---
title: "AI Investment Enters an 'Era of Divergence': The Six Most Contentious Questions on Wall Street"
type: "News"
locale: "en"
url: "https://longbridge.com/en/news/294753081.md"
description: "The AI narrative is tearing itself apart: the same capital expenditure is seen as both a signal of growth and a financial risk; SaaS stocks have fallen more than 20% year-to-date, yet ServiceNow's quarterly report remains solid; open-source large models are scaling up, but this has not hindered OpenAI's revenue explosion. Barclays analysts bluntly state that investors' concerns are \"contradictory.\" The real question is no longer whether AI is useful, but who captures the incremental revenue and who bears the expansion risks—each dichotomy remains suspended in the air, awaiting judgment"
datetime: "2026-08-04T01:13:17.000Z"
locales:
  - [zh-CN](https://longbridge.com/zh-CN/news/294753081.md)
  - [en](https://longbridge.com/en/news/294753081.md)
  - [zh-HK](https://longbridge.com/zh-HK/news/294753081.md)
---

# AI Investment Enters an 'Era of Divergence': The Six Most Contentious Questions on Wall Street

The trouble with the AI theme has evolved from "whether it exists" to "who will win," with an opposing narrative standing beside every piece of good news. As the narrative continues to splinter, the same batch of positive data is being interpreted by the market simultaneously as reasons to buy and reasons to sell.

According to Zhuifeng Trading Desk, Raimo Lenschow, Barclays' US software analyst, wrote in a research note on August 3: **“In discussions with investors about the AI theme, we find that many of their concerns are contradictory.” This statement essentially summarizes the awkwardness of the current AI trade: the market does not lack consensus, but rather, the same consensuses cannot hold true simultaneously.**

Cloud providers' capital expenditures hitting record highs can prove that demand is real; it can also prove that balance sheets are under pressure. Large-cap SaaS stocks have generally fallen more than 20% year-to-date, significantly underperforming the S&P 500's 11% gain over the same period, but ServiceNow's latest quarterly report remains solid, with operating metrics not deteriorating in sync. Valuations for foundation models are elevated, while open-source large models continue to scale up. These signals, which should ideally corroborate each other, are instead canceling each other out.

For example, at the individual stock level, Oracle and Microsoft are facing two different types of pressure. Oracle is being penalized for its excessive capital expenditures and strained balance sheet; once the market worries about slowing growth, it is further penalized for "insufficient growth." Microsoft's situation is slightly better, but it is also caught in the middle: on one side, some complain that its AI momentum is not strong enough, while on the other, long-term capital is concerned about the return on invested capital (ROIC) of its massive capital expenditures.

The contradictions in the software sector are even more direct. Large-cap SaaS stocks have generally fallen by at least 20% year-to-date, while the S&P 500 rose 11% over the same period; however, the operating metrics of SaaS companies have not deteriorated in sync, and ServiceNow's performance a few weeks ago was quite solid. The market is now betting that "AI will disrupt software," but it has not yet seen sufficient financial evidence that large SaaS platforms are being truly displaced.

## Contradiction One—The More Capital Expenditure, the Better; The Less, the Better

AI capital expenditure was initially viewed as direct evidence of strong demand. Stocks like Oracle once surged on this logic: **the bolder a company is in expanding data centers and purchasing computing power, the more it indicates that the underlying orders and cloud demand are genuine and credible.**

However, after the debt market began to reprice, this logic reversed. The expansion of AI infrastructure requires financing arrangements such as debt and leases; the larger the financing scale, the more apparent the pressure on the balance sheet, making "spending less" a plus factor.

Subsequently, Google GCP and Amazon AWS delivered accelerated cloud growth results in the last quarter, shifting sentiment back toward strong demand. In recent weeks, discussions about the semiconductor cycle and "whether AI can be achieved at lower costs" have heated up, bringing concerns back to the forefront.

The result is that the same capital expenditure can be interpreted either as an investment in growth or as a financial risk. The market has not yet decided which one to reward.

## Contradiction Two—Oracle and Microsoft: Two Different Pincers

**Oracle is in the most difficult position. When capital expenditures are high, the market worries about the balance sheet; when growth expectations are downgraded, it worries that the AI story is not robust enough. These two penalties overlap, and the logic is not entirely self-consistent.**

But there is another side to Oracle's situation. It started late in cloud computing, and the expansion of AI infrastructure has given it a window to catch up, so it must be more aggressive than some of its peers. If future new data center construction encounters increasing resistance regarding site selection and power supply, companies that have already locked in sites and capacity may possess a first-mover advantage.

Microsoft's situation is relatively more comfortable. Azure, Copilot, and Office 365 collectively provide multiple paths to participate in AI growth, without needing to make the most aggressive bets in every AI track. This, however, has brought new two-sided skepticism: some complain that its momentum is not fast enough, while others worry that the return on invested capital for its massive capital expenditures remains unclear. The former thinks it is not aggressive enough, while the latter worries it is already too aggressive.

## Contradiction Three—SaaS Pricing Logic: Stock Prices Fall, But Metrics Remain Intact

The decline in large-cap SaaS stocks already implies fairly heavy pessimistic expectations. A drop of at least 20% year-to-date, significantly underperforming the S&P 500, indicates that the market is repricing the software business model itself—not just adjusting valuation multiples.

**The problem is that operating data has not aligned with this pricing. ServiceNow's latest quarterly performance was solid, and at least so far, there is no financial evidence of customer churn or revenue erosion among large SaaS platforms.**

What truly needs verification are two things: first, whether large SaaS platforms are indeed easily replaceable; second, whether these companies can transform AI into product upgrades and pricing upgrades, rather than having their seat-based revenues suppressed by AI. Some companies have already begun shifting towards more usage-based billing models, making the path of "AI replacing SaaS" far more tortuous than imagined.

## Contradiction Four—"AI Kills Software" vs. "AI Overinvestment": Two Narratives Cannot Both Be Maxed Out

This is the set of contradictions with the greatest logical tension in the AI trade.

**If AI is truly going to replace the existing software system, it essentially means rebuilding the entire current software stack, which itself requires a massive amount of AI computing power.** Especially since a large portion of current AI workloads are still concentrated on the training side and have not fully shifted to the inference side, if inference demand explodes on a large scale in the future, the demand for computing power will not naturally disappear.

The market must choose between two narratives: either the impact of AI on software is not as fast as imagined, and SaaS platforms can still retain a significant portion of their value; or AI is truly rewriting the software world, in which case it is difficult to simultaneously assert that future AI capacity will be significantly oversupplied. The phrases "AI kills software" and "AI infrastructure bubble" cannot both be pushed to the extreme.

## Contradiction Five—Foundation Models: High Valuations and Commoditization Cannot Both Be Realized

There is a similarly difficult-to-reconcile contradiction at the foundation model layer. On one hand, foundation model vendors enjoy high valuations, implying that they will capture a significant portion of the economic value in the AI industry chain. On the other hand, discussions about open-source models, multi-model configurations, small models, and specialized models are heating up, suggesting that model capabilities are moving towards commoditization.

Recent news has further amplified this divergence. OpenAI CFO Sarah Friar told employees that annualized recurring revenue in July exceeded the entire second quarter, with growth momentum coming from the GPT-5.6 series, the enterprise Agent product ChatGPT Work, and the AI coding tool Codex—providing support for the notion that "frontier closed-source models still have explosive revenue potential."

But at the same time, Moonshot AI released Kimi K3 with 2.88 trillion parameters, hailed as the largest open-source AI model to date, adopting a MoE architecture, supporting a 1 million token context window and native multimodal capabilities. The continuous scaling of open-source large models is the core fuel source for the narrative of foundation model commoditization.

Microsoft and other large tech companies are increasing their emphasis on multi-model configurations. If the model layer accelerates towards commoditization, the pricing power of single closed-source foundation models will be significantly weakened; if current valuations hold, the model layer should not rapidly degrade into ordinary infrastructure. Between the two, the market has not yet picked a side.

## Contradiction Six—Real Enterprise AI Demand: Not Just "The Stronger the Model, the Better"

Several recent enterprise AI developments jointly outline a picture more complex than a "model arms race."

Elastic expanded its partnership with OpenAI, combining OpenAI models with Elasticsearch, focusing on enterprise data retrieval, search, governance, permission control, and reducing unnecessary data transmission and token costs. Snowflake launched Cortex AI Gateway, aiming to manage how AI Agents access enterprise data and tools, while preventing enterprise costs from spiraling out of control, and introduced identity and security partners such as 1Password, SailPoint, and Saviynt.

The common keywords in these two cases are not computing power, but permissions, trust models, and cost caps. The core consideration in enterprise AI procurement is shifting from "model capability" to "governance, security, and controllable costs."

**This is also the deepest divergence in current AI investment: the bull narrative is not without logic, nor are the bear concerns baseless. The real question is no longer "whether AI is useful," but who ultimately captures the incremental revenue brought by AI, and who bears the cumulative cost risks of AI expansion.** The explosion of inference revenue versus falling token prices, the deflationary effect of AI versus rising electricity and data center costs, winner-takes-all versus lowering entry barriers—each dichotomy remains suspended in the air, awaiting data to pass judgment.

### Related Stocks

- [NOW.US](https://longbridge.com/en/quote/NOW.US.md)
- [NOWL.US](https://longbridge.com/en/quote/NOWL.US.md)
- [BCS.US](https://longbridge.com/en/quote/BCS.US.md)
- [.SPX.US](https://longbridge.com/en/quote/.SPX.US.md)
- [ORCL.US](https://longbridge.com/en/quote/ORCL.US.md)
- [MSFT.US](https://longbridge.com/en/quote/MSFT.US.md)
- [OpenAI.NA](https://longbridge.com/en/quote/OpenAI.NA.md)
- [ESTC.US](https://longbridge.com/en/quote/ESTC.US.md)
- [SNOW.US](https://longbridge.com/en/quote/SNOW.US.md)
- [SAIL.US](https://longbridge.com/en/quote/SAIL.US.md)
- [BARC.UK](https://longbridge.com/en/quote/BARC.UK.md)
- [ORCL-D.US](https://longbridge.com/en/quote/ORCL-D.US.md)

## Related News & Research

- [ServiceNow targets December 2026 rollout for AI specialists to automate vulnerability remediation, compliance monitoring](https://longbridge.com/en/news/294829055.md)
- [Why shares of ServiceNow stock were rising this week](https://longbridge.com/en/news/294541409.md)
- [How the 'Nostradamus of AI' failed to spot a market crash](https://longbridge.com/en/news/294580418.md)
- [The AI bros are lonely](https://longbridge.com/en/news/294942232.md)
- [3 genius artificial intelligence (AI) stocks to buy right now](https://longbridge.com/en/news/294816326.md)