Data Infrastructure Software – the control layer of the AI stack

Data infrastructure has decoupled from the rest of enterprise software in the first half of the year. While seat-based SaaS de-rated violently in Q1 2026 — the “SaaSpocalypse” that erased roughly US$1trn of aggregate enterprise software market capitalisation amid fears that AI agents could cannibalise per-seat licensing — consumption-priced data platforms have been more resilient than their peers. Companies such as Datadog (+81.60% YTD), Snowflake (+51.87% YTD) have materially outperformed the IGV ETF while MongoDB (+10.2% YTD) has delivered a more modest gain.

Agentic AI changes the unit of demand. A human analyst may run a handful of queries a day; an agent orchestrating a workflow can run hundreds continuously, while requiring governed, current and machine-readable data to do so. Under a per-seat model, this represents a cost.

Under a consumption model, it represents revenue. Data infrastructure vendors therefore sit on the right side of this pricing asymmetry — provided the workloads run on their engines rather than on a hyperscaler’s platform or open-source infrastructure.

Figure 1: YTD performance

The Single and Most Direct Catalyst – Data Infrastructure in an AI World

Unlike traditional software companies that charge based on the number of seats, data infrastructure companies get paid based on how much work is performed, rather than how many people are working. Therefore, More computational activity can support consumption revenue, although the benefit depends on workload retention, pricing and customers’ ability to optimise usage.

From the Internet era to the mobile era and now the AI era, there has always been a need for a data layer. AI may change existing interfaces and reduce some labour-intensive processes; it does not eliminate the need for reliable underlying data. An agent that answers, “Which SKUs underperformed in Johor?” still has to retrieve the numbers from somewhere, and that somewhere needs to be current, consistent and governed. These are precisely the capabilities that data infrastructure companies provide.

The AI model is the worker, not the warehouse — and replacing the worker with a robot does not empty the warehouse.

Simply put, rising AI deployment within enterprises should benefit data infrastructure companies across the various layers of the data stack. Analytics pulls the actual numbers — positions, transaction history and portfolio aggregates. Observability records how long an entire exchange took, how much it cost in tokens, and alerts someone when something breaks. Databases store and serve the underlying information. These are all offerings provided by data infrastructure companies.

Agentic AI creates another structural push for the sector. A chatbot is largely stateless: ask, answer, forget. An agent is the opposite. It needs to remember what it has done, read and write business records, maintain session state across interactions, retrieve context, and remain observable when it fails. Every one of these functions generates database or telemetry operations.

Hyperscaler’s RPO provide an indirect indicator. As of the second quarter, all major hyperscalers reported an acceleration in cloud revenue, with Google Cloud growing 82% y/y, Amazon AWS 37% y/y and Microsoft Azure 43% y/y. Order backlogs, as measured by RPOs, provide strong visibility into future demand.

While a significant portion of these backlogs is associated with AI labs, which will not directly benefit data infrastructure companies, the expansion of cloud capacity provides the foundation for future increases in enterprise AI deployment. As enterprises adopt AI more broadly, this should increase demand for storage, networking, managed services and the infrastructure required to run ordinary corporate applications — This could expand the addressable workload base for data infrastructure vendors data infrastructure companies.

Companies are seeing strong demand for their offerings. More than 13,600 accounts now use Snowflake’s AI capabilities, while the number of accounts using Snowflake Intelligence has more than doubled quarter-over-quarter. Cortex Code is now deployed across more than 7,100 accounts.

Datadog reported 750 AI customers, including start-ups and hyperscalers using Datadog for their in-house AI labs. Additional data points include 31 customers generating more than US$1m in ARR and eight customers spending more than US$10m in ARR. During the quarter, Datadog landed two AI labs in seven-figure deals and renewed its largest AI customer in a nine-figure deal, with the customer using 17 Datadog products.

Figure 2: Still strong top-line growth

This may be as good as it gets

The room for imagination remains huge for this sector, but current valuations suggest that this may be as good as it gets. Among the companies, Snowflake and MongoDB are still loss-making. Even using future estimated earnings, forward P/E multiples suggest that these companies are trading at 48x–85x 2028 earnings, compared with around 21x forward P/E for the broader software sector.

While we remain positive on the future earnings trajectory of these companies, such stretched valuations suggest that the margin for error is low and that companies need to deliver near-perfect execution. Currently, investors are willing to pay a premium as these companies are viewed more as “beneficiaries” of the AI transformation, compared with the broader software sector, which is increasingly viewed as a “victim”.

Looking ahead, this strong momentum could continue as AI adoption progresses. However, we see several factors that could derail the momentum, which we discuss below. One of these is interest rates. The market is currently pricing in the possibility of one rate hike this year. Higher rates would negatively affect high-valuation companies, as higher discount rates are applied to their future cash flows.

Figure 3: Valuations are not cheap 

No clear earnings evidence yet on AI contribution

No vendor currently discloses AI-attributable revenue separately. Larger players such as Snowflake, MongoDB and Datadog have not broken out AI-driven consumption from their underlying growth. Snowflake reported that more than 13,600 accounts are using its AI capabilities, but provided limited details on the incremental revenue generated from these customers. This could suggest that the contribution remains relatively insignificant at this stage.

MongoDB is the more cautious dissenter. Management’s view is that enterprises remain in the early stages of AI adoption, with a broader impact expected over the next 12–18 months. The company continues to guide towards strength in its core workloads, suggesting that the impact of AI adoption is not yet material at the operational database layer.

The only notable exception is Databricks, which disclosed that its AI products have reached a US$1.7bn revenue run-rate. The AI business grew from more than US$1bn in December 2025 to US$1.4bn in February and US$1.7bn in June — roughly 70% growth in seven months. However, Databricks’ AI offering comprises Mosaic AI training, provisioned-throughput model serving and pay-per-token Foundation Model APIs. Put simply, Databricks is partly purchasing model capacity and reselling it to customers. This means that its AI exposure is, to some extent, exposure to the AI capex cycle rather than purely to durable demand for governed data infrastructure.

In short, the aggregate evidence is encouraging, but the attribution evidence remains weak. Databricks provides the clearest evidence of AI-driven revenue, but its AI business is not directly equivalent to demand for governed-data infrastructure. The lack of clear evidence that AI is driving incremental growth means that execution uncertainty remains high for the sector.

Table 1: Disclosed AI Revenue

	AI revenue disclosed	Growth trajectory	Margin signal
Databricks	$1.7bn run-rate, ~25% of ARR	55% → 65% → 80% (accelerating)	80%+ → mid-70s
Snowflake	None	~34%, accelerated	Stable
Datadog	None	36%, guided to 28–29%	80.9% → 79.6%
MongoDB	None	~25%, steady	Improving
Elastic	None	~16%	Improving

Source: Company announcements. Data as of 31 July 2026.

High AI Usage Does Not Guarantee Higher Vendor Revenue

This is arguably the most important risk for the sector. Datadog’s largest customer, despite recently renewing its contract, recorded lower usage that was incorporated into management’s guidance. Even if the customer’s underlying AI activity remained strong, lower Datadog usage shows that customers can optimise observability spending as infrastructure bills increase.. Once AI infrastructure expenditure becomes sufficiently material, customers begin optimising it. As a result, AI usage and vendor revenue can move in opposite directions precisely where AI adoption is highest.

In other words, the customer may remain, but revenue growth can still slow. The renewal itself is not the issue — Datadog highlighted that the customer had recently renewed even as usage declined. As such, the deterioration never appears in retention metrics or customer counts; it only becomes visible in the revenue growth rate, often with little advance warning. This helps explain why the market reacted so negatively to the latest results, given its weaker guidance as it raised questions over whether customer optimisation could slow growth despite otherwise healthy operating indicators (Datadog’s Q3 guidance implied revenue growth of 28–29%, down from 36% in Q2).

The implication is clear: as AI bills become larger, customers are likely to become more cost-conscious and increasingly optimise their spending. Higher AI adoption does not necessarily translate into proportionally higher revenue growth for infrastructure vendors.

A Databricks IPO Could Put Lofty Valuations to the Test

Although Databricks has yet to file for an IPO, with market speculation pointing to a potential listing in 2027, its eventual debut could have meaningful implications for the entire sector.

The first impact is on valuation multiples. Snowflake, trading at around 18x forward revenue, is partly valued as the investable proxy for lakehouse architecture and AI-driven data demand because its faster-growing private competitor is unavailable to public investors. A Databricks listing would remove that scarcity premium, increasing the risk of multiple compression across the group.

The second impact is on transparency. Databricks is currently the only major player in the sector that publicly reports a dedicated AI revenue line. An eventual S-1 filing would require audited financial statements, greater segment disclosure, customer concentration data, and — most importantly — a breakdown of gross margins between its high-margin data platform and its lower-margin pass-through inference business. This would provide investors with the clearest answer yet to a critical question: how much AI revenue is genuinely being generated by databases and data infrastructure, rather than by reselling AI compute?

A lot to prove, low margin of error

Data infrastructure enters the second half with a genuinely robust outlook: consumption pricing means these vendors are paid for machine work rather than headcount, agentic workloads multiply that work, and accelerating hyperscaler cloud revenues and order backlogs point to a widening enterprise deployment base — which is why the market has re-rated the group as an AI beneficiary while de-rating seat-based SaaS as an AI victim.

The problem is that the re-rating has run ahead of the evidence. At 48x-85x 2028 earnings against 21x for the broader software sector, the group is priced for execution perfection, yet no vendor other than Databricks discloses an AI revenue line, MongoDB itself puts broader enterprise impact 12-18 months out, and Datadog's largest customer shows AI usage and vendor revenue can move in opposite directions once the bill gets big enough to optimise. Add a potential rate hike and a Databricks IPO that would strip away the scarcity premium and force disclosure of how much AI revenue is genuinely governed-data demand, and the conclusion is that the sector's future earnings — good as the trajectory looks — are not yet strong enough to underwrite current valuations.

In short, we recommend caution when investing in this segment.  The cautious stance applies to the listed data infrastructure names covered in this note — Snowflake (SNOW), Datadog (DDOG), MongoDB (MDB) and Elastic (ESTC) — which trade at 48x–85x 2028 earnings and are therefore priced for near-perfect execution. Databricks is not rated here, as it remains privately held; we treat its potential 2027 listing as a catalyst risk to the group's multiples rather than as an investable instrument. While the long-term outlook remains bright, current valuations leave limited room for disappointment. Forward earnings growth does not yet appear sufficiently differentiated from peers to justify a BUY rating on the sector.

We would therefore look for either a more attractive valuation entry point or clearer evidence that AI adoption is generating incremental, high-margin revenue before turning more positive.

Figure 4: Estimated Earnings growth