Generative AI and enterprise software: What’s the revenue uplift potential?
Gen AI is coming to enterprise software, but expect competition between vendors who want to charge per user and IT departments that believe generative AI features should be free.
Duncan Stewart
Baris Sarer
Gillian Crossan
Jeff Loucks
At the start of 2023, excitement about generative AI prompted three big questions for enterprises: Will gen AI be embedded in enterprise software? How will vendors charge extra for gen AI tools inside software? And, how big an uplift in revenues will gen AI create for the enterprise software industry?
As we enter 2024, Deloitte has three projections: First, we predict almost every enterprise software company will embed gen AI in at least some of their products. Second, we predict there will be a mix of pricing models: explicit per-seat pricing (per user per month or PUPM), consumption-based pricing, a hybrid approach, implicit pricing (retaining whatever model they have today but charging more), or free, at least for now. Finally, we predict that the revenue uplift for enterprise software companies (in addition to the cloud providers of gen AI processing capacity) will be approaching a US$10 billion run rate by the end of 2024.
That’s lower than the US$14 trillion (not a typo) that one fund manager has predicted for gen AI software by 2030,1 but still significant, even though it’s a fraction of the US$1.6 trillion in global enterprise IT spending projected for 2024,2 and also smaller than the expected hardware uplift for chips and servers that perform gen AI of more than US$50 billion in 2024.3 Given the excitement about gen AI enterprise software tools in 2023, how did we arrive at this number, and why isn’t it higher?
The market potential for 2025 and beyond looks robust, and the revenue uplift for enterprise software companies will likely be tens of billions of dollars. But 2024 is effectively a transition year. The various kinds of enterprise software tools that are expected to include gen AI are not launching until late 2023 or early 2024. Some companies have cautioned analysts that adoption and revenues in the second half of the year will be much stronger than in the first half.4 More than 70% of companies are experimenting with gen AI, but less than 20% are willing to spend more on it.5 Putting that all together for 2024, we predict a revenue uplift approaching a $10 billion run rate by the end of 2024.
We predict that for most enterprises, the “gateway to gen AI” will likely be through the three categories outlined below wherein gen AI features are embedded in existing software, often going unnoticed by users.
Broad enterprise productivity software suites: There were predicted to be 1.14 billion knowledge workers worldwide in 2023,6 which suggests a total addressable market of almost US$400 billion annually, assuming every knowledge worker needs at least one gen AI-powered set of enterprise software tools and that they pay US$30 PUPM.
Enterprise software tools: There are various software tools such as database and analysis solutions, enterprise resource planning (ERP) solutions, customer relationship management (CRM) solutions, creative and document management solutions, and more. As of the time of writing, Deloitte analysis of publicly available announcements suggests that 100% of the 50 largest enterprise software companies are planning to offer a version of their software that has gen AI features, with some using PUPM pricing models, usage pricing, or offering for free, at least for now.7
Engineering, design, and software development tools: There are a number of new tools in which gen AI is not merely an enhancement but is, in fact, core to the new product. Multiple chip design companies are offering gen AI-enhanced versions that design chips (one of our 2023 TMT Predictions),8 perform functional verification, and test silicon.9 In the computer-aided design space, many players offer gen AI features.10 And some software development tools charge around US$10 PUPM.11
Gen AI is expensive to operate but its value to customers may not yet be clear
At one level, it makes sense that software companies want to charge for gen AI. In addition to growing revenues, offering gen AI is expensive. They are often spending billions of dollars either buying the chips that are needed for gen AI training or needing to buy instances from the cloud or chip companies. As an example, looking at some of the largest cloud players, they are expected to spend from 3%–13% of their 2023 capex on AI.12 Further, the operating costs are non-trivial, with estimates that each gen AI query costs from US$0.01 to US$0.36 per query. As an example, one service that costs US$10 PUPM is rumored to be losing US$20 monthly, with some users costing the provider more than $80.13 Both chip pricing and operating costs are expected to come down sharply over time but likely not until the current gen AI chip shortage eases, which Deloitte predicts will not occur sooner than the second half of 2024 (see gen AI chips prediction.)
Meanwhile, at least some buyers of enterprise software are pushing back. To quote a June 2023 US survey of buyers, although the long-term potential of gen AI features was very high, some respondents said that AI features were “table stakes” (that is, every vendor needed to offer it) but “good luck trying to get me to pay for it.”14 Other customers may not yet see the value of gen AI in their workflows: They may not want to pay to use it but might take free trials that later shift to paid services.
The bottom line
At a high level, companies likely won’t pay for gen AI-enhanced tools unless they generate a positive ROI. A research study from the fall of 2023 suggests that ROI might be very strong, with knowledge workers using gen AI (direct access, not through embedded AI inside enterprise software) to do more, faster, and at a higher quality than those not using the tools.15 If improvements of those magnitudes are seen in the real world and across multiple industries, it seems likely that the uplift in revenue over the longer term could be much larger than the early innings of 2024 might suggest.
However, if ROI gains are lower (or take time to be demonstrated) vendors are likely to see slow adoption or buyer pushback on pricing. One potential alternative between high per-seat per-seat-per-month pricing and free “table stakes” gen AI could be a hybrid model: a relatively low per-seat month price (<US$10) but combined with a consumption charge (aka usage-based pricing), which would allow vendors to recapture some of their per-request operating costs.16 Effectively, the more you use it, the more you pay.
Regulation and concerns around privacy, IP ownership, accuracy/confabulation, and more could be roadblocks. Any one of these could be enough to slow or even halt the adoption of gen AI-enhanced enterprise software solutions. Some proposed EU rules are so restrictive that many current-generation AI software tools might not be allowed in that market. (See Paul Lee’s 2024 Prediction on regulation.) One potential solution for some of those barriers could be for companies to build their own models, then train and run them on gen AI cloud services. This is likely to be a multibillion-dollar market for processing, software, and services over time. (See Chris Arkenberg’s 2024 Prediction on Private LLMs.)
Currently, leading gen AI accelerator chips are in shortage and on allocation (see Duncan’s 2024 Prediction on gen AI chips), and this may be complicating the ability of companies to meet demand for gen AI-inside features. They need thousands or even tens of thousands of these US$40,000 chips to meet anticipated demand, and some companies cannot get enough (or any) gen AI via the cloud.17 Capacity is expected to ramp into H1 2024, but in the first part of that year, gen AI-inside software revenues could be capacity-constrained.18 That chip scarcity may allow enterprise software companies to charge higher prices for gen AI features, as users of the software likely cannot build their own gen AI solutions due to chip scarcity.
But it may then become likely that gen AI accelerator chip prices will fall in the next 18–24 months. That could happen abruptly as supply comes online and new entrants emerge. Multiple players are expected to announce new data center and edge processing gen AI chips,19 and although it is unclear what market share these chips could garner, alternative chips are likely to reduce the current hardware shortages and high prices.
Some larger companies that have the capacity to build their own solutions either on top of hardware they buy, or through cloud gen AI capacity, may be deferring building/buying gen AI capacity until the prices come down. They are seemingly happy to be fast followers rather than paying up to be at the leading edge. Relatively few companies are expected to buy the hardware—many will get gen AI from the bigger cloud providers. Further, companies that are planning to use the hybrid model may see customer pushback on usage-based pricing if costs fall sharply, and may need to reduce that pricing.
A gen AI revenue uplift of up to US$10 billion is a significant positive for vendors. But it should be placed in context: Global spending on cloud services is large and still growing, but that growth will likely slow. Public cloud was a US$546 billion industry in 2022, up 22% from the year before,20 but growth was down to 16% in Q2 of 2023,21 and although major cloud players are launching gen AI as a service, it’s unclear from publicly reported numbers just how much money they are making from those services in 2023 or will in 2024. Will gen AI revenues for cloud companies be big enough and fast enough to re-accelerate overall growth to the 20% annual range?
Another question is whether companies will pay for more than one gen AI-inside enterprise software per employee. Most knowledge workers use multiple software tools, and at US$10–30 per month, cumulative “gen AI stacking” spending could be more than US$100 monthly. Could we see a new class of software that sits on top of everything else and does gen AI on everything … a universal co-pilot?
Ultimately, generative AI features have to show efficacy for a number of tasks very quickly and are contributing to enormous amounts of spending and strategic planning from large companies. Providers are confronting their own costs while figuring out pricing for customers, both for pure-play gen AI processing via the cloud and gen AI-enabled service features. End users for their part may take some time to determine how these capabilities translate into direct value for their businesses, but likely will soon realize gains and develop a better sense of its value and how much they’re willing to pay for it. From what we’ve seen so far, this could raise prices across the board.