AI disruption concerns have been overdiscounted, and low valuations in the software sector are creating re-rating opportunities

Overview

The report assesses disruption risk, growth prospects, cost structure, valuation divergence, and stock ranking in the U.S. software and AI infrastructure industry. The core judgment is that investors are overgeneralizing the impact of AI on application software and infrastructure software, while overlooking the barriers formed by enterprise systems of record, existing workflows, data context, maintenance responsibilities, and deterministic outcome requirements. In the near term, the industry still faces pressure from seats, competition from adjacent products, rising inference costs, and capital expenditure burdens, but AI feature monetization, usage-based pricing, declining model costs, and personnel efficiency improvements are expected to bring medium-term revenue and earnings reacceleration.

Core views

First, the actual level of disruption in the software industry may be significantly lower than what is priced by the market; risk is lowest at the infrastructure layer, and highest at the application layer where interface value accounts for a large share and integration with systems of record is limited. Second, AI products have moved from deployment and adoption improvement toward monetization, and revenue models will gradually expand from pure seat-based models to a combination of seats and usage. Third, current consensus expectations do not incorporate meaningful growth acceleration, so even if the industry merely maintains healthy but relatively low growth, valuations of around 4x enterprise value/revenue are supported; if growth reaccelerates from the second half of 2026 to the first half of 2027, the timing of sector re-rating may come earlier. Fourth, open-weight models, small language models, and model portfolio optimization are expected to reduce inference costs and repair gross margins. Fifth, AI infrastructure has excellent demand prospects, but its high capital intensity and ongoing financing needs give it a risk-return profile different from that of asset-light software companies.

Analysis framework

The report combines the location of disruption risk within the software technology stack, the Rule of 40 and Rule of X growth-profitability frameworks, enterprise value/forward revenue valuation, historical share price and valuation divergence, consensus revenue revisions, global compute supply and demand, sales and R&D expense ratios, revenue per employee, and pricing model migration to compare and rank covered companies across layers.

Methodology notes

• Growth and Profit QualityRule of 40

  The sum of forward revenue growth and forward free cash flow margin

  Used to measure the balance between growth and cash flow for asset-light software companies; companies building physical infrastructure with pressured free cash flow require different benchmarks.

• Growth and Profit QualityRule of X

  Two times forward revenue growth plus forward free cash flow margin

  By increasing the weight of revenue growth, it emphasizes the importance of growth durability to software valuations; this framework further highlights DDOG, SNOW, and NOW.

• Relative ValuationEnterprise Value/Forward Revenue Multiple

  Compare current valuations, historical ranges, and differences between application software and infrastructure software

  The current sector trades at around 4x enterprise value/revenue, far below the pandemic-era peak of around 20x; application software valuations have fallen to about one-third of infrastructure software valuations.

• Industry StructureRisk Layering in the Software and Infrastructure Technology Stack

  Disruption risk usually increases from the underlying infrastructure layer toward the upper application interface layer

  The closer a product is to compute, the data layer, systems of record, and critical workflows, the higher the replacement barriers generally are; low-integration products whose main value is concentrated in the user interface face relatively higher risk.

• Operating EfficiencyRevenue per Employee and Expense Ratio Analysis

  Use revenue per employee, sales expense ratio, and R&D expense ratio to assess room for cost leverage

  Median revenue per employee among covered companies is about $484,000, with large differences across companies, indicating that many companies can still use AI to improve R&D, sales, and operating efficiency.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Microsoft (MSFT)

  Report top pick, OW rating

  Strengths

  Balanced portfolio of application software and AI infrastructure; able to use cash generated by high-margin application businesses to support Azure capital expenditure and create additional growth opportunities through Copilot.

  Weaknesses

  Physical infrastructure buildout depresses near-term cash generation and requires continued capital investment.

  Comparison

  Compared with other hyperscale cloud providers, its internal high-margin application business can reduce external financing pressure for AI infrastructure; compared with Oracle, it has richer software monetization opportunities.

  Risks

  Compute investment returns fall short of expectations, AI demand conversion is slower than expected, and gross margins come under pressure.

• ServiceNow (NOW)

  Key preferred name

  Strengths

  As an AI entry point in enterprise IT environments, it is deeply embedded in deterministic workflows; AI annualized contract value has exceeded $1 billion, and its Rule of 40 performance is outstanding.

  Weaknesses

  Organic business growth is currently mainly stable in the high teens, and usage revenue is still needed to drive reacceleration.

  Comparison

  Compared with CRM and DT, it has better combined growth and profitability performance, while its free cash flow valuation is discounted by more than 50% versus its long-term average.

  Risks

  Delays in enterprise AI project deployment, usage monetization falling short of expectations, and slowing seat growth.

• Datadog (DDOG)

  AI observability beneficiary

  Strengths

  Closer to cloud-native and DevOps use cases, and expected to achieve higher growth from increasing AI workloads and rising system complexity.

  Weaknesses

  R&D investment is high, and its valuation as an AI beneficiary is not cheap.

  Comparison

  The report prefers DDOG over DT because its cloud-native and DevOps positioning provides higher growth potential.

  Risks

  High valuation, fluctuations in cloud usage, intensifying competition, and growth falling short of expectations.

• Snowflake (SNOW)

  Data-layer AI beneficiary

  Strengths

  Well positioned in the enterprise data layer, with AI innovation already showing momentum; over 90% of revenue is contributed by usage-based products.

  Weaknesses

  The usage model makes revenue more sensitive to customer compute demand and budget adjustments.

  Comparison

  The Rule of X framework highlights its high growth quality; compared with application software, its infrastructure attributes and data barriers reduce disruption risk.

  Risks

  Customers optimizing cloud spending, slower adoption of AI products, usage volatility, and high valuation.

• Oracle (ORCL)

  Low-valuation AI infrastructure opportunity

  Strengths

  Its plans to expand compute capacity give it the ability to compete with hyperscale cloud providers, and its valuation is relatively cheap.

  Weaknesses

  Capital needs are high, free cash flow metrics are pressured during the buildout period, and software monetization opportunities around physical infrastructure are lower than Microsoft’s.

  Comparison

  Valuation is cheaper than Microsoft’s, but the business mix provides relatively limited self-funding capacity and incremental software monetization opportunities.

  Risks

  Capital expenditure execution, financing costs, rising debt, and insufficient realization of compute demand.

• Salesforce (CRM)

  Valuation recovery candidate

  Strengths

  As a system-of-record platform that provides multiple solutions to a single customer, its actual disruption risk is lower than what the market prices; progress toward the Rule of 50 can support re-rating.

  Weaknesses

  Some businesses such as Tableau still face AI substitution pressure, and overall growth is relatively limited.

  Comparison

  Valuation and growth metrics are lower than NOW’s, but low expectations and system-of-record attributes provide a larger expectation gap.

  Risks

  Seat compression, damage to adjacent products, insufficient AI monetization, and continued growth slowdown.

• Cloudflare (NET)

  AI infrastructure beneficiary but maintained at Neutral

  Strengths

  Positioned at the infrastructure layer, it can benefit from growth in AI traffic and compute demand, with a Rule of 40 of 41.4%.

  Weaknesses

  Its 32.6x enterprise value/revenue valuation is significantly higher than most covered companies.

  Comparison

  Also a beneficiary of AI infrastructure software, but its valuation is more expensive than preferred names such as DDOG and SNOW.

  Risks

  Valuation compression, growth not meeting high expectations, and intensifying competition.

• Physical AI infrastructure companies

  Strong demand but high volatility

  Strengths

  The compute supply-demand gap is expected to widen, and customers are accepting higher compute prices and premiums, supporting medium-term demand and revenue growth.

  Weaknesses

  Capital expenditure must occur before revenue, continued reliance on capital markets is required, and more enterprise value may accrue to creditors.

  Comparison

  Demand visibility is higher than for most application software, but capital intensity, financing risk, and share price volatility are also significantly higher.

  Risks

  Tighter capital market access, rising interest rates, debt burden, construction delays, and insufficient long-term investment returns.

Key data

• Current valuation of the software sectorAround 4x enterprise value/revenueSignificantly below the pandemic-era peak of around 20x, but above the trough of around 1.5x during the global financial crisis.
• Relative valuation of application softwareAround one-third of the valuation multiple of infrastructure softwareYears of underperformance in share prices have caused a significant valuation divergence between the two types of software companies.
• Average share price performance of infrastructure companies in 2025+18%Average share price performance of application software companies over the same period was -11%.
• Cloudflare valuation and Rule of 4032.6x enterprise value/revenue; 41.4%Not included in the corresponding matrix display because its high valuation would affect the chart scale.
• ServiceNow AI annualized contract valueOver $1 billionAI outcomes have already begun to be monetized, and non-seat models account for about half of new business.
• Median revenue growth of covered companies in the second half of 20266.3%The consensus expectation threshold is low and has not yet incorporated broad industry growth reacceleration.
• Global computing capacity supply-demand gap in 2030241,324 megawattsThe report table shows demand of about 449,813 megawatts and supply of about 208,489 megawatts in 2030.
• Revenue per employee of covered companiesAverage about $679,000, median about $484,000Large differences across companies indicate room to further improve productivity and unlock cost leverage.
• Snowflake usage-based product revenue shareOver 90% of total revenueIts business model mainly charges directly based on compute and storage usage.
• Twilio usage-based revenue share74% in 2025Higher than 72% in 2024 and 71% in 2023, but it also increases sensitivity to macro cycles.

Impact & implications

For investors, the current opportunity is not simply to bet on a synchronized rebound across all software companies, but to prioritize companies that combine system-of-record attributes, data or workflow barriers, a clear AI monetization path, strong growth quality, and reasonable valuations. Infrastructure software benefits more directly from growth in AI workloads, but valuations for some names are already high; application software faces greater risks, but its significant discount creates a larger expectation gap. Physical AI infrastructure has the strongest demand, but financing capacity, cost of capital, debt as a percentage of enterprise value, and investment returns should be the main risk constraints.

Risks

• AI-native products may erode low-integration application software and adjacent businesses whose main value lies in the user interface more quickly.
• Enterprise customers may reduce seat counts and conduct stricter return-on-investment assessments between purchased software and internal development.
• Rising inference and token costs may continue to pressure software companies’ gross margins before AI revenue scales.
• If industry revenue fails to reaccelerate from the second half of 2026 to the first half of 2027, valuation recovery may be delayed.
• High interest rates may continue to suppress software valuations and increase financing costs for AI infrastructure companies.
• AI infrastructure requires continued capital investment, and debt growth may dilute the enterprise value upside captured by shareholders.
• While usage-based pricing increases growth elasticity, it also increases revenue sensitivity to customer budgets, the macro environment, and usage intensity.
• Even with low valuations, some application software companies may still become value traps due to structural growth slowdowns.

What to watch

• Whether software company revenue in the second half of 2026 and the first half of 2027 reaccelerates beyond low consensus expectations.
• After AI features shift from trials and adoption to paid usage, their contribution to new contract value, usage revenue, and net retention rates.
• Whether revenue pressure from seat reductions can be offset by usage growth and AI add-on charges.
• Whether open-weight models, small language models, and model routing can reduce unit inference and token costs.
• Whether industry gross margins stop declining and recover as AI revenue scales.
• The speed at which enterprise customer AI projects move from pilots to production environments and evidence of investment returns.
• The global compute supply-demand gap, compute prices, and the sustainability of customers paying premiums.
• Changes in capital expenditure, financing channels, debt ratio, and free cash flow for physical infrastructure companies.
• Whether the divergence in valuations and share price performance between application software and infrastructure software begins to converge.
• Whether MSFT, NOW, DDOG, SNOW, ORCL, and CRM can deliver on their respective AI monetization paths.