Deal flow in 2024 has picked up significantly compared to last year, and AI deals have predominantly dominated headlines due to megadeals like those involving OpenAI, Grok, and others.
While later-stage companies have seen a bulk of the funding, seed activity in the space has also seen a spike, with a focus on AI hardware.
With a focus on early-stage startups, David Goldman, partner at Celesta Capital, has shared key insights with CI. Celesta Capital, a VC firm, actively invests in the deep technology space with key portfolio companies in AI, semiconductors, biotechnology, and cloud infrastructure, and it has funded over 100 technology companies with 29 exits to date.
David discussed why investment firms are taking an interest in AI-related sectors that can support the foundational AI ecosystem.
He commented on what role smaller players have in the industry, and how the entryways into AI markets allow for a more cautious take in investments.
He touched on the reason why the “hole” in series A and B rounds is causing many startups to solve for capital to remain viable. Goldman went on to talk about the AI hardware and infrastructure companies that are upending particular industries, as well as how to access valuation when it comes to AI solutions.
Our discussion with David Goldman is shared below.
Crowdfund Insider: Why have investment firms taken an interest in AI-related sectors that can support the infrastructure for the AI ecosystem?
David Goldman: Many firms have taken a more cautious approach to investing directly in foundational AI models and large language models (LLMs) due to the massive capital requirements, which often cost billions of dollars. This high-cost barrier makes it challenging for venture-backed startups to compete in this space, especially with open-source competition like Meta’s LLaMA model and Mistral.
Rather than investing directly in the foundational AI models, select VC firms have approached opportunities around the broader AI tech stack, such as hardware, infrastructure software, and enabling technologies. These technologies can help improve the training, deployment, management, and utilization of large AI models, providing a more feasible investment opportunity.
New solutions have emerged that seek to integrate within existing AI ecosystems rather than developing the foundational AI models themselves. By investing in enabling technologies, startups can help create a more robust and accessible infrastructure for companies to build on without the need to shoulder massive costs.
These integrators have already delivered and will continue to deliver unexpected, disruptive AI applications, similar to the impact of the iPhone and App Store.
Crowdfund Insider: What role do smaller players have in the AI ecosystem, and how do the entryways into these markets allow for a more cautious take in investments?
David Goldman: While large players like Nvidia have distinct advantages in certain AI applications, there are opportunities for smaller, more specialized players to succeed. Since the AI ecosystem is rapidly evolving, there are specific use cases and application needs that larger players do not address, creating openings for startups.
Another avenue for entering the space is by complementing existing organizations. This “win-win” approach, exemplified by Nvidia’s acquisition of Octo, allows smaller players to find a path to value creation without needing to challenge market leaders directly.
Crowdfund Insider: How can valuation be accessed regarding AI solutions? Is AI ready to deliver real value?
David Goldman: Quantitative value is difficult to evaluate for some of these AI startups. Predictive, recurring revenue streams can be highlighted in traditional SaaS business models to show a company’s trajectory. However, the impact of AI infrastructure can be exponential.
The value of AI technology integrators can come from their potential to complement an acquirer’s products and services, creating significant uplift and transformative impact. Due to the unique ways companies can leverage AI technology, it can be challenging to determine if an AI company is overvalued, as the value it can create for an acquirer may be “immeasurably large.”
Just as we are seeing AI getting applied to legal analysis and customer service, there is going to be another wave of capabilities enabled by this that people would not immediately think of. If you think about it, we had the launch of the iPhone, then you had the launch of the App Store, and then you had the capability for real-time tracking, like the very first thought no one had, was going to be let’s create Uber or Airbnb?
Just as the launch of these tools enabled unexpected, disruptive applications like Uber and Airbnb, we anticipate that the continued development of AI capabilities will similarly lead to unforeseen, innovative applications that entrepreneurs with unique perspectives and experiences will create.
Crowdfund Insider: What AI hardware and infrastructure approaches are being made to upend particular industries?
David Goldman: To support emerging AI products and services, the foundation for these technologies is being heavily revamped to address this new future. This shift can be seen with the transformation of the data center.
Traditional data center architectures were focused on CPU-based designs. Now, they are shifting towards GPU-based and accelerated computing architectures to support AI workloads. These new AI-focused data centers require adjusted cooling, connectivity, and control software compared to previous centers.
New industry approaches include:
-
- New interconnect companies and technologies, which are critical for the new data center architectures required to support AI workloads. This includes optical interconnect and accelerator links.
- Alternative power management technologies to help mitigate the massive power usage of accelerated compute.
- New software and firmware that improve GPU utilization and ROI for AI customers.