Growth of High Liquidity companies, part 3: AI Infra

In our list of growth champions among liquid private companies, five standout players belong to the AI infrastructure segment. These companies saw remarkable growth in 2023, and while 2024 has shown more tempered performance, their results remain impressive. Valuations have also surged, naturally drawing the attention of secondary investors.

However, AI infrastructure is one of the most complex sectors for investors. In this article, we explore the key drivers behind the growth of this segment, how it differs from software investing, and the strategies investors employ to manage the sector's inherent complexity.

The AI infrastructure layer is vast, but how significant is the liquid-private portion? It’s clear that AI, especially AI infrastructure, is emerging as the new "cloud," contributing heavily to public market returns in 2024.

According to Coatue, the AI infrastructure market cap has already reached around $6 trillion, with the infrastructure layer being the biggest winner in the AI race. Public companies dominate this space, naturally, but the sector’s massive TAM (Total Addressable Market) is also attracting a surge of early-stage startups.

However, scaling in AI infrastructure is particularly challenging for private companies. Moreover, investing in those companies is complicated for growth-stage tech investors, both primary and secondary.

As a result, we see a dichotomy: large public players on one side and a wave of early-stage startups on the other. In the middle, there are relatively few liquid private companies—just 13 liquid and semi-liquid players currently operating.

Secondary platforms like Launchbay are recognized as a one-stop shop for investors, offering critical insights such as company metrics, segment benchmarks, and funding data. This empowers secondary investors to apply a VC-style approach in their decision-making, an approach that works seamlessly in sectors where VCs are the primary investors. However, that’s not the case in AI infrastructure.
So how do we support investors in making informed decisions in such a complex sector?

This raises key questions:

• Who is investing in AI infrastructure, and where are secondary investors drawing their insights from?
• What are the fundamental differences between investing in software and infrastructure, and how are former software investors navigating the complexity of this segment?

Who is investing in AI infra?

Big tech companies are the largest spenders in AI infrastructure, both as investors and as major clients. Their capital expenditures (Capex) in this area have grown significantly, reflecting their deep commitment to AI-driven innovation and scale.
Take, for example, the Capex growth of Meta, Microsoft, Amazon, and Google—all of which have massively expanded their data centers, AI capabilities, and cloud infrastructure to support AI and machine learning applications.
On the investment side, industry giants like Nvidia, Cisco Systems, Dell, AMD, Intel, and Qualcomm are not only advancing their own AI capabilities but also investing in private companies in the AI infrastructure sector.

Dylan Patel, who runs SemiAnalysis, a leading publication and research firm on AI hardware, compares Big Tech's investment in AI to Pascal’s Wager.
“This is a matrix of like, do you believe in God? Yes or no. If you believe in God and God's real and you go to heaven, that's great. If you don't believe in God and God is real, then you're going to hell.
This is psychologically what's happening, right? Satya said it on his earnings call. The risk of under-investing is worse than the risk of over-investing.”

The second major cohort of investors consists of investment giants, often partnering with Big Tech.

Blackstone has become a leading global financier of data centers. Just last week, Reuters reported that Blackstone plans to invest $8.2 billion to develop data centers in Spain’s Aragon region, following similar moves by Microsoft and Amazon in the area. Earlier, Blackstone announced the acquisition of a large data center developer in the Asia-Pacific region for $16 billion.
Even before these recent deals, Blackstone’s data center portfolio was valued at $55 billion, according to CEO Steve Schwarzman in the company's July earnings report.
Blackstone isn’t the only one making waves in this space. BlackRock has partnered with Microsoft on a $30 billion data center fund.

Additionally, investment giants from the Middle East making significant moves in the AI infrastructure space.

Finally, VCs are relative newcomers to the AI infrastructure sector.

Looking at cap tables from the 2024 funding rounds of private AI infra companies, it's clear that a wave of generalist VCs—many of whom had previously avoided infrastructure and hardware investments—are now stepping into the space for the first time.

The main reason VCs are entering the AI infrastructure space is their strong belief in the market size and timing. Many point to frameworks like Coatue's AI S-curve to guide their investments, which outlines how the market is entering a phase of explosive growth and opportunity.

Additionally, and crucial for growth-stage investors, AI infrastructure companies are showcasing stronger business fundamentals compared to Model Layer players. CoreWeve reports 85% gross margin, VAST Data - 90%.

What are the critical differences between investing in software and AI infrastructure?

The first challenge lies in the sheer technical complexity of AI infrastructure. The semiconductor industry, in particular, represents one of the most intricate and vast fields in human endeavor. With so many layers and processes involved, it becomes difficult to fully grasp competitive advantages, scaling laws, or even to accurately assess technical risks.
Jon Y, who runs Asianometry, ‘the world’s best YouTube channel on semiconductors and business history’: “In some ways, nobody knows the whole stack. It's important to state that semiconductor manufacturing and design is the largest search space of any problem that humans do because it is the most complicated industry that humans do.”

Secondly, this immense complexity opens up numerous points for potential disruption. Given how many components and systems remain under-optimized, there’s a vast opportunity for breakthrough innovations at multiple levels of the infrastructure stack.
Dylan Patel: “there's a tremendous opportunity to bring breakthrough innovation simply because there are so many layers where things are unoptimized.”

Thirdly, there's a significant lack of publicly available data in the AI infrastructure and semiconductor industries. Much of the knowledge is tightly held within a few key companies, making it difficult to access or analyze. Unlike other sectors where white papers and public research are abundant, the semiconductor space operates in a more closed environment. This limits our ability to leverage AI or other tools to learn and understand the intricacies of AI infrastructure from external sources, adding another layer of difficulty for investors and innovators alike.

Fourth, the AI infrastructure sector's dependency on big tech companies is both a key to its success and a source of potential risk. Big tech firms play a crucial role as both major investors and primary clients, driving growth and demand. However, this reliance can also create vulnerabilities. Striking the right balance between big tech involvement and diversifying the client and investor base is essential to mitigate risks and ensure long-term stability in the sector.

Last but not least, the sector faces significant geopolitical limitations and risks. A prime example is Cerebras’ recent IPO delay, which was caused by a national security review of its foreign investments. Such regulatory scrutiny highlights the growing geopolitical challenges that AI infrastructure companies must navigate, potentially slowing growth and complicating expansion plans in the global market.

How are VCs navigating this complexity?

VCs need to establish new metrics and evaluation criteria tailored to AI infrastructure companies. This should involve evaluating the scalability of AI models, the efficiency of compute resources, and the potential for breakthrough innovations. However, these frameworks are not developed yet. When discussing their AI infrastructure investments, VCs rarely share detailed methods or highlight specific technological advantages of the companies they choose. Instead, they tend to focus solely on their belief in the total addressable market (TAM).

Also, in the absence of relevant frameworks, the recent VC-favored data-driven approach to due diligence is not applicable in this sector.

First applicable strategy is just following smart money. VCs engage more in more co-investments or partnerships with strategic investors who have deep technical expertise in AI.
This strategy has gained such popularity that companies often need to limit the number of investors involved, as seen in the case of OpenAI. However, the secondary market continues to provide alternative avenues for participation.

The second strategy is to learn through small-ticket investments. This approach entails making smaller investments to secure a "seat at the table," which provides valuable access to knowledge about abstraction layers. This includes insights into industry-specific challenges, market dynamics, and technological advancements. This strategy is used with smaller funds.

Finally, VCs are adopting a strategy of making shorter-term bets on the most liquid tickets. This approach resembles that of public investors, leveraging the recent developments in the secondary market to capitalize on available opportunities.

Other articles:

Growth Benchmarks for High Liquidity Private Companies: Part 2, AI at the Model+Application layer
Growth of private companies in High liquidity sector: part 1, SaaS