Over the past 10 years, investment in artificial intelligence has accelerated at a quick pace, reaching hundreds of billions of dollars.
3 Factors Holding Back the Return on AI Investments
- Much of the investment in AI infrastructure is geared toward the future.
- Outside of OpenAI, Claude and a few others, there is limited consumer adoption of AI tech.
- The tech is early in the adoption curve for enterprises, too. Adoption on a large scale is still on the horizon.
Economic returns, though, have yet to match the investments. Nascent technological breakthroughs such as large language models still need to be fully adopted within most enterprises and although the technology has seen one of the fastest adoption curves, currently the technology is costly to develop.
This pattern is typical for emerging technologies. For instance, sequencing the human genome initially cost $1 billion, whereas it now costs about $100.
While OpenAI has surpassed an estimated $3 billion in revenue, many other AI startups and ventures struggle to exceed the $100 million mark. The current market focuses heavily on developing foundational frontier models and technologies, enabling products like AI companions such as Friend.
In the AI wrappers space, where startups develop products around AI lab APIs, competition is fierce. These startups often struggle to exceed the $100 million revenue mark, even when fine-tuning models for specific use cases.
A major risk is the emergence of new AI models that can perform these specialized tasks inherently, potentially rendering the fine-tuned solutions of these startups obsolete. For instance, when ChatGPT came out, jasper.ai lost subscribers, resulting in staff cuts, and copy.ai now operates in an extremely crowded market. This challenge underscores the volatility and rapid evolution of the AI industry, making it difficult for smaller ventures to achieve significant market traction and differentiation.
AI and the Competition Gap
This big-vs-small company situation creates a considerable gap between major players like OpenAI, MidJjourney and Anthropic and the rest of the ventures within the industry. That’s because there is limited consumer adoption of AI technologies, outside of a few key products such as Claude, ChatGPT, MidJourney and Runway.
However, operating these models is capital-intensive, with rumors suggesting that running ChatGPT costs a staggering $700,000 per day. That’s not even including all the staffing and expenditure that goes into R&D and training costs of new models. The high costs and investments exclude many companies from competing.
This has raised concerns in the market and provoked certain rumors, one being that OpenAI might run out of cash within a year. While this seems unlikely, the company needs to keep attracting investment and expanding its operation so that it has a clear road to profitability. But that’s not the goal for now.
Investing in AI Infrastructure
Because AI development is still in its early stages, companies like Microsoft, Amazon and Google are leading the charge with substantial investments in AI and data center infrastructure. The VC ecosystem is highly active in AI investments, too. Firms like Sequoia Capital and Andreessen Horowitz are among the most active and prominent investors in the AI space, particularly in generative AI startups.
Investments in infrastructure ensure that AI labs can stay ahead by pushing out the newest models and remaining competitive. Building this infrastructure is crucial for the future, as it enables the development and deployment of even more advanced AI technology.
Investing in Compute
One of the main infrastructure components is compute, with investments potentially surpassing a staggering $1 trillion over the next few years. Major tech companies, including Microsoft, Google and Amazon, are heavily investing in this sector, with each data center costing around $2 billion to build. This field is still nascent, as companies are learning how to set up these specialized GPU data centers. These centers will be equipped with the latest chips, like the H100. However, these chips will quickly become outdated as more powerful chips emerge, requiring ongoing reinvestment to meet the increasing computational demands of new AI models.
While one can argue that certain labs have advantages in models, algorithms or data, competing in this space is challenging. Researchers often move between AI labs, transferring knowledge and reducing competitive edges. One of many examples is Dario Amodel, former vice president of research at OpenAI, who co-founded Anthropic in 2021. When it comes to returns on capital expenditure, what are AI labs and their investors really betting on?
The Future Cost of Intelligence
Although AI is not yet on the roadmap of all corporations, AI labs are counting on decreasing the cost of intelligence and its value for companies seeking to acquire it. Currently, companies invest heavily in recruiting top talent, which is a significant expense. While current AI models are akin to clumsy interns or junior employees, they are improving and becoming cheaper.
For example, OpenAI’s GPT-4o-mini is 97 percent cheaper for input tokens and 96 percent cheaper for output tokens compared with GPT-4. This reduction translates to a 97 percent decrease in the cost of a clumsy intern’s intelligence. Imagine if this intelligence reaches Ph.D.-level capabilities; the implications for cost savings and efficiency would be immense.
In the near future, digital workers, also known as AI agents, will collaborate with humans and other AI agents. Initially, they will automate mundane tasks, but eventually, they will handle higher-value activities. This shift could allow humans to focus on more significant problems, potentially reducing the need for as many human workers. Smaller groups of humans, supported by thousands of digital agents handling non-strategic tasks, could produce more valuable outputs and tackle complex issues more efficiently.
When considering the investment in AI, one might ask if capturing only a small part of the tasks that humans perform today will yield significant returns on investment. We can support this thesis by examining past innovations that optimized human productivity, such as electricity, the personal computer and the internet. These technologies revolutionized industries, leading to substantial efficiency gains and cost reductions.
Similarly, AI has the potential to transform various sectors by decreasing the cost of intelligence, thereby creating significant economic value and improving productivity across the board.
