The proverbial low-hanging fruit in the sector is addressing the huge 10.000
administrative burden within the sector. It is estimated that 25–50% of a
doctor’s time is taken up by administrative work, which contributes to 8.000
burnout. 6.000
• The proliferation of domain-specific models will facilitate innovation up
4.000
and down the GenAI technology stack and further adoption within the
industry. This includes proliferation in large foundational models, such as 2.000
Med PaLM, hosted on hyperscaler cloud-based platforms and domain-
specific generative models. 0.000
2023 2024 2025 2026 2027 2028
• Expect the rollout of GenAI to begin by automating administrative tasks such
Predictive Generative
as documenting and summarizing doctor–patient interactions, medical image
processing, and further adoption of GenAI tools in drug discovery. Source: Omdia Artificial Intelligence Software Market Forecasts – 2H23 © 2023 Omdia
Increased spending on GenAI will fuel implementation and competition
rating
Enterprises GenAI has the potential to address several of the sector’s pain Enterprises should prepare for implementation through robust
points: costly and tedious administrative tasks, poor +2 strategic and tactical planning, strong governance, and risk
productivity in drug R&D, the patient experience, and +1 mitigation policies and procedures. The pace of innovation is
inefficient processes across from drug discovery and − high up and down the tech stack, so enterprises will need to
manufacturing through to care delivery. Deployment of GenAI -1 monitor the evolution of the GenAI and ML landscape for
solutions to address the sector’s pain points is not without -2 innovation and to see how other healthcare organizations are
challenges but for the enterprises that get it right the payoff leveraging the technology. Enterprises must also develop
will be profound, particularly in productivity and operational KPIs/metrics to measure GenAI’s performance to internal goals.
efficiency.
Vendors Spending on AI in general and GenAI will continue to increase +2 Vendors should continue to invest in lowering the bar to
in 2024 and throughout our current forecast period to +1 adoption for enterprises and in building out their advisory
2028. Healthcare organizations will seek to further engage − capability, as the sector will be turning to the vendor community
with vendors and professional service providers to continue -1 for guidance. Vendors must be clear about their value
experimentation and deployment of GenAI in low-risk, high- -2 proposition and look to partner and collaborate across the
ROI use cases. No one can go it alone; it will require vendor ecosystem. Healthcare organizations are more willing
collaboration between industry and the tech community. The than ever to work with a diverse set of providers to navigate the
competitive landscape for GenAI products and services will GenAI space.
intensify as will the impact of the open-source community in
shaping enterprise deployments.
GenAI developments in
Asia & Oceania will be
driven by the need for
localization
The GenAI ecosystem in Asia & Oceania prioritizes localization

• The Asia & Oceania region, in general, is lagging slightly behind the US in terms GenAI software revenue in Asia
of breakthrough GenAI developments. 20,000
– China is an exception. The government has made AI a strategic priority, Chinese 18,000
big tech players such as Baidu, Alibaba, and Tencent have deep AI expertise
while certain start-ups are driving innovation, for example, Baichuan’s LLM. 16,000
• In broad terms, the GenAI ecosystem in Asia & Oceania is more strongly focused 14,000
on localization than achieving global dominance in GenAI, with language 14,739
12,000
localisation top of the agenda. Most LLM used for GenAI development use
Revenue ($m)
English language and/or other mainstream Western languages. Additional 10,000
languages are typically not well-supported by the major open-source and
8,000
proprietary GenAI models. This is prompting Asian and Oceanian vendors and 10,823
developers to create models optimized for local language use cases. 6,000
7,597
• As of 2023, Alibaba, Baidu, Huawei, Tencent, SK Telecom, and KT are developing 4,000
their own AI chipsets, infrastructure, frameworks, services, and solutions on 4,906
3,588
GenAI. China Mobile, China Telecom, Naver, NEC, and NTT have launched LLMs in 2,000 2,803
2,479
Chinese, Korean, and Japanese, respectively. 562 1,018 1,633
0
• Alongside their own technology stack, these vendors and telcos are building their 2023 2024 2025 2026 2027 2028
own ecosystem with system integrators and distribution partners.
• Omdia expects this trend to continue in 2024, with a major push from large Central & Southern Asia Oceania, Eastern & South-Eastern Asia
markets such as India, Indonesia, Thailand, the Philippines, and Vietnam.
Source: Omdia Artificial Intelligence Software Market Forecasts – 2H23 © 2023 Omdia
It will take an entire ecosystem to implement GenAI localization

Player type What will the impact be? Impact rating How should players respond?
Vendors Since initial fine-tuning and prompt engineering tools +2

Vendors must identify the specific localized business needs and
(non-telco) were only available in English, major vendors in China, +1 opportunities for GenAI in their markets. Once the needs have
South Korea, and Japan have taken the lead in creating − been identified, invest in corresponding R&D capabilities,
GenAI solutions in local languages. Others will follow. -1 including attracting the right AI engineers. Develop LLMs in
-2 local languages and find the right local system integrators and
distribution partners.
Telcos Telcos in China and South Korea have already Telcos have infrastructure that is an asset in supporting GenAI.
+2
demonstrated technology leadership in GenAI. Other But before launching any GenAI solutions, telcos must set up
+1
Asian telcos will follow these pioneers in terms the right data operation and governance frameworks. They
−
of localized GenAI models. These models will be heavily also need to have a clear understanding of how best to deploy
-1
used in customer services, revenue assurance and GenAI: build their own model, fine-tune an existing foundation
-2
generation, network optimization, sales and marketing, model, collaborate with other telcos, use off-the-shelf
and internal knowledge sharing. solutions.
Enterprises The initial wave of GenAI solutions in English makes them Conduct market research to better understand customer
+2
inaccessible for small and medium enterprises that are needs. Investigate and identify the best GenAI solution
+1
serving local customers. Enterprises should start −
suppliers that provide the right levels of local support in
demanding localized solutions from AI vendors, -1
performance, accuracy, governance, compliance and
particularly for virtual agents and customer experience. -2
transparency.
Appendix
Appendix
Methodology
The Trends to Watch predictions in this report is draws on primary data and analysis from Omdia AI related trackers, forecasts, consumer and enterprise
surveys and research reports. The Trends to Watch predictions also benefit from insights from AI stakeholders and the wider Omdia analyst network.
Further reading
• Artificial Intelligence Software Market Forecasts – 2H23 Data (November 2023)
• Telco positioning in the generative AI ecosystem (November 2023)
• Consumer AI Survey: 2023 – Generative AI Analysis (August 2023)
• AI Processors for Cloud and Data Center Forecast Report – 2023 Database (August 2023)
• AI Market Maturity 2023: Data (April 2023)
• Generative AI: Tech Provider Viewpoints (March 2023)
• Generative AI: Market Landscape 2023(March 2023)
• The Rise of Generative AI: A Primer (February 2023)
Author
Eden Zoller, Chief Analyst, Applied Intelligence
askananalyst@omdia.com
Appendix
Omdia Consulting
We hope that this analysis will help you make informed and imaginative business decisions. If you have further requirements, Omdia’s consulting team may be
able to help you. For more information about Omdia’s consulting capabilities, please contact us directly at consulting@omdia.com.
Citation Policy
Request external citation and usage of Omdia research and data via citations@omdia.com.
Disclaimer
The Omdia research, data and information referenced herein (the “Omdia Materials”) are the copyrighted property of Informa Tech and its subsidiaries
or affiliates (together “Informa Tech”) or its third party data providers and represent data, research, opinions, or viewpoints published by Informa Tech,
and are not representations of fact.
The Omdia Materials reflect information and opinions from the original publication date and not from the date of this document. The information and
opinions expressed in the Omdia Materials are subject to change without notice and Informa Tech does not have any duty or responsibility to update the
Omdia Materials or this publication as a result.
Omdia Materials are delivered on an “as-is” and “as-available” basis. No representation or warranty, express or implied, is made as to the fairness,
accuracy, completeness, or correctness of the information, opinions, and conclusions contained in Omdia Materials.
To the maximum extent permitted by law, Informa Tech and its affiliates, officers, directors, employees, agents, and third party data providers disclaim
any liability (including, without limitation, any liability arising from fault or negligence) as to the accuracy or completeness or use of the Omdia Materials.
Informa Tech will not, under any circumstance whatsoever, be liable for any trading, investment, commercial, or other decisions based on or made in
reliance of the Omdia Materials.
Get in touch
Americas Europe, Middle East & Africa Asia Pacific
customersuccess@omdia.com customersuccess@omdia.com customersuccess@omdia.com
08:00 – 18:00 GMT -5 8:00 – 18:00 GMT 08:00 – 18:00 GMT + 8

2024 Trends To Watch Generative AI GenAI PDF v2

Uploaded by

Copyright:

Available Formats

2024 Trends To Watch Generative AI GenAI PDF v2

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2024 Trends To Watch Generative AI GenAI PDF v2

Uploaded by

Copyright:

Available Formats

2024 Trends to Watch:

innovation, disruption and anxiety 2028 2024 2023

months, and to a significant degree for 37% of respondents. Energy 380

It will displace job functions, put people out of

Enterprises Enterprises (continued)

Vendors Vendors (continued)

Governance frameworks struggle to

• Voluntary AI Principles will come to the fore and proliferate, attempting to

• Despite efforts to make GenAI content identifiable, 2024 will see a

GenAI governance extends beyond advanced models and high-risk systems:

Open-source projects will slash the

GenAI models will become smaller, specialized, and possibly yours

new hardware approach markets: 2022–28

prompt using an external data source like a search engine.

• Omdia expects to see more interest in middle-weight AI accelerators with

GenAI drives demand for I/O across AI hardware

GenAI will accelerate spend on enterprise data quality, management and

The success of GenAI in the enterprise hinges on data quality

In healthcare GenAI is moving from experimentation to deployment

implementation, beginning with low-risk, high-return on investment (ROI) use 18.000

Increased spending on GenAI will fuel implementation and competition

The GenAI ecosystem in Asia & Oceania prioritizes localization

It will take an entire ecosystem to implement GenAI localization

Vendors Since initial fine-tuning and prompt engineering tools +2

You might also like