I want to thank everyone for your inquiries over the past couple of months. I took a break from regular blog posts, in order to focus on a few side projects. These included new research areas, doubling down on angel investing and another guest lecture for the London Business School. A common thread within these pursuits was exploration of current trends associated with AI. Without stating the obvious, AI is touching every industry and will have huge ramifications for software infrastructure going forward. This trend is too large to treat as an extension of the existing growth vector for software, which has largely projected through the rise of the Internet and mobile device proliferation.
While today’s AI is the outcome of years of steady, but familiar, efforts in machine learning, it has become significantly more accelerated, broad-reaching and impactful. We are entering an exciting period where software is inflecting from helping humans retrieve information more quickly (traditional Internet) to actually performing tasks on our behalf (generative AI, co-pilots and ultimately autonomy). This shift dramatically alters the value proposition and economic benefit, particularly as AI-driven systems start to perform work that has traditionally been addressed by salaried humans.
The ramifications will be far-reaching for society and naturally create a number of investment opportunities. I am considering these within the context and through the lens of software infrastructure. At the simplest level, AI training and operations will drive a step-change in the consumption of compute and data. The same technology functions that benefitted from the growth of mobile devices will see increased demand driven from enablement of new AI services, except that the magnitude will likely be 10x the influence of mobile.
As it relates to this blog, I plan to continue the same scope of coverage, but will account for the impact of AI on software development and infrastructure. Look for upcoming posts on the business implications for the companies that provide data infrastructure, application delivery, security, observability and core hosting. It is an exciting time to be in the software space – even surpassing the acceleration we experienced during Covid.
Implications for Software Infrastructure Companies
Fortunately for investors, the market has taken notice of these opportunities over the last few months. Combined with a curtailing of optimization headwinds, we have seen many of the companies covered on this blog revisit their 52 week highs. These include SNOW, NET, DDOG, MDB and ESTC, along with security companies like PANW, ZS and CRWD. While I rarely take credit for prior calls, these trends reflect much of what I posted earlier in 2023. The optimization impact curve that I charted in my review of Hyperscaler results from Q2 is largely playing out as expected.
As we transition into 2024, I expect these trends to continue. There will likely be some volatility, but the general forces will persist. Optimization headwinds will taper off, as there will be fewer workloads remaining that were over-provisioned during Covid-19. The stabilization of macro factors, marked by an expectation for stable if not lower interest rates, will give enterprises confidence to increase their investment in IT. Digital transformation projects will be dusted off and the cloud migration will continue.
Further, I think that commercial use cases from AI will emerge across multiple industries, extending beyond the early success of the current breed of co-pilot offerings. Co-pilots to date have largely leveraged publicly available content on the Internet and in open code repositories. As enterprises with proprietary data and industry-specific knowledge move out of experimental mode into production AI services, they will launch new monetized “expert systems” that create competitive advantage, allowing their customers to save time, lower costs and capitalize on totally new capabilities.
The majority of these new AI-driven services will still be delivered over the Internet, providing another tailwind for cloud hosting demand. And with data at the core, enterprises will invest in modernizing their data infrastructure stack. Data management will experience growth across a few dimensions – more data sources, longer retention and faster updates. All of this will drive incremental demand for associated data service providers. Among independents, this favors SNOW, MDB, ESTC, CFLT and Databricks, as some examples.
Another important factor to consider will be the impact of generative AI on software developer productivity. Copilots for developers, like those available from GitHub and GitLab, are reducing development times by 30%-50% or more. This has two impacts for enterprises. First, they can work through their backlog of software projects more quickly. Most enterprises have a long wishlist of digital transformation initiatives that are prioritized and awaiting developer resources. As coding velocity increases, internal software teams will complete these projects in a shorter amount of time. This should drive a pull-forward in the need for application hosting services.
Second, enterprise IT teams will be able to complete more work with the same staff. While the near term effect will be an increase in project completion velocity, once enterprise software teams catch up somewhat on their backlogs, budget allocations for the following year will be adjusted to capture some of this efficiency. This means that hiring rates for new developers will slow down and staffing costs will contribute a smaller percentage of the overall IT budget. This allows money to shift into other areas, like software hosting and supporting infrastructure. IT teams will manage more data to power increasingly sophisticated and proprietary AI services, while continuing to leverage new tooling to increase developer productivity.
Enterprises will deliver more digital transformation projects, increasing their consumption of software infrastructure services, including hosting runtimes, data processing, delivery, observability and application security. This incremental cost will be offset by savings in headcount growth. As IT budgets increase again, less money will be allocated to staffing costs and more can be spent on application infrastructure. This will benefit those companies providing cloud and software infrastructure services.
London Business School Guest Lecture
Incorporated into my research, I prepared a guest lecture for the London Business School (third one in two years) in which I discussed investment opportunities for AI. To help consider the impact of AI on various components of the software stack, I proposed a model of expanding concentric circles over time. The center circle and first set of companies to benefit are those delivering core infrastructure – providers of compute (GPUs), model hosting and runtimes (hyperscalers). Examples are Nvidia, AWS, Microsoft Azure, GCP and AMD.
Author’s Presentation to London Business School, December 2023
As time progresses and the technology matures, the impact of AI investment expands outward to data infrastructure services next. These vendors deliver services that allow enterprises to prep, secure and distribute the inputs and outputs of AI models. This primarily takes the form of data. As Snowflake’s CEO quipped on their Q3 earnings call, “there’s no AI strategy without a data strategy.” Data infrastructure providers are the key beneficiaries here – SNOW, MDB, CFLT, ESTC, Databricks, etc.
Moving further outwards, all sorts of SaaS businesses will be able to deliver new capabilities and lower costs by incorporating AI into their existing offerings. Leveraging their proprietary data sets and industry knowledge, they will develop unique services for their customers that add significant value. This will support new products and extensions of existing ones, driving incremental revenue. There are many examples of beneficiaries. I covered just one in my talk in the interest of time, which was Intuit (INTU), but the concepts can be applied to any industry.
Finally, we have to acknowledge that AI will have the power to radically transform existing industries in the physical world. While it is outside of my coverage, I briefly explored how AI will accelerate the capabilities of autonomous devices (robots), rapidly improving their ability to perform complex tasks. This will have significant implications for manufacturing, transportation, health care, consumer services and most other industries. The key is that these capabilities will be enabled by large amounts of data and require significant compute. These costs will be justified by a step-change improvement in capabilities.
I used Tesla as an example use case here. While Tesla currently derives most of its revenue from selling cars, I think the majority of their earnings in the future will be generated from new services enhanced by their leading investments in AI and data collection. Examples of new AI-driven services will include FSD (sold internally and licensed externally), robo-taxis, the Optimus robot and other advancements in manufacturing operations, energy distribution and transportation. Given that value with AI systems is primarily conferred to those companies with the most proprietary data, Tesla has rapidly assembled the largest relevant data set for autonomous driving. That approach can be applied to other activities in the physical world.
I expect Tesla’s cycle of data collection, AI model creation and new service development to continue. In the same way that selling books allowed Amazon to eventually create AWS, Tesla will leverage their start in EV manufacturing to launch a bevy of new services. Beyond Tesla, we will see other companies, currently operating in mundane industries, leverage AI to supercharge their existing product offerings and launch brand new ones.
With that overview, I plan to resume regular coverage in 2024. As AI exerts a significantly larger influence on software infrastructure and services going forward, I expect to revisit many of these themes in future posts. In the meantime, enjoy the final whispers of 2023 and look for more posts in 2024. In full disclosure, I am considering some form of monetization to offset the costs associated with running the SSI service (hosting, content subscription, financial feeds). Finally, as in the past, if you are interested in a copy of my deck from the London Business School presentation, you can email me at analysis@softwarestackinvesting.com. I am happy to share it on request.
NOTE: This article does not represent investment advice and is solely the author’s opinion for managing his own investment portfolio. Readers are expected to perform their own due diligence before making investment decisions. Please see the Disclaimer for more detail.
I am very gratified to read your new message, Peter. Selfishly so I acknowledge because now I know you are okay, still with us, still engaged, and that 2024 promises many more prototypical updates and commentaries from you.
Your news that the ‘macro headwinds of optimization’ have not dulled the London Business School’s esteem for your insights heartens me, as it should all investors, for obvious reasons. Godspeed.
Happy New Year 2024, Peter! May it bring good health and prosperity for you and your family – and for us all.
David
and the same from me!
Thanks for your kind works, David. I appreciate it and wish you a great 2024 as well!
Hi Peter,
Glad you back to posting here. You’re example of transparently is encouraging. I just can get myself to give Common Stock or Plaid or anyone my broker Username and Password. What am missing? How is this not dangerous?
Thanks, Jason. Regarding sharing brokerage account details with a third party, I agree that it represents a risk. Most brokerage accounts employ additional security measures for actions like trades and withdrawals, usually checking the device of origin and requiring a PIN or text response to complete a transaction. My particular brokerage account for Commonstock’s tracking is read-only and locked down, but agree that this may not be suitable for all accounts.
Gartner estimated that 85% of AI and ML projects fail to produce a return for the business (Forbes, Jan 2023, from a search result). Is that percentage likely to go up or down much? I’m thinking that maybe there’s a rush to use GenAI by many enterprises with little experience of using any kind of AI, which could make failure even more likely. From me, it’s just a hypothesis.
A few thoughts here. First, I think that enterprises will try a lot of experiments before settling on the few initiatives that will receive additional investment and have an expected return. The Gartner statistic may include every experiment. Second, I agree with the sentiment that a lot of companies will start AI projects without a good sense for how to manage, execute and measure it. There probably are a lot of failures. But, I think the percentage that succeed likely become large drivers of revenue. So this could be a bit of the 80/20 rule, where 20% of the new AI initiates generate 80% of the benefit.
Thanks!
Thank you Peter for your article, and glad to hear you’re doing fine. Looking forward to more articles from you in 2024.
I’m also interested in learning if you have made any adjustments to your portfolio as you learn more about AI, and its implications on companies.
Hi Prateek – Thanks for the feedback. In terms of my portfolio, it is largely made up of the same core companies. These include NET, SNOW, MDB, CFLT and DDOG. I think all of these companies will benefit from the AI tailwinds discussed in this post. Additionally, I have starting building positions in NVDA and TSLA for similar reasons.
Thank you Peter.
Why would you invest in NVDA at this price?
I haven’t valued this company as I don’t know enough about it, but usually when a stock rises 300% in a year it’s nearly always going to produce terrible future returns because the market has either over-hyped it or all of it’s future cash flow is priced in.
This looks to me like the equivalent of investing in Amazon in 1999.
> The optimization impact curve that I charted in my review of Hyperscaler results from Q2 is largely playing out as expected.
As you said below this, this is mostly a result of inflation falling faster than expected which has reduced their WACC, not due to optimisations really, although I agree the optimisations will produce better returns for them in the future and it’s a good thing they did this.
> Copilots for developers, like those available from GitHub and GitLab, are reducing development times by 30%-50% or more
I’m a senior developer and use CoPilot and ChatGpt, I would say it’s saved 20% development time max for me. I don’t think these tools are anywhere close to 50% time saves for non-junior developers, although they are absolutely amazing and I wouldn’t go back to not having them.
50% time save would be an insane productivity gain that would take decades to materialize.
The rest of your blog I agree with.
Hi – thanks for the feedback and you raise great questions. Here are a couple of thoughts:
1. NVDA. While the stock has appreciated a lot in 2023, that followed a 50% drop from late 2021 to late 2022. Relative to the prior peak 2 years ago, the stock is up about 50%. At this point, I think the valuation is reasonable, in spite of the growth. The PE ratio for the prior fiscal year should finish at about 40. I think growth estimates for the next fiscal year (FY2025 / CY2024) are reasonable, bringing the forward PE to 24. The big question will be around the sustainability of growth from there, but FY2026 (calendar 2025) calls for just 20% revenue and earnings increases, per analyst estimates. Optimistically, I think those are beatable.
2. Optimization. I think we are saying the same thing. My point around optimization and hyperscaler performance is that I think a lot of cloud workloads were over-provisioned during the Covid surge of 2020-2021. By 2022, with macro pressure to reduce IT budgets, customers had plenty of opportunities to reduce workload resource consumption for those overly sized clusters. I think a lot of XL instances got downsized to M or L, cutting spend immediately. As the low-hanging fruit of these optimizations gets worked off, then the large negative impact on total resource consumption moderates. I think this is causing hyperscaler growth rates to trough, after dropping rapidly over the past year.
3. Developer productivity. Fair point – this data is anecdotal and based on individual experiences. Even a 20% productivity boost, though, would have a large impact on future budget allocations. I am anticipating a shift where a larger portion of IT budgets can be allocated to the infrastructure needed to host more applications, offset by savings in salaries. Also, some companies measure overall product development productivity beyond developer work. Intuit, for example, discussed at their September Investor Day how homegrown Gen-AI tools yielded speed increases of 2x to 6x for end-to-end product development processes, like translating new tax rules into code (6x) and data analysis tasks (2x).
Hi peter, great post as usual!
Question about commonstock, is it not up to date anymore?
As I see you mention NVDA and TSLA here as positions. Thanks
Hi – thanks for the feedback. The Commonstock feed broke when my brokerage account at TDAmeritrade was acquired by Schwab. I have to look into how to support API access for Commonstock to access it. I may just shift back to posting the portfolio make-up manually. In the meantime, the position distribution is roughly the same, with some smaller new positions in NVDA and TSLA. All total, it includes NET, SNOW, DDOG, MDB, CFLT, NVDA, TSLA.
Thanks a lot for the transparency!
Hi Peter – glad you’re back and happy new year! Could you please share some quick comments on CFLT? Per their earnings call their loss of two large customers (maybe more this quarter) and macro uncertainty will cause lower growth and they don’t know how long it’ll last. There’s real opportunity cost waiting for them to recover from this drag. Do you still plan to hold? Thanks.
Hi Chris – I am actually working on a large update for SNOW, MDB and CFLT now. The short version is that I am still holding a mid-sized position in CFLT. While it might take a quarter or two, my thesis is that Confluent will get growth back on track in second half of 2024 and exit Q4 with about 30% annual growth. Plus, profitability measures (like EPS and FCF) will improve noticeably. If both of those happen, then the stock could return to the $30-$35 range by end of year. That would be about a 50% return from today. In terms of opportunity cost, you could compare that risk/reward with other companies, but I feel like the rest of software infrastructure has appreciated a fair amount already, while CFLT has not. Yet, the underlying drivers are similar (unless something existential is happening with Confluent).
Hi Peter – Thank you for continuing to freely share your written analysis. Your writing is clear, concise and easy to read. Most importantly, the quality of material in your SSI articles inspires trust in your work.
Your offer to share your London Business School presentation deck is very generous. I would appreciate a copy, if it is still available.
Happy New Year to you and your family.
Hi Peter,
I hope you’re doing well, and am glad you are posting again. I wanted to follow up on my previous email and express my gratitude for your unique and valuable insights.
Peter — it’s great to have you back! Knowing that we will continue to have the benefit of your deep analysis and clear, articulate writing is a great way to start the year. Thanks for everything so far and I look forward to future postings. It’s great to hear that we’ll get your insights into the impact of AI, as well.
All the best!
Hi Peter,
Glad you are back, Thank you for your sharing as always.
I am learning a lot from your post; it is so much value with sincerely.
Thank you very much,
Nadal