AI Infrastructure Thesis-Based Deal Tracking
Global League (G-L.Ventures) is committed to a data-driven top-down research approach to analyze market trends, landscape and dynamics for investors and startups. Our thesis currently prioritizes the intersection of AI infrastructure, energy transition, semiconductors, and national security.
Briefing:
Data-Driven Thesis-Based Tracking?
We curate and track a large number of companies and investors in focused theses. Our scope is global and across both private and public companies. In the US, big tech companies’ R&D investment outpaces that of overall US venture funding by a wide margin. This underscores the scale these companies play in dominating the market. (More: Magnificent Seven to Watch for the Next AI Era)
We filter companies on all available data from different sources. Curated companies and updates will be shared with Global League Club members. The purposes are:
Bottlenecks in AI Infrastructure?
Who is dominating our current AI era? Where are the bottlenecks in AI computing now? The public market can provide answers.
A new AI era needs new infrastructures, and building them takes time. AI servers and data centers are the “picks and shovels” for all the desired developments with generative AI, including AI software infrastructure, applications, and struggling efforts to find AI business models. Big funding from VCs or tech titans going into software has only increased the overwhelming demand for AI servers and data centers. Hardware is more a bottleneck than software now, and energy is also becoming a bottleneck. No bits can run without atoms and electrons moving. (AI Infrastructure Hardware and Software Accrue the Most Value in AI Stack)?
Companies are rushing to build AI data centers, and mentions of AI and data centers skyrocketed on earning calls this year. To elaborate on the term, Jensen Huang, CEO and founder of NVIDIA, has long maintained that the data center is the new unit of computing.
These new centers are being built with capacities of 100 to 1,000 megawatts, or about the same loads that can power from 80,000 to 800,000 homes, notes the Electric Power Research Institute in a 2024 white paper. In this paper, EPRI analyzes AI and data-center energy consumption and predicts that if a projected high growth rate of 10% per year continues, data centers will annually consume up to 6.8% of total U.S. electricity generation by 2030—versus an estimated 4% today. The International Energy Agency estimates that global electricity consumption by data centers could surpass 1,000 terawatt-hours by 2026, more than twice the amount used in 2022.
In this AI era, access to power is a differentiator. The number one operating cost of data centers is power! GenAI makes the data center demand go from a megawatt scale to a gigawatt scale. That adds extra stress to grids or is even prohibitive without some change. And, regulations will require disclosure of carbon emissions. Sustainability is more than a buzzword. Using a combination of grid-sourced power and a portfolio of other renewables such as small modular nuclear reactors, wind turbines and batteries is a possibility.?
Leveling up energy efficiency is happening in EVERY aspect of computing infrastructures, including new materials in different layers, advanced semiconductor techniques, compute resource utilization improvement, data center design and heat management, efficient machine learning architectures, and models, energy source integration and coordination strategies, and even using AI itself to optimize the power grid efficiency. (More: Next AI Infrastructures for New AI Decade)
Cooling accounts for 40% of a data center’s energy consumption. Everyone is talking about advanced cooling technologies such as liquid cooling, which has been used in hyperscale data centers. The liquid cooling market is expected to approach $3 - 5B by 2028 driven by AI data center builds that will bring significantly higher rack power densities. Liquid cooling accounts for roughly 2% of overall cooling solutions within the data center today and is expected to increase to almost ~30% by 2028.
Data centers have evolved from physical servers and large buildings to virtualized systems and now composable infrastructure, where resources such as storage and persistent memory are disaggregated from the server.?Therefore, networking, or interconnectivity, is becoming more important than ever. Broadcom is a leader in this landscape, but many emerging players are solving bottlenecks by offering complementary innovations for collaboration instead of competing with incumbent players. That’s a smart move. (More: The WHY behind Impressive IPO of AI Infra Startup)
Another key differentiator is semiconductor capability, including advanced packaging techniques, advanced manufacturing nodes, new dielectric or substrate materials, new circuit architectures or data processing techniques, custom chips, and different fundamental physics leapfrogging of the current PPAC (Power, Performance, Area, Cost) constraints. For example, Co-Packaged Optics (CPO) and Si Photonics are using light to replace electrons. (More: The WHY behind Impressive IPO of AI Infra Startup). The international semiconductor sovereignty race started quite a while ago because computing power is a national security issue.
Speaking of national security, cybersecurity is certainly at the top of the list. The double-edged digital sword of GenAI could shift the playing field in favor of both cybersecurity defenders and attackers depending on the circumstances. Where are startup companies leading the way in harnessing the technology? What are cybersecurity companies to watch in the quantum computing era?
Prioritized Segments in AI Infrastructure Tracking?
In the AI data center space, we prioritize tracking companies and pre-IPO early-stage deals in these emerging segments, though many overlap:?
We’ll keep an eye on relevant emerging opportunities across computing spaces. For example, quantum computing funding surged to a record high in 2023, bucking the broader venture downturn for 2 years in a row. More enterprises are running pilots with quantum startups. Interest in the intersection between quantum computing and AI is heating up as quantum advances can enable more powerful AI models down the road.
AI is a very broad term. Machine learning and deep learning have had meaningful breakthroughs in the past decade, which has led to the latest major emerging opportunity: generative AI (GenAI). GenAI has special infrastructure demands, but it has overlapped with other spaces such as high-performance computing (HPC), cloud computing, cognitive computing, quantum computing, supercomputer, edge computing, etc. Interestingly, cryptocurrency mining could be relevant as well. In general, computing capability is an arms race between enterprises and nations.
Data Tracked and Data Sources
Several variables are prioritized among tons of possible data in evaluating a startup.
Data sources could include market intelligence platforms (such as Netzeroi the leading climate innovation intelligence platform, Pitchbook, Crunchbase, etc.), DoD, DoE, SBIR, research organizations (such as RMI, Lux Research, etc.), innovation communities (such as National Technology Alliance), specialized industry media (such as Semiengineering), and industry expert interviews. Different market intelligence tools have different strengths. Although they are all incomplete, their values are appreciated. And, so far no “AI for VC” tools are usable after trying quite a few with inquiries for our thesis.?
Recent News to Review
Microsoft has partnered with investment giant BlackRock to launch the Global AI Infrastructure Investment Partnership (GAIIP), a fund exceeding $30 billion aimed at developing critical infrastructure for artificial intelligence advancements.
The collaboration includes Global Infrastructure Partners (GIP) and MGX, and is focused on building data centers and energy projects necessary to power next-generation AI technologies.
The investments will primarily target the United States, supporting the expansion of computing power and energy resources, with additional investments planned in partner countries.?
The goal of GAIIP is to unlock $30 billion in private equity capital initially, potentially reaching $100 billion when combined with debt financing.?
Oracle today announced the first zettascale cloud computing clusters accelerated by the NVIDIA Blackwell platform. Oracle Cloud Infrastructure (OCI) is now taking orders for the largest AI supercomputer in the cloud—available with up to 131,072 NVIDIA Blackwell GPUs.
Oracle founder and Chairman Larry Ellison told investors and media that it planned to build a gigawatt-scale data center that will be powered by three nearby SMR reactors.
Broadcom CEO Hock Tan has predicted his hyperscale semiconductor customers will continue building AI clusters for another three to five years, with each generation of machines to double in size.… Hock Tan reckons the silicon sales cycle is about to swing up, sharply, too
Constellation signs its largest-ever power purchase agreement with Microsoft, a deal that will restore TMI Unit 1 to service and keep it online for decades; add approximately 835 megawatts of carbon-free energy to the grid; create 3,400 direct and indirect jobs and deliver more than $3 billion in state and federal taxes.
Global League Club
There is a quality AI infra deep tech deal flow and closed social space for collaboration. Share this newsletter with investors you like to collaborate with.
Continue to read: Tracking AI Data Center Cooling Tech Companies