Insights | What Investors Should Know About AI

Jennison's Owen Hyde discusses how investors may think about the long-term investment opportunity as AI adoption evolves and matures. In Jennison's view, "we are still in the opening stages of an AI–driven investment cycle—one we believe could reshape the global economy."

They don't think the ride, however, will be linear. In 2025, DeepSeek reignited fears that powerful, low‑cost models could pressure pricing and disrupt monetization. Early 2026 brought another wave of volatility, as investors reassessed what AI means for software business models and long‑term profitability. Against this backdrop, Jennison's Owen Hyde, CFA, a large cap growth equity research analyst focused on technology companies, shares his perspective—what's driving the volatility, and where Jennison sees the potential opportunities materializing over the long term.

The Current State of AI: CapEx and Concerns about a Bubble

Q: We hear about the extraordinary amount of CapEx being directed toward AI. Why is this happening?

A: At a high level, AI requires a different kind of computing infrastructure than the CPU, or central processing unit, ‑based architecture that underpinned the build‑out of traditional cloud platforms. Training and running modern AI models aims to rely on accelerated compute (GPUs, or graphics processing units, and custom AI chips), and that drives major investment in chips, networking, power, and data‑center capacity.

Fundamentally, AI seeks to change how people interact with computers. Tools like ChatGPT highlighted that you no longer need to be a programmer to get value, you could interact in natural language and receive natural‑language outputs. That's been expanding rapidly from "Q&A" into AI agents that aim to do work on your behalf.

It's also helpful to separate training from inference. Training is the long‑running process of teaching a model using large datasets; inference is when the trained model responds to a prompt (for example, generating an answer or completing a task). A key development is that inference is increasingly the dominant driver of compute demand—reflecting rising usage and the growth of enterprise applications.

Q: Do you see evidence that AI is a bubble?

A: I don't see evidence of that today. We certainly acknowledge that the CapEx spending is large and may feel uncomfortable, but what matters is whether capacity is being utilized and whether we see real product market fit in consumer and enterprise end markets to soak up the incremental capacity. In addition, adoption still appears early relative to what could come next—especially as more "agentic" systems (AI that takes actions, not just answers questions) become mainstream, which could be significantly more compute‑intensive but may accelerate the productivity of nearly every person and company in the world.

Many people concerned about a bubble in AI may find it difficult to understand the spending because they appear to be missing the exponential improvement we see in several new areas such as AI systems' ability to carry out tasks on a computer and handle long‑form work. If you assume progress is linear, you may underestimate how quickly capabilities can expand, and why capital spending can remain elevated as the opportunity set grows. Conversely, we also acknowledge that humans may only adopt new AI/software tools so fast, and we are cognizant of the fact there is a notable gap between the new model capabilities and the way the majority of consumers and enterprises use the technology now. Even if model progress was to stop today, there would still be many years of AI‑driven software product improvements and incremental adoption.

Q: What catalysts could convince skeptics that the spend is worthwhile, especially for hyperscalers and AI labs?

A: For hyperscalers, the path to proof could be more direct: AI demand can show up in revenue growth, backlog, and margin outcomes within cloud businesses. For model providers, it may take longer because there's more uncertainty around long‑run economics, competitive dynamics, and how expectations are set as these businesses mature.

Software: Where We See Opportunity vs. Risk

Q: Can you provide context for the recent volatility in the software space? What long‑term factors should investors be paying attention to?

A: We expect the market to become much more selective within software. In our view, the long‑term debate is not "does AI make software irrelevant," but rather what types of software benefit from AI. In application software (traditional SaaS), a key risk is that an AI‑driven "context layer" could sit on top of systems of record (e.g., CRM or HR systems), changing how end users could interact with those platforms and potentially pressuring pricing or stickiness for some software providers. Conversely, some parts of cybersecurity and infrastructure software may become direct beneficiaries of the increase in the capabilities and volume growth in software and agentic systems.

Q: Where is AI already changing workflows—and what's next?

A: We've seen the most visible early impact has been in software development. We're seeing strong product market fit for AI programing tools, and new models that are exceptionally strong at coding at larger scale with improved accuracy. Models, when given a good "harness," may accomplish tasks with higher level directions and complete tasks with less direct human intervention. This has changed the job of the software developer, flattening the organizational structure as specialists now become more generalist managers of complex AI generated systems. Coding is now done in natural language with iterative workflow. This has dramatically accelerated development cycles, increasing the volume of software created.

Q: Where is AI looking to create attractive long‑term opportunities?

A: In our view, areas leveraged to growth in total software being built and deployed, particularly infrastructure software and the public cloud, remain especially attractive.

We believe the cloud businesses within hyperscalers could exceed expectations if capacity remains highly utilized and AI workloads continue to scale, particularly as AI moves from experimentation to production use cases. When it comes to cloud businesses, our "north star" is utilization. To date, we continue to see very high utilization of deployed compute and little evidence of slack capacity. Over time, the adoption of more agentic systems could further increase compute intensity and reinforce demand for cloud infrastructure.

We also view cybersecurity as an important area. More software creation (including by non‑professional developers) and new AI‑enabled workflows can introduce vulnerabilities and shift security requirements. We believe that many organizations prefer to buy proven cybersecurity products rather than try to build them internally.

Q: Beyond the pace of change, what has been most surprising as you assess AI's implications?

A: One major shift is how AI‑driven productivity in software development could reshape competitive dynamics across many industries, not just enterprise software. If building digital products becomes materially easier, the traditional advantage of having large engineering teams may matter less. In many cases, moats, or a company's ability to maintain competitive advantages to protect its long‑term profits and market share from competitors, may increasingly depend on other factors such as network effects, physical infrastructure, proprietary data, or structural "choke points" in an industry.

Key Risks and Perspectives on Current Trends in AI

Q: What are the biggest risks to the AI investment story, both for companies and investors?

A: One risk we monitor is whether a major step‑change in model efficiency reduces compute intensity, or a measure of how much computational work is performed relative to the amount of data accessed or moved in a computation, more than expected. If an algorithmic breakthrough or a new model architecture materially increases the amount of work you could do with the same compute, it could shift the industry from being compute‑constrained to having slack capacity, potentially changing the trajectory of infrastructure demand.

We still see very high utilization of deployed compute and continued constraint across parts of the supply chain, which is an important data point. Another risk to watch, particularly for model providers, is how durable "frontier" advantages are in a hyper competitive environment that includes open‑source alternatives. If progress slows materially, lower‑cost options could look more compelling versus premium models.

Q: What's your view on hyperscalers building custom AI chips? Does that threaten incumbent chip providers?

A: We're seeing more specialization. Many custom chips seem to be designed primarily for inference, and only a handful of companies have the scale and engineering teams to develop them. These efforts may gain share over time, but the overall market is also expanding rapidly. In addition, software ecosystems and tooling remain important barriers to adoption for some custom architectures. In practice, a large portion of the market still relies on general‑purpose GPU, or graphics processing unit, capacity provided directly or through cloud platforms. Both GPUs and custom ASICs, or application‑specific integrated circuits, have a place in the new accelerated compute stack.

Q: Where do you think long‑term value accrues across the AI stack?

A: One open question is whether model providers can move beyond general chat interfaces into workflow‑specific enterprise products, and whether they could scale their economics over time. Today, selling model usage may look structurally different from traditional enterprise software, including lower gross margins driven by compute costs. A key question is whether these businesses can scale, raise sufficient capital, and build product suites that likely improve unit economics and capture more value. We believe enterprise AI adoption will be a very important metric to watch in '26/'27, with increased focus on new verticals outside of coding.

More broadly, I expect AI to become something people interact with as frequently as they do their phones today, and the gap between what the models are capable of and how they're currently used remains very wide.

Owen Hyde, CFA

Managing Director

Owen Hyde is an equity research analyst covering technology companies. Before joining Jennison in 2021, he was an analyst at J.P. Morgan Asset Management where he focused on the technology sector. Prior to J.P. Morgan, Owen was an equity research analyst at Morgan Stanley and Pacific Crest Securities where he covered internet and software companies, respectively. Owen earned a BS in economics from the University of Oregon and is a CFA charterholder.

Important Information

The views expressed herein are those of Jennison at the time the comments were made. They may not be reflective of their current opinions, are subject to change without prior notice, and should not be considered investment advice. These views are not necessarily those of the Harbor Investment Team and should not be construed as such. The information provided is for informational purposes only.

Past performance is no guarantee of future results.

Stock markets are volatile and equity values can decline significantly in response to adverse issuer, political, regulatory, market and economic conditions. Investing in companies linked to artificial intelligence and technology involves risks, including rapid innovation cycles, intense competition, valuation volatility, regulatory uncertainty, uncertainty regarding monetization, infrastructure overcapacity, and potential disruption from lower‑cost or open‑source technologies.

Capex, or capital expenditures, refers to the money a company spends to acquire, upgrade, or maintain long‑term physical or fixed assets that provide value over multiple years.

CPU‑based architecture refers to the core component of a computer responsible for executing instructions, performing calculations, and controlling data flow between system components. Even in a simplified architecture, it consists of several key parts working together to process information efficiently.

Jennison is in a strategic partnership with Harbor Capital.

Jennison is not affiliated with Harbor Capital.