5 Software Stocks Target AI Token Costs as Budgets Tighten

For nearly two years, semiconductor stocks have dominated investor attention, with shares of chip designers and equipment makers climbing as artificial intelligence (AI) demand surged. In contrast, enterprise software companies have often been priced as if large language models might soon make their products obsolete. That narrative is now being challenged, not by the software firms themselves, but by analysts who see a new role for these platforms in controlling AI costs.

Piper Sandler recently identified five infrastructure software companies-Elastic, GitLab, MongoDB, Snowflake, and Atlassian-as well positioned to address a growing pain point for chief information officers: the unexpectedly high cost of running AI agents at scale. According to a report from Piper Sandler, these companies can leverage the customer data they already manage to reduce the number of tokens AI models need to process, directly lowering operational expenses.

Token Costs and Enterprise AI Budgets

AI models typically charge by the token, with each token representing a small chunk of text. Both input and output tokens are counted, and as models become more sophisticated, the number of tokens consumed per query can rise sharply. While the price per token has dropped-output tokens on newer models are about 50% cheaper than previous generations-overall AI bills have climbed as organizations run more complex queries and expand usage.

Piper Sandler's analysis suggests that by feeding clean, well-organized proprietary data into AI agents, companies can cut token usage by 50% to 75% in early deployments. This efficiency not only reduces costs but also allows for broader AI adoption without blowing through budgets. The mechanism is straightforward: AI models require fewer tokens and deliver faster responses when provided with structured, relevant data instead of unfiltered information.

Shifting Revenue Models

The move toward consumption-based pricing is changing how software vendors generate revenue from AI. Instead of charging per user or seat, vendors are increasingly billing based on actual AI usage-specifically, the number of tokens processed. This model aligns vendor revenue with customer activity: if a company's AI agents handle more queries, the software provider earns more, even if the number of users remains flat.

For this thesis to hold, three conditions must be met: enterprises need to keep expanding AI agent deployments, context-layer technology must remain difficult for AI model vendors to replicate, and consumption-based revenue must outpace any decline in traditional seat licenses. Piper Sandler notes that early conversations with management teams and channel partners indicate a shift toward using software to make AI more efficient, but the transition is still in its early stages.

Stock Performance and Market Context

The five software stocks highlighted by Piper Sandler have not moved in lockstep. MongoDB has been the standout, with its market capitalization reaching approximately $27.7 billion in mid-July, up more than 60% from the previous year. Elastic, on the other hand, has seen its share price decline, while GitLab remains the weakest performer among the group, trading just below analysts' average 12-month target. Snowflake sits in the middle, with a strong buy consensus and an average target price above its current level. Atlassian trades well below its average analyst target, reflecting ongoing skepticism about seat-dependent business models.

These developments come against a backdrop of broader weakness in software stocks. The S&P 500 software industry index has dropped more than 25% from its October highs, and the iShares Expanded Tech-Software Sector ETF (IGV) is down 13% year-to-date, even as the S&P 500 overall has gained nearly 10% over the same period. This divergence has prompted some analysts, including those at Morgan Stanley, to argue that sentiment on software has become overly negative, especially as AI reshapes the sector's economics.

Risks and Uncertainties

There are significant risks to the consumption-based AI thesis. AI model providers could build their own retrieval and memory features, reducing the need for third-party context layers. Some labs have already started integrating these capabilities, which could erode the competitive advantage of the five software companies highlighted by Piper Sandler. Additionally, the projected 50% to 75% token savings are based on early use cases, not audited results across a broad customer base.

Another limitation is that none of these companies currently break out context-layer revenue as a separate line item in their financial filings. Investors are relying on analyst estimates rather than disclosed numbers, and consumption pricing can cut both ways-if AI budgets are reduced, revenue could fall more quickly than with annual seat contracts.

What to Watch Next

The upcoming earnings cycle will provide more clarity. Snowflake, MongoDB, and Elastic all report consumption metrics that investors can use to assess whether AI-driven usage is translating into revenue. For GitLab and Atlassian, which are more dependent on seat-based pricing, the key question is whether they can grow consumption or credit revenue even if seat counts remain flat. Net revenue retention-how much existing customers are spending as they expand AI deployments-will be a critical metric to watch.

For investors weighing these trends, it's important to recognize that the five companies carry very different risk profiles, despite sharing a common narrative around AI token cost savings. As the market continues to digest the implications of AI for software business models, the next few quarters could prove pivotal. For those interested in the broader landscape of AI model investing, recent moves by companies like Anthropic are also drawing attention, as seen in coverage of their potential IPO at Anthropic's plans to open direct AI model investment.

According to company filings and market data, MongoDB's market capitalization stood at $27.7 billion as of July 2026, reflecting a 62% increase from the prior year. Snowflake's average analyst target price was $302.26, while Atlassian's was $139.70, compared to its trading price near $86. The S&P 500 software industry index declined over 25% from October 2025 highs, and the IGV ETF fell 13% year-to-date through July, even as the broader S&P 500 rose nearly 10%.

Consumption-based pricing in enterprise software represents a fundamental shift from traditional seat-based models. Instead of tying revenue to the number of users, vendors now align their income with how much customers actually use AI-powered features. This approach can create more flexible, scalable revenue streams, but it also introduces new volatility-especially if customers cut back on AI spending. For investors and companies alike, understanding the mechanics of token usage, context layers, and consumption metrics will be essential as AI continues to reshape the software landscape.