After Nvidia, where does the AI money go next?

Key takeaways

• Nvidia confirms strong AI computing demand, but the spending chain is becoming much broader.

• Proprietary data and workflows may gain value as powerful AI models become easier to access.

• More autonomous AI creates new needs around identity, security and recovery.

Nvidia’s results on 26 August 2026 confirmed that demand for artificial intelligence (AI) computing remains enormous. But this earnings week also suggested that the AI opportunity is starting to spread well beyond the graphics processing unit (GPU).

Across chips, software and cybersecurity, a clearer picture is emerging of what comes next. AI increasingly needs three things to scale: compute to run, context to become useful, and control to be trusted.

Compute is becoming more specialised

Marvell designs chips and networking technology for data centres. Revenue rose 37% to a record 2.74 billion USD, with Data Center revenue up 46%.

Its results also show the gap between winning a design and earning the revenue. Marvell secured a major custom-chip agreement with Google, but management said the biggest contribution should arrive later, particularly from fiscal 2029.

Synopsys sells software and intellectual property used to design complex chips and systems. Revenue reached 2.48 billion USD, up about 42%, while management raised its outlook and pointed to AI-driven complexity.

As hyperscalers build more customised infrastructure, value can spread towards companies that help design, connect and optimise specialised systems.

Context may become software’s new moat

Powerful AI models increasingly have similar general capabilities. What they do not automatically have is a company’s customer history, payroll records, engineering designs and business rules.

That gives established software companies an interesting asset: context.

Salesforce stores customer and sales data. Agentforce and Data 360 annual recurring revenue (ARR) reached nearly 3.9 billion USD, up more than 210%. Its expanded Anthropic partnership makes the logic clear: Salesforce can make Claude more useful by connecting it to trusted data, workflows and governance.

Workday manages human resources and finance processes. AI products generated more than 25% of new annual contract value, while more than 5,500 customers now use at least one Workday agent.

Autodesk extends the argument into engineering and construction. Its software sits inside project workflows, giving AI access to rich technical context. Management argues that useful AI in the physical world requires trusted project data.

If intelligence becomes easier to access, the moat can migrate from owning the model to owning what makes the model useful.

Control becomes more valuable as AI gains autonomy

The third layer appears when AI stops answering questions and starts taking actions.

A company running thousands of agents must know which agent is acting, what it can access and how to recover if something goes wrong.

CrowdStrike protects devices and cloud systems. Its Falcon Flex ARR exceeded 2.29 billion USD, up 101%, suggesting customers increasingly want a broader security platform.

Okta manages digital identities and access. Revenue grew a more modest 11%, which is a useful reminder that a convincing AI story is not the same as proven monetisation. Still, every agent needs an identity and permissions.

Rubrik adds recovery and data protection. Subscription ARR rose 33% to 1.66 billion USD, while its strategy increasingly includes securing AI-agent activity.

More autonomous AI can therefore create some of the security spending needed to make further adoption possible.

The risks are moving too

Custom-chip projects can take years to scale. Software companies may own valuable data without successfully charging more for AI access. Security vendors may face tougher competition as larger platforms bundle more tools.

Watch delayed production ramps, weak AI-related contract growth and rising usage that fails to become recurring revenue.

Investor playbook

• Separate AI exposure into compute, context and control rather than treating every AI company as the same trade.
• Track conversion, not announcements. Design wins, pilots and agent usage matter when they become revenue and cash flow.
• Ask whether AI strengthens an existing moat, such as proprietary data or embedded workflows.
• Keep diversification in mind. The AI value chain is widening, but expectations can widen even faster.

From compute to context and control

Nvidia remains the clearest proof that the AI infrastructure boom is alive. Yet this earnings week suggests the opportunity is becoming broader and more layered. Marvell and Synopsys help build specialised infrastructure. Salesforce, Workday and Autodesk give AI the business context it needs. CrowdStrike, Okta and Rubrik help companies trust it enough to act.

That does not make every company in the chain a winner. It does give investors a more useful map. The first phase rewarded scarce computing power. The next may reward scarce context and control. Compute makes AI possible, context makes it useful, and control makes it deployable at scale.