Nvidia’s Q1 2026 earnings—$22.1 billion in revenue, 26% year-over-year growth—cemented its dominance in AI accelerators, but Wall Street’s money is quietly shifting to a darker corner of the software stack: enterprise automation platforms like ServiceNow. While Nvidia’s H100 and Blackwell GPUs remain the bedrock of AI infrastructure, a structural misalignment is emerging between hardware hype and the actual economics of deployment. The real play isn’t just about training larger LLMs; it’s about operationalizing them at scale, and that’s where ServiceNow, Salesforce, and Workday are positioning themselves as the “hidden beneficiaries” of AI’s infrastructure boom.
The AI Infrastructure Paradox: Why Nvidia’s Earnings Mask a Software Power Shift
Nvidia’s financials are a masterclass in platform lock-in. The company’s data center revenue—now 87% of its total—isn’t just about selling GPUs. It’s about selling an ecosystem where every layer, from the NPU (Nvidia Processing Unit) to the CUDA-X stack, is optimized for its hardware. But here’s the catch: the companies actually using those GPUs—enterprise software vendors—are the ones reaping the indirect benefits. ServiceNow, for instance, isn’t building its own AI chips. It’s embedding Nvidia’s inference engines into its workflow automation tools, then charging customers a premium for “AI-native” service desks.
This isn’t vaporware. ServiceNow’s AI-powered Now Platform already ships with pre-trained models for IT ticket triage, cost optimization, and even cybersecurity incident response. The models run on Nvidia’s A100/A10G GPUs in customer data centers, but the value capture happens at the software layer. Wall Street’s buy recommendation isn’t just about ServiceNow’s AI capabilities—it’s about the operational leverage it gains by sitting between Nvidia’s hardware and enterprise IT budgets.
The 30-Second Verdict
- Nvidia’s dominance is real, but its margins are being cannibalized by software layers. The company’s 26% YoY growth is impressive, but the real AI infrastructure play is happening in SaaS.
- ServiceNow’s AI isn’t revolutionary—it’s ruthlessly practical. No custom LLMs here. Just fine-tuned models for existing enterprise workflows, wrapped in a $100K/year subscription.
- Wall Street’s money is flowing to companies that solve the “last mile” of AI deployment. Training models is easy. Making them useful in a regulated, multi-cloud environment? That’s where the money is.
Under the Hood: How ServiceNow’s AI Stack Actually Works
ServiceNow’s AI isn’t built on proprietary architectures. It’s a composition of Nvidia’s CUDA cores, open-source transformers (like Mistral 7B), and a proprietary Workflow Orchestration Layer that routes API calls between models and legacy enterprise systems. The key innovation isn’t the model—it’s the latency-optimized API mesh that ensures sub-500ms response times for IT ticket classification, even when running on mixed x86/ARM infrastructure.
Here’s the breakdown of their current stack (as of the latest GitHub docs):
| Component | Technology | Latency (Avg.) | Deployment Model |
|---|---|---|---|
| Model Inference | Nvidia TensorRT + Mistral 7B (quantized to INT4) | 350ms (on-prem A100) 800ms (cloud) |
Customer-managed (GPU-accelerated) |
| Workflow Orchestration | ServiceNow’s FlowEngine (custom Rust/C++) |
120ms (API call routing) | SaaS-hosted (multi-region) |
| Data Pipeline | Apache Kafka + Snowflake (for training data) | N/A (batch processing) | Customer-controlled |
The real genius? ServiceNow doesn’t sell you the model. It sells you the integration. Their AI Core API lets enterprises plug in their own LLMs (e.g., Meta’s Llama 3) but defaults to ServiceNow’s fine-tuned variants for compliance-heavy tasks like HR case routing. This is strategic vendor lock-in disguised as flexibility.
“The enterprise AI market isn’t about who has the biggest model. It’s about who can make the model disappear into existing workflows without breaking IT governance. ServiceNow’s play is textbook—take the hype cycle, strip out the fluff, and sell the operational utility.”
Ecosystem Bridging: The Software Stack’s Silent War
This isn’t just a ServiceNow story. It’s a platform war between three factions:
- Nvidia: Controls the hardware and the inference layer. Its business model relies on selling GPUs, not software.
- ServiceNow/Salesforce: Controls the application layer. Their margins come from licensing, not hardware.
- Open-Source Communities: The wild card. Projects like Ollama and vLLM are building lightweight inference engines that could bypass both Nvidia’s and ServiceNow’s ecosystems.
The tension is already visible. Nvidia’s NIM (Nvidia Inference Microservice) is designed to lock customers into its ecosystem, but ServiceNow’s API-first approach lets enterprises choose their hardware—even if they’re incentivized to pick Nvidia. Meanwhile, open-source projects like MLCommons’ Inference Benchmark are pushing for standardized performance metrics that could force Nvidia to compete on price, not just proprietary features.
“We’re seeing a fragmentation of the AI stack. Nvidia owns the GPUs, but the real innovation is happening at the edges—where companies like ServiceNow are building vertical-specific AI layers. The open-source community is the only thing that can disrupt this, but so far, the enterprise market has zero tolerance for ‘works on my machine’ problems.”
Why This Matters: The Hidden Cost of AI Infrastructure
Here’s the dirty secret: most enterprises don’t want to build their own AI models. They want plug-and-play solutions that integrate with their existing tools. ServiceNow’s play is a masterclass in abstraction. By hiding the complexity of LLMs behind familiar UIs (like their Now Platform), they’re making AI adoption seamless—but at a cost.
The real economics of AI infrastructure aren’t in the training phase. They’re in the operationalization phase. And that’s where ServiceNow, Salesforce, and Workday are winning. Their pricing models (typically per-user, per-month) align perfectly with how enterprises budget for software. Nvidia’s GPUs, by contrast, require capital expenditure—a harder sell in a recessionary climate.
This is why Wall Street is bullish on ServiceNow. It’s not just about AI. It’s about recurring revenue in a market where hardware sales are cyclical and software subscriptions are sticky.
The 90-Second Takeaway: What This Means for You
- If you’re an enterprise IT buyer: The “AI” in ServiceNow’s pitch isn’t the headline—it’s the enabler. You’re not paying for a model; you’re paying for integration. Audit their API latency guarantees before committing.
- If you’re a developer: ServiceNow’s stack is closed, but their
AI CoreAPI is documented. If you’re building open-source alternatives, focus on interoperability—not just performance. - If you’re a hardware vendor: Nvidia’s lock-in is real, but the software layer is where the real competition will play out. Watch how ServiceNow’s API evolves—it’s the blueprint for the next generation of enterprise AI.
The Road Ahead: Will Open Source Break the Lock?
The wild card in this equation is open-source inference engines. Projects like Science-Llama and Llama Recipes are proving that you don’t need Nvidia’s hardware to run state-of-the-art models. But here’s the catch: enterprise adoption still favors stability over innovation.
ServiceNow’s play is a reminder that in the AI stack, the companies that control the user experience will always win. Nvidia can sell you the fastest GPU, but ServiceNow can sell you the illusion of simplicity. And in enterprise IT, simplicity is currency.
For now, the money is flowing to the software layer. But if open-source inference matures, the entire equation could flip. The question isn’t whether AI will disrupt enterprise software—it’s who will control the disruption.
