In a widely circulated essay published on July 12, Microsoft (MSFT 0.74%) CEO Satya Nadella identified a fundamental imbalance in how companies deploy artificial intelligence (AI). He argued that enterprises essentially pay for intelligence twice, first through tokens and again with the proprietary knowledge they must reveal to make AI useful.
Every prompt and correction an organization feeds a model becomes “exhaust,” a trail of insights about how an organization operates that leaks out, “trace by trace.” To counter this “Reverse Information Paradox,” Nadella argued that companies must take control of their data by running AI within their own “tenant boundary.”
Microsoft CEO Satya Nadella. Image source: Microsoft Corporation.
The cost of protecting enterprise IP
Today, companies use three primary paths to access frontier AI models. They can use large cloud platforms such as Azure OpenAI Service or Amazon‘s (AMZN +0.85%) AWS Bedrock, a direct application programming interface (API) from model providers, or a hybrid of both.
The cloud path doesn’t require expensive hardware, and the direct route is well established by companies like Anthropic, which built a multibillion-dollar business on direct enterprise contracts. Regardless of the path, the risks Nadella described remain.
The model provider’s infrastructure inevitably captures the usage patterns and query data that reveal how an enterprise operates. In response, sovereign nations and regulated entities are increasingly investing in their own hardware.
Hewlett Packard Enterprise (HPE 0.02%) closed its second quarter with a $6 billion AI server backlog, roughly 60% of which came from sovereign nations and enterprise clients. Through its acquisition of Juniper Networks last year, HPE can now sell an integrated on-premise stack of servers, storage, and networking to organizations building their own AI capabilities.
Meanwhile, Dell Technologies‘ (DELL 0.59%) backlog continues to grow. The company reported over $24 billion in AI orders last quarter and exited with a $51 billion backlog. Management noted that customers are seeking integrated solutions they can deploy on infrastructure they control, with specific products designed to keep sensitive data and IP on-premise.
The hyperscalers are positioned to profit either way
Once enterprises own the hardware, they need orchestration tools to route between models without getting locked into a single provider. Microsoft’s Azure AI Foundry, AWS Bedrock, and Alphabet‘s Vertex AI offer model-agnostic capabilities.
The architecture supports hybrid adoption, enabling companies to run workloads across multiple clouds and on-premises systems. Nadella’s essay makes more sense against this backdrop, as his company positions itself to capture value from the shift he warned about.
After a volatile week in the markets, AI-related stocks are taking a breather following a historic run. With shares trading at just 13 times this year’s earnings estimates, HPE warrants a closer look.
Hewlett Packard Enterprise
Today’s Change
(-0.02%) $-0.01
Current Price
$45.81
Key Data Points
Market Cap
Day’s Range
$45.33 – $46.31
52wk Range
$19.64 – $64.25
Volume
98.2K
Avg Vol
26.1M
Gross Margin
31.38%
Dividend Yield
1.22%
The company assembles AI server systems, but margin expansion depends on its ability to attach networking and storage contracts. At this valuation, the stock offers a reasonable entry point for investors seeking exposure to enterprise infrastructure spending.
