At a gathering in San Francisco in October, Sequoia Capital partner Pat Grady shared how the fund was betting on the next generation of billion-dollar companies to emerge from the AI boom — and how these will be making AI applications rather than building models.
The firm has invested around $150m in OpenAI, Safe Superintelligence and xAI, all of which build expensive foundational models. But Grady said that Sequoia had invested “in order of magnitude more dollars” at the application layer. This, despite many of these applications failing to generate much revenue, if any, in the short term.
It appears that AI is a long-term play across the board in the tech industry. Grady comments demonstrate a widespread belief in the industry, certainly among the investor community, that future AI unicorns will emerge from the application layer.
Over the past couple years, venture capitalists and leading platform providers invested heavily in funding large language models (LLMs). These infrastructure provider services have been computational and capital-intensive to build, explains Charlotte Dunlap, GlobalData research director.
Big Tech companies are investing billions of dollars for training and inference of their AI models. According to Stanford University’s 2024 Human Centred Artificial Intelligence report, the training costs of state-of-the-art AI models have reached unprecedented levels with Google’s Gemini Ultra costing an estimated $191m worth of compute to train, and OpenAI’s GPT-4 an estimated $78m.
Meta’s decision to release its open-source model Llama, Llama 3 launched in April, has further hampered its competitors’ ability to generate revenue from their proprietary models. OpenAI is said to have made a $5bn loss in 2024.
“Now, we see investors turning their attention to the application layer of generative AI (GenAI) innovation. That’s because applications are becoming far more complex, reflecting not just one but multiple LLMs used to help power advanced apps and automatically route prompts to the most appropriate and numerous LLMs,” says Dunlap.
“The next-gen apps include a heavy emphasis on the data integration layer, with features such as text to SQL, and other new forms of data access to improve the workflow layer for vastly improved common business processes in new ways, ultimately improving the UX and CX,” adds Dunlap.
GlobalData’s Artificial Intelligence Executive Briefing (fourth edition) estimates the total AI market will be worth $1,037bn in 2030. Within that market, the global specialised AI applications market will be worth $512bn in 2030, up from $39bn in 2023. In the early years, AI investments will be dominated by computer vision and conversational platforms. But could those companies developing AI applications have already lost their first mover advantage?
Indeed, some estimates (see graph) put ROI on AI already peaking for venture capital investments made in 2013. Snap Inc. has been developing augmented reality applications for its SnapChat platform and is an early investor in AI to drive its applications. The company heavily invested in computer vision engineering, working on augmented reality applications for future hardware modalities such as computer vision glasses.
With a confluence of factors like consumer appetite for greater platform immersion, refined hardware and advances in computer vision engineering, Snap could be one of those AI application companies positioned to become the next generation of Big Tech.
Qi Pan leads Snap’s computer vision engineering and its AR product suite of powerful creation tools driven by AI. In the past, a programmer would have had to write a lot of code to create one simple effect, explains Pan, something a creator on the Snap platform can do almost instantly.
“If you'd asked me five years ago, about the things that we could do in computer vision – actually we're probably five times better today than then where I was expecting because there's just such an explosion in technology,” he says.
Snap has 350,000 lens creators, and they have built three and a half million lenses, which have been viewed by trillion times, which is phenomenal numbers for creator generated content, says Pan. Snap has 443 million daily active users, steadily increasing user numbers since 2019 and its October results showed a 15% increase in revenue year-on-year.
AI's business utility lies in the application layer
Ignasi Barri, global head of data and AI at GFT leads a team building advanced AI applications tailored for the financial services and manufacturing sectors. He notes that it is important for companies to identify the right use cases to leverage LLMs or Large Vision Models (LVMs), or a combination of both.
“The number of business opportunities that will be created at the application layer will far exceed what has been seen in the last couple of years at the infrastructure and service layers. This is where technology meets business needs, enabling companies of all sizes and sectors to break even and get a return on investment,” says Barri.
Specialisation and agility allow smaller players to target niche markets or solve specific problems with AI-driven solutions and while Barri admits he doesn’t have a crystal ball to predict the near future, he draws a parallel with the advent of cloud technology.
“When we began adopting cloud technology we only referred to the big players like AWS, GCP, and Azure, but this new paradigm eventually sparked sector-transformative value propositions from companies like Spotify, Netflix, Slack, or Shopify. The AI application layer could give birth to new giants, offering tools and experiences tailored to industries such as healthcare, education, and logistics,” he says.
Vijay Guntur, is CTO of software services giant HCLTech which offers its customers strategically designed AI products and applications. Guntur is on the coalface of AI transformation, advising customers on application development. He says there are many examples of ROI for both vendor and client at the AI application layer within vertical industries like healthcare and pharma.
“For example, the drug discovery process is very risky and unpredictable process, with many stages of approvals and clinical trials. AI can be used to predict early on in the cycle whether a drug is likely to go through the entire process only to fail delivering efficacy,” says Guntur.
While the AI boom may spawn giants, companies will see a ROI on developing their own applications. “A company that used to launch five new drugs a year, can now launch 13 new drugs a year. From a very narrow view of return-on-investment for that firm developing its own application, the payoff is obviously there,” says Guntur who advises business that they should be proactive and figure out ways to create ROI on AI for themselves. Either way, it is becoming increasingly clear that AI's business utility will truly come to the fore with further development at the application layer.
"Future AI unicorns will emerge from the application layer" was originally created and published by Verdict, a GlobalData owned brand.
The information on this site has been included in good faith for general informational purposes only. It is not intended to amount to advice on which you should rely, and we give no representation, warranty or guarantee, whether express or implied as to its accuracy or completeness. You must obtain professional or specialist advice before taking, or refraining from, any action on the basis of the content on our site.
