Software strikes back: From AI panic to recovery
Enterprise software has been the subject of sustained investor debate over the past two years. Rapid advances in artificial intelligence (AI) have raised questions around pricing power, the durability of seat-based licensing as AI agents take on work previously performed by users, and the extent to which customers may choose to build functionality internally rather than license it. These are legitimate questions, and the answers are unlikely to be uniform across the sector.
By the middle of this year, market pricing reflected a notably negative interpretation of those questions. Software, which has traded at a premium to the broader market for much of the past two decades derated significantly, and several established software names were trading on single digit or low double-digit price-to-earnings multiples.
Since the end of June, part of that has reversed, supported by company results and by several strategic developments across the sector. In this piece we look at what drove the derating, what has changed, and how we have positioned our portfolios.
What happened to software
Software and semiconductors have different business models. Software is asset light, earns recurring revenue and high stable margins. Semiconductors are cyclical and capital intensive. Despite those differences, the two have moved together for most of the past 25 years, since both have the same underlying technology demand cycle. Over the long run software has performed slightly better, which is consistent with value in a technology chain accruing closest to the end customer.
Source: Fairtree, Bloomberg as at 10 September 2026.
That relationship changed materially from 2023, shortly after the launch of ChatGPT. Capital rotated towards semiconductors seen as the direct beneficiaries of AI, and we saw hyperscalers and software used as the funders of that trade. From the start of 2024 to the end of June 2026, the Semiconductor Index outperformed the Software and Services Index by more than 200%, with software down 20% over the period.
Behind that move sits a genuine question, namely whether established providers can continue to monetise longstanding customer relationships as functionality becomes cheaper and more widely available.
Our concern was less with the questions themselves than with how uniformly they were applied. Companies of materially different quality, competitive positioning and customer relationships were repriced in the same direction and at a similar pace. In our experience, that pattern tends to indicate that factors other than fundamental company analysis are influencing prices.
A disconnect between price and fundamentals
Salesforce is a company we have followed closely, and over the past two years its forward earnings estimates (orange line) have risen, while the share price (black line) did the exact opposite, falling by more than half from its 2024 high. A multiple (green area) that had been above 70x forward earnings in 2021 reached close to 10x at this year’s low. The derating therefore reflected a change in what investors were prepared to pay rather than any deterioration in reported results.
Graph 2: Salesforce share price versus forward EPS estimates, with forward P/E below
Source: Fairtree, Bloomberg as at 11 September 2026.
Salesforce was not an isolated case, and by the middle of this year several established software names had derated substantially even as their earnings continued to grow. That raises the question of what these businesses should be worth now. We would not expect them to trade at the multiples they commanded over the past decade, since AI has changed the economics and moats of the industry, but the levels reached were at extreme lows relative to business fundamentals.
Where the value sits
These depressed valuations seem to underestimate where value in enterprise software is created. These systems are deeply embedded in customer workflows and must be integrated, secure, reliable and continuously maintained. Over time they accumulate customer specific configurations, data and regulatory requirements that are costly to replace, particularly in mission critical or regulated environments. A small business may be able to build a basic internal tool. A large enterprise operating with decades of embedded data, supplier integrations and institutional knowledge faces a very different problem, where replacing a core system is costly, risky and can take years to execute. The question is not whether AI can write code, but whether it can replace the accumulated expertise embedded within enterprise software and infrastructure.
The providers are also building AI into their own products rather than waiting to be disrupted by it. Used effectively, AI may strengthen many of these businesses by accelerating product development, enhancing functionality and improving customer outcomes, supporting retention rather than threatening it. Salesforce made the point directly, with Marc Benioff describing AIforce, its enterprise harness, as “unlocking the data, workflows, business logic, actions, and governance inside Salesforce and making it available to every agent, model, and interface”.
We would however not extend this to the entire software sector. AI will likely produce winners and losers within the space, sharpening the difference between businesses that can defend their position and those that cannot. It does seem unlikely, though, that all established providers are structurally impaired in the way mid-year pricing implied.
Salesforce: from perceived disruption risk to distribution partner
Salesforce sat close to the centre of the disruption concern. The argument was that AI would commoditise seat-based software, and that companies owning the customer relationship would be displaced by those owning the models. Developments over recent months have pointed in a different direction.
In late August the company announced Claudeforce with Anthropic. Claude now serves as a reasoning model within Agentforce’s Atlas Reasoning Engine and as the default model across Slack AI, Headless 360 and Agentforce Coworker, running inside the Salesforce trust boundary, which matters for regulated industries. Salesforce data and workflows also become accessible from within Claude itself, a capability currently in pilot. This follows the partnership with OpenAI announced last October that made Agentforce 360 available within ChatGPT. Dario Amodei described the appeal from Anthropic’s side, noting that companies can point Claude at the customer information and business context “they’ve been building in Salesforce for decades” and use it to run their businesses.
Amodei’s framing points to where the commercial value sits. Models are becoming cheaper and more widely available, while the governed, permissioned customer data they need to be useful in an enterprise setting is considerably harder to replicate, as is the workflow they operate within. Salesforce owns that layer, and these agreements suggest it is being treated as a distribution partner rather than an incumbent to be displaced.
Recent results are consistent with that. Second quarter revenue grew 11%, current remaining performance obligations grew faster at 14%, and combined Agentforce and Data 360 annual recurring revenue reached approximately US$3.9bn, having more than tripled over the year. Full year guidance was raised, and Robin Washington described new annual order value growth as “the strongest it’s been in four years”, pointing to revenue reacceleration over the coming quarters.
Evidence across the software sector
Salesforce is not the only company reporting improving trends, which matters, since a single set of results does not make a sector view. Adobe delivered record third quarter revenue of US$6.76bn, up 13% year on year, with subscription revenue up 14% and total annual recurring revenue of US$27.5bn. Its AI-first annual recurring revenue passed US$650m and grew more than 150% year on year.
SAP shows a similar picture with current cloud backlog reaching €22.9bn in the second quarter, up 26% in constant currency, and cloud ERP suite revenue up 27%. Christian Klein attributed this to AI related demand, noting that customers are choosing SAP “to enable accurate and compliant AI outcomes grounded in their most critical business processes and data”.
ServiceNow has raised its subscription revenue guidance twice this year and passed US$1bn in annual contract value for its AI products. Workday reports that AI is contributing more than a quarter of new annual contract value. Snowflake raised its full year product revenue forecast following a 37% increase in the second quarter, attributing approximately half of the acceleration to AI.
Software’s recent recovery
Multiples across enterprise software have started to recover since the end of June. Among the names highlighted below the average fell 46% into the low and has since regained 17%, leaving every one of them below the level it averaged over the three years to the end of 2025.
Graph 3: Software 12-month forward P/E multiples: three-year average, June 2026 trough and 15 September 2026
Source: Fairtree, Bloomberg as at 15 September 2026.
Salesforce shows the pattern most clearly, having averaged 25.2x forward earnings over the three years to the end of 2025, fallen to 10.6x by the end of June and recovered to 15.6x by 15 September. Adobe follows the same shape from a lower base, moving from a 23.9x average to 7.9x at the trough and 9.5x on 15 September. In both cases the recovery has been substantial in percentage terms while leaving the multiple far below where the businesses have historically been valued. The same pattern is visible in prices, with the S&P Software and Services Index rising 13% in August and standing roughly 21% above its end-June low. How AI ultimately reshapes software economics remains an open question, but even after the recovery off the June lows we still see value in the sector.
How we are positioned
Our investment philosophy is guided by a set of core principles that shape how we identify opportunities, manage risk and deliver long term value for our clients. Among them is a disciplined approach to valuation, investing only when the price reflects a favourable risk-reward profile. Enterprise software did not meet that test through 2021 and 2022 when the sector traded at elevated multiples, so we did not own these names.
The derating on the back of the “AI scare trade” changed that. Multiples on businesses with recurring revenue and deep integration into customer operations fell to levels that implied structural decline. This is when we became interested because we like the business models and they were trading at a substantial discount to the broader market and their own histories. We began selectively adding enterprise software in the fourth quarter of 2025 and continued through the first half of this year as valuations fell further and operating results began to improve.
We are also highly active, and the same discipline applies as we see these multiples recover. A rerating of this speed changes the risk and reward attached to a position, and where share prices move ahead of what the results support, we would take some profits and redeploy into opportunities we find more compelling.
Where this leaves us
Prices have recovered some ground since June, and the case for owning these businesses has begun to play out. The questions raised about pricing power, licensing models and competitive dynamics remain open, and we expect meaningful differentiation between companies over the coming years. Some software businesses will be structurally challenged by AI while others emerge stronger for it, and separating the two is where the work lies.
Periods of dislocation between price and fundamentals are where opportunities for valuation disciplined investors tend to arise, though they are uncomfortable while they persist and the outcome is never assured. We cannot forecast precisely how AI reshapes this industry, but understanding these business models deeply and staying disciplined on the price we pay is what has guided our decisions throughout.
We remain anchored to the principles that guide our investment philosophy: a disciplined approach to valuation, a style agnostic and well diversified portfolio, and the willingness to act as the opportunity set or facts change. They are what kept us out of the sector at elevated multiples and what allowed us to build exposure once valuations had fallen to levels we found attractive.
Author
Karena joined Fairtree in 2024 as a Global Investment Specialist in the Investment team. She began her investment career in 2019 as an Investment Professional at Investec Asset Management, followed by a role as an equity analyst at Denker Capital. In 2021, she joined PSG Asset Management as an equity analyst, gaining further experience in equity markets and financial modelling. Karena holds a Bachelor of Science degree in Mathematical Sciences from the University of Johannesburg.
FAIRTREE INSIGHTS
You may also be interested in
Explore more commentaries from our thought leaders, offering in-depth analysis, market trends and expert analysis.
Navigating a momentum-driven market
Global equity markets are currently being driven by a powerful momentum cycle, reshaping the opportunity set across our funds. The AI capex boom has seen beneficiaries such as industrials and semiconductors post exceptional returns.
The countries beneath the index
Emerging markets offer some of the most dynamic and attractive opportunities in global equities. After more than fifteen years of underperformance relative to developed markets, driven by a strong US dollar, weaker earnings growth and corporate governance concerns, the opportunity set looks fundamentally different today.
Fairtree Global Emerging Markets Fund Q2 2026 Commentary
The quarter was defined by an exceptionally powerful but highly concentrated recovery in emerging market equities, led by a small number of semiconductor and memory names, following the March US-Iran shock and the subsequent ceasefire that unwound the geopolitical risk premium.
Disclaimer
Fairtree Asset Management (Pty) Ltd, Registration Number: 2004/033269/07, is an authorised Financial Services Provider (FSP25917) under the Financial Advisory and Intermediary Services Act (No.37 of 2002), to act in the capacity as investment manager. This information is not advice, as defined in the Financial Advisory and Intermediary Services Act (N0.37 of 2002).
Information and content
The information in this publication is provided by Fairtree for general information purposes only. It is not financial, tax, legal, investment or other professional advice, and should not be relied upon as the sole basis for any investment decision. Investors should seek advice from a qualified financial adviser regarding their individual circumstances. While reasonable care has been taken to ensure the accuracy of the information, Fairtree makes no representations or warranties and accepts no liability for any loss, damage, cost or expense arising from its use. Nothing herein constitutes an offer, solicitation, or recommendation to invest.
Management company
Prescient Management Company (RF) (Pty) Ltd is registered and approved under the Collective Investment Schemes Control Act (No.45 of 2002).
Performance
Collective Investment Schemes in Securities (CIS) should be considered medium- to long-term investments. The value of a CIS may go up as well as down, and past performance is not necessarily a guide to future performance. CISs are traded at the ruling price and may engage in scrip lending and borrowing. A schedule of fees, charges, and maximum commissions is available on request from the Manager. A CIS may be closed to new investors to allow it to be managed more efficiently in accordance with its mandate.
Performance has been calculated using net NAV-to-NAV numbers, with income reinvested. The performance shown for each period reflects the return for investors fully invested for that period. Individual investor returns may differ due to initial fees, the actual investment date, the timing of reinvestments, and dividend withholding tax. Full performance calculations are available from the Manager on request. There is no guarantee of capital or returns in a portfolio. Annualised performance represents the average return per year over the period, rescaled to a one-year period. Actual annual figures and the highest and lowest one-year returns since inception are available on request. NAV (Net Asset Value) represents the assets of the Fund less its liabilities.
Fees
Permissible deductions may include management fees, brokerage, securities transfer tax, auditor’s fees, bank charges, and trustee fees. A schedule of fees, charges, and maximum commissions is available on request from Fairtree. For more information on annual management fees, please refer to the fund’s factsheet or the Fairtree website.
For any additional information, such as fund prices, fees, brochures, minimum disclosure documents and application forms, please go to www.fairtree.com.
About you…
By proceeding, I confirm that:
- To the best of my knowledge, and after making all necessary enquiries, I am permitted under the laws of my country of residence to access this site and the information it contains.
- I have read, understood, and agree to be bound by the Terms and Conditions and Privacy Policy.
Important notice: Please beware of fraudulent WhatsApp groups and social media accounts pretending to be affiliated with Fairtree or Fairtree employees.
If you do not meet these requirements, or are unsure whether you do, please click “Decline” and do not continue.