SaaS Disruption: “We’re Seeing Clear Winners and Losers”

More than three years after the launch of ChatGPT, the question is no longer whether OpenAI will dominate the AI sector (it won’t), but rather who will ultimately be the winners and losers.

In this interview, Zac Gill, Equity Research Analyst at Jennison Global (the company invested in Nvidia, Microsoft, Amazon, Netflix or Meta), offers insights into the key economic developments surrounding LLMs and related technologies.

How is the team thinking about AI exposure across the portfolio, and where are you seeing the most compelling opportunities?

Zac Gill: We’re approaching AI comprehensively across infrastructure, software applications (both enterprise and consumer), and select adjacencies. Our framework emphasizes flexibility given the rapid pace of change in this space.

On the infrastructure side, we remain constructive. There’s still a significant compute shortage affecting both training and inference workloads. Companies are throttling AI products significantly below their potential capacity, which tells you something about underlying demand.

Within semiconductors specifically, we’re focused on where the acute shortages are building—primarily in clean room capacity (wafer manufacturing) and memory. We see attractive opportunities in semiconductor capital equipment makers like Lam Research and ASML, where backlogs are just beginning to inflect. TSMC remains well-positioned given the shortage in advanced node logic manufacturing and their lack of real competition. NVIDIA continues to benefit from being both the first-choice buyer and the buyer of last resort for compute capacity, and the valuation is now more reasonable than it has been recently.

On the consumer side, we see Google and OpenAI as the clear leaders in bundled platforms that will serve as comprehensive AI assistants. Google has done an excellent job catching up, and it’s a significant position for us. The number one AI product in the world by far is actually Google Search—there are billions of people using AI through Google who never downloaded Gemini and don’t show up in the ChatGPT user trackers. Google is already monetizing this effectively with better conversion rates than traditional search.

What are Jennison’s views on recent volatility in the software industry?

In software, we’re seeing clear winners and losers emerge. Jensen Huang’s observation that “AI is the new software” is playing out as he described. We’re moving from the deterministic, rules-based, software to contextual, implicit and flexible interfaces. This shift is creating opportunities but also challenges for traditional seat-based software models.

We’re deliberately avoiding software companies that are in “the eye of the storm.” We are avoiding vertical software-as-a-service (SaaS) companies and traditional seat-based application software companies. Instead, we favor the enablers. The enablers are software companies helping enterprises integrate AI safely and effectively, like Cloudflare, CrowdStrike, Palantir, and data infrastructure plays like Snowflake and Datadog.

Given the rapid evolution we’re seeing in frontier models and the proliferation of open source alternatives, how do you think about the monetization potential and competitive positioning for model providers?

This is one of the key dynamics we’re monitoring closely. We’ve learned that this isn’t a traditional network effect business like we initially hoped—using a model more doesn’t make it meaningfully better, which is different from how search evolved. And switching costs are currently quite low.

However, we think competition will ultimately come down to three factors: trust, product quality, and ecosystem integration.

Trust will become increasingly important as these platforms gain access to email, calendar, medical records, and payment information. There’s a significant trust benefit that accrues to established players or companies that have built credibility. This represents a meaningful barrier for new entrants.

Product differentiation matters more than raw model performance. Most users can’t distinguish between the latest generation models. What will matter is which platform has the best features, the most seamless functionality, the best integrations, and can anticipate user needs most effectively. Google has significant advantages here given their existing ecosystem.

We expect switching costs to rise over time—more like the Apple ecosystem than Instagram’s pure network effects. It won’t be impossible to switch, but it will become increasingly difficult as users integrate these agents more deeply into their workflows.

On monetization, we expect advertising will be the primary revenue stream for consumer products, potentially supplemented by transaction-based revenue. The subscription model alone isn’t sufficient for profitability. But advertising is already working well in Google’s AI-enhanced search with better conversion rates than traditional keyword search.

On the enterprise software side, how do you see the traditional software companies navigating the transition to AI?

The most likely outcome is that the large incumbent software companies will adapt and remain significant players in the AI-enabled software landscape, similar to how Microsoft, SAP, and Oracle navigated the cloud transition. Enterprise customers aren’t replacing cloud-based core platforms like Salesforce with internally-built on-premise alternatives in the near term.

However, transitions like these take time to play out fully.

What we expect to see is seat-based software growth will moderate because enterprises don’t need as many licensed users when AI improves productivity. Margins will face some pressure because implementing AI effectively requires investment (even with efficient open-source models, inference costs need to be managed). Software companies will need to absorb some of these costs while delivering value to customers.

The recent volatility in the software space is driven by uncertainty surrounding how these dynamics will play out. As long-term investors, our job is to look past the short-term noise that is currently affecting the entire space and focus on the fundamentals that make companies well positioned for future success. For example, some companies that got caught up in the broader software sell off are likely to be winners longer term in our view. And we’re focused on distinguishing between companies facing structural challenges and those experiencing temporary sentiment headwinds.

How are you thinking about the capital intensity of this buildout and the financing dynamics, particularly around some of the leading AI companies?

The AI buildout requires significant capital, but we think the financing will come together through a combination of sources. The model will involve multiple parties like cloud providers, infrastructure partners, and credit markets, each of which will share the capital requirements. This is similar to how early cloud capacity was built through partnerships rather than any single company bearing the full cost.

These infrastructure projects will likely be structured as traditional discounted cash flow-based investments with predictable revenue streams, similar to data center REITs and co-location facilities that enabled the cloud transition.

In our view, the key factors to monitor are execution and revenue growth. The leading AI companies are entering large, lucrative markets and have been growing at impressive rates. If they continue executing on their product roadmaps and expanding into new verticals, the financing should follow.

Companies with strong balance sheets like Google (Alphabet) don’t face meaningful financing constraints. For the broader ecosystem, we’re monitoring how capital flows into the space, and thus far the appetite from both strategic partners and financial investors remains robust.

Does the emphasis on product features over raw model performance change the investment case for infrastructure?

LLM model quality still matters, particularly for complex reasoning and decision-making applications, and especially while switching costs remain relatively low. But as product differentiation becomes paramount, the dynamics are expected to shift.

That said, we see continued investment in training better models given the significant long-term opportunity. It’s also important to note that training isn’t limited to text-based models anymore—there’s substantial work on video, virtual environments, and other modalities.

The key insight is that even with current model capabilities, there’s significant work ahead in better utilizing what we already have. This is why we believe inference could ultimately represent a larger market than training.

Importantly, the compute shortage affects both training and inference. On the consumption side, AI products are throttled well below their potential capacity. As compute becomes more efficient and cost-effective, we should see proliferation of inference workloads across enterprises. Organizations will run AI in the background for countless applications such as security, automated processes, and analytics. And it makes sense that these companies use appropriately-scaled models for each task.

As a result, even as efficiency improves, we expect this dynamic to create sustained demand for compute infrastructure because the addressable use cases should expand substantially.

What are the key factors you’re monitoring as the AI theme continues to develop?

We’re tracking several important indicators across the ecosystem.

The first is user adoption and engagement. These metrics remain critical, particularly for the leading consumer AI platforms. Growth trajectories and product iteration cycles give us insight into commercial traction.

Developments in China are also important to monitor. There’s significant AI talent and innovation happening there, and the resource-constrained environment is driving interesting approaches to model efficiency. China has led in certain technology themes previously, so it provides useful perspective.

Enterprise adoption patterns are also very important to monitor, and we’re focusing on how quickly businesses integrate AI capabilities and the types of applications gaining traction. This informs our views on which enablement and infrastructure companies are best positioned for the future.

We also view capital formation and partnership announcements in the space as positive indicators. When a major infrastructure project moves forward, it demonstrates continued conviction and financial support for the AI buildout.

Product launches and capabilities from leading AI companies help us assess the pace of innovation and competitive positioning across the landscape.

Overall, we remain positive on the investment opportunities across the whole rapidly evolving AI ecosystem, while staying flexible and focused on identifying the leaders within each part of that ecosystem.