Riding the generative AI wave

The generative AI start-up landscape is undergoing a remarkable transformation.

Driven by recent advancements in large language models (LLMs) and foundation models, hundreds of startups are emerging, leveraging generative models to unleash a wave of new innovations. Amidst the current excitement, enterprises and venture capital (VC) investors are rushing to pour money into the sector, hoping to capitalise on the potential of this emerging field.

But whilst generative AI startups are attracting headlines and investor money, will they become sustainable businesses? There are multiple hurdles facing generative AI startups, and whilst some will undoubtedly find a route to longer-term value, many will not survive.

Here, we’ll delve into some of the factors shaping the generative AI landscape, exploring drivers behind startup growth and their potential relevance across industry. We also explore some of the challenges faced by generative AI startups and examine how this will impact their longer-term viability.

Growth drivers

Growth of the generative AI startup landscape has been enabled by two key factors.

First, the development of LLMs and wider foundation models. AI models typically require intense computing power and significant resources to train, which previously created major barriers-to-entry for startups. Today, large pre-trained foundation models and LLMs can be accessed and fine-tuned for a wide range of downstream specialised use cases, enabling startups to build, experiment and launch AI applications flexibly and at lower costs. 

Second, VC activity has surged. Despite recent depressions in the tech sector, investments in generative AI startups have skyrocketed. According to Pitchbook, in 2022 globally, generative AI received $4.8B of funding across 375 deals, and in Q1 2023 alone, received $1.7B across 46 deals. This compares to just $1.5B in 2020 and $0.4B in 2018. The allure of potential high returns and the belief that generative AI will disrupt multiple industries is creating a climate of speculative investing, which we expected to continue in the near future.

Zooming in on the start-up landscape

The first category of generative AI startups is model providers – the ones responsible for the very existence of this market, and that power AI products such as OpenAI’s ChatGPT and Anthropic’s Claude. Second, we have a tooling layer, which enables developers to build applications on top of foundation models quicker and more efficiently. An example startup operating in this layer is HumanLoop. The third category are applications, which is the layer where everyday consumers generally interact with generative AI products. There has been an explosion of offerings in this area, with companies such as Jasper, Runway and Harvey experiencing rapid growth. Most startups across this layer are using generative AI to boost operational efficiency, for example by automating repetitive processes and helping with text-based processing and content creation. Other use cases include generating marketing images, writing code, creating personalised gaming experiences, discovering new drugs, and more.

Hype does not mean success

The growth of generative AI startups is notable, and there are undoubtedly startups in this space with the potential to generate real impact and value. However, growth alone is not sufficient to build durable companies, and does not guarantee longer-term success.

For example, generative AI in its current form would likely not exist without advanced LLMs and foundation models, but we expect profit margins for firms operating at this layer to be challenging, in part due to high development, training and inference costs. The California-based startup Inflection AI, for instance, raised a $255M seed round and repurposed the majority of this capital just to develop computing power for its model. Moreover, model developers may face uncertain longer-term differentiation, as models are currently trained using similar datasets, architectures and approaches, and it may be difficult to prevent competitors from replicating any short-term advantages.

Similarly, startups in the application layer may find it hard to maintain a competitive edge. This market is already crowded, and we expect many companies will struggle to move beyond providing “wrappers” around foundation models. There are also profit margin and continuity concerns at this layer; startups are reliant on foundation model providers, which introduces risks from price increases, changes to model access and model performance, and potential model unavailability if a provider go out of business at any time.

In addition, generative AI is still a nascent technology, and faces significant challenges and uncertainties around ethics, security, reliability, regulation, and more. Data used to train foundation models, for example, reflects historic biases, including around gender and race, and without further research and corrections, will act to accentuate existing systematic biases and discrimination in society. AI-generated content also raises concerns around the spread of deepfakes, counterfeit videos, misinformation, and broader threats to liberal democracy. Moreover, the regulatory landscape is also evolving in real-time, and there are on-going legal challenges in the use of copyrighted data to train models, which in turn threatens future model development and availability. These factors – and others – add to significant on-going uncertainty, and further contribute to the longer-term challenges faced by generative AI startups.

Concluding thoughts

Startups are at the cutting edge of the generative AI revolution, propelled by advancements in LLMs and foundation models, and by significant investor funding. While there is no denying the current excitement, it is important to separate hype from the reality and to question: how many of these startups will succeed, and how many are simply cashing in on the hype?

We expect many generative AI startups will fail. Challenges around differentiation, low profit margins, ethical concerns, security risks, misinformation, and more will push many existing ventures out of the market. The ones that remain will be those that surpass these challenges and focus on solving valuable problems for industry and society.

Given these risks, it is essential that enterprises and investors carry out thorough due diligence before engaging in potential partnerships and investment opportunities. Organisations must successfully distinguish between startups that are riding the current wave of hype from those grounded in strong business fundamentals.

Here in Deloitte, we have several teams that specialise in helping our clients navigate startups and the associated investments. If you are interested to hear more about our startup capabilities and services, please do get in touch.