On March 28, 2026, Salesforce delivered a stark message during its Q4 FY2026 earnings call, one that rippled far beyond its own shareholder base. The enterprise software giant set its fiscal year 2027 operating margin guidance at 18.5%, a material decline from the 20.2% achieved in the fiscal year just ended.
CEO Marc Benioff was unambiguous about the cause: the escalating computational and integration costs required to power its Einstein AI platform at scale.
This 170-basis-point compression is not merely a rounding error; it is a direct challenge to the foundational investment thesis of the software-as-a-service (SaaS) model, which has long been predicated on the principle of near-zero marginal cost for each new customer.
The margin pressure is not a simple line item but a complex interplay of new cost drivers endemic to artificial intelligence. The primary culprit is AI inference—the process of running trained models to generate predictions or content.
Unlike the one-time capital expenditure of traditional software development, inference is a continuous operational expense, a computational tax on every user query.
These costs flow directly to the providers of the underlying infrastructure: cloud hyperscalers like Amazon Web Services and Microsoft Azure, and the chipmakers, principally NVIDIA, whose H100 and B200 GPUs power the data centers.
Microsoft itself signaled this trend, reporting a 28% year-over-year increase in capital expenditures to $14B in its most recent quarter, largely to build out AI capacity.
For Salesforce and its peers, this means every AI-powered feature now has a variable cost of goods sold (COGS) attached, fundamentally altering the sector's economic architecture.
This margin shift presents two competing interpretations for the SaaS sector. The bear case argues that Salesforce's guidance is a canary in the coal mine. The 170bps compression is a preview of a sector-wide re-rating, where the market must finally price in the non-trivial marginal costs of AI.
Companies with less scale or pricing power than Salesforce, such as smaller players in the marketing automation or HR tech spaces, could face even more severe pressure.
This view suggests that Wall Street has enthusiastically modeled AI-driven revenue growth, as seen in the 35% average premium for stocks in the Goldman Sachs Non-Profitable Tech Basket since 2024, while largely ignoring the corresponding impact on COGS.
A portfolio heavily weighted toward SaaS names that have aggressively marketed AI features without a clear monetization strategy is exposed to significant valuation risk.
The bull case, conversely, frames this as a temporary investment cycle, not a permanent impairment of the business model. Proponents argue that the current high cost of inference will decline over time due to both hardware and software efficiencies.
Innovations in model quantization, the development of smaller specialized models, and the eventual commoditization of AI accelerator chips beyond NVIDIA could drive down per-query costs. More importantly, this view hinges on pricing power.
Microsoft’s early success with its $30 per user per month Copilot for Microsoft 365 is the key exhibit. This represents a 53% to 83% price uplift for its core enterprise customers.
If Salesforce can successfully tier its offerings and convince its 150,000+ customers to pay a premium for its Einstein 1 Platform, the incremental revenue will more than absorb the increased compute costs, leading to an expansion of both gross profit dollars and, eventually, margins.
This evolving landscape creates a clear bifurcation of winners and losers. The most obvious beneficiaries are the infrastructure providers selling the proverbial picks and shovels. NVIDIA’s Data Center revenue, which surged to $18.4B in its last reported quarter, is a direct reflection of this demand.
The cloud providers—AWS, Azure, and Google Cloud—are also primary winners, converting capital expenditures on servers into high-margin recurring revenue streams. Among software companies, those with unique, proprietary data sets will thrive.
An AI model is only as good as the data it’s trained on, and firms that can leverage decades of customer data, like Salesforce with its Data Cloud or Adobe with its creative asset libraries, can build defensible moats that justify premium pricing.
An investor holding a broad-based tech ETF like the Technology Select Sector SPDR Fund (XLK) has exposure to both layers of the stack, from Microsoft and NVIDIA to a basket of software firms facing these new pressures.
The losers will be SaaS companies with undifferentiated products in crowded markets. Those with limited pricing power and thin pre-existing margins—say, below 10% on an operating basis—will find it nearly impossible to absorb AI costs without either sacrificing profitability or pricing themselves out of the market.
Another vulnerable category includes firms that rely exclusively on third-party models, like those from OpenAI or Anthropic, without adding a significant proprietary data or workflow layer.
They risk becoming thin wrappers around a commoditized service, with their margins dictated by the API call fees charged by their suppliers.
An investor holding $100K in a portfolio of mid-cap SaaS companies, each with operating margins under 15%, would need to scrutinize their AI monetization strategies with extreme prejudice following the Salesforce announcement.
To navigate this dynamic, portfolio managers must monitor specific forward-looking signals in upcoming earnings reports.
Pay close attention to the Q1 and Q2 2026 releases from other SaaS bellwethers like ServiceNow, Workday, and Intuit for any commentary on AI-related costs or revisions to long-term profitability targets. The key metric to watch is gross margin, not just operating margin.
A decline in gross margin is a clear indicator that AI compute costs are being classified as COGS, confirming a structural change to the business model. A firm that instead classifies these costs under R&D may be signaling that it views the expense as a temporary investment rather than a permanent cost of revenue.
Furthermore, the pace and success of new product monetization will be critical. Watch for announcements of premium, AI-enabled subscription tiers and track the commentary on their adoption rates.
For example, if Adobe were to announce a 'Firefly Pro' tier for its Creative Cloud suite at a 25% price increase, its adoption by the first one million users would be a powerful data point on the market's willingness to pay. Finally, continue to monitor the capital expenditure guidance from the cloud hyperscalers.
Amazon, Microsoft, and Alphabet collectively guided for over $150B in capex for fiscal 2026. Any upward revision to these already historic figures signals their continued confidence in enterprise demand for AI, which translates directly into future costs for the SaaS companies they serve.
The Salesforce guidance marks a pivotal moment.
The era of 'software's near-zero marginal cost,' a concept that has underpinned tech valuations for two decades, is giving way to 'AI's non-zero computational cost.' This introduces a new variable into the financial physics of the sector, one that rewards scale, proprietary data, and pricing power while penalizing undifferentiated products.
The profit and loss statement is the final arbiter, and it is beginning to render its judgment on the true price of intelligence.
This article is for informational purposes only and does not constitute financial advice. Past performance is not indicative of future results. Consult a qualified financial adviser before making investment decisions.