Oracle Widens Chip Sourcing Beyond Nvidia, Adds Google and Cerebras to Cloud Lineup

Oracle's latest cloud infrastructure announcement tells two stories at once, and investors should read both carefully.

The headline move is an expansion of Oracle's existing chip supply agreement with Nvidia. The companies said Tuesday that Oracle will scale up to "multi-gigawatt" deployments using Nvidia's newest "Vera Rubin" AI server generation. Neither side disclosed deal value. For context, Nvidia CEO Jensen Huang has previously pegged Vera Rubin systems as capable of generating up to $40 billion in revenue per gigawatt for the chipmaker. If Oracle's commitment reaches even a fraction of multi-gigawatt scale, this is a material revenue input for Nvidia over coming fiscal years.

But the quieter story matters more for how we should think about Oracle's long-term positioning. Oracle is diversifying its chip stack in ways that reduce its single-vendor dependency. Even within its Nvidia-based systems, Oracle said it will bypass some of Nvidia's own networking chips in favor of its in-house "Acceleron" networking silicon. That's a direct margin and control play: fewer dollars flowing to Nvidia's networking stack, more value captured internally.

More significant still, Oracle will begin offering Google's tensor processing units for the first time. It's also adding chips from Cerebras Systems and Intel-backed SambaNova Systems, both specialists in AI inference rather than training. Mahesh Thiagarajan, president of Oracle Cloud Infrastructure, framed the logic plainly: "Different AI workloads have different compute needs. That choice gives customers the flexibility to choose the right infrastructure for the performance, throughput and economics of each AI workload."

Why Diversification Changes the Calculus

Nvidia's dominance in AI training chips remains largely unchallenged. The inference market, where trained AI models actually respond to user requests, is a different competitive landscape entirely, and it's considerably more fragmented.

That fragmentation is exactly what Oracle appears to be exploiting. By adding Cerebras and SambaNova to its inference offerings, Oracle positions itself as a multi-vendor marketplace rather than a single-architecture shop. Cerebras CEO Andrew Feldman was candid about what this means for his company's market reach. "That's a massive expansion of our customer base of tier-one customers," Feldman told Reuters. "We're being deployed in other people's data centers at large scale."

For Cerebras, landing inside Oracle's infrastructure is a credibility and distribution win that's hard to overstate for a company still establishing itself against Nvidia's scale. For Oracle, it's a hedge. Cloud providers that lock themselves into a single chip architecture expose customers, and their own margins, to whatever pricing power that single vendor commands. Diversified sourcing gives Oracle negotiating leverage and insulates it somewhat from any single supplier's production constraints or price increases.

There's a financing dimension worth flagging too. Oracle has not disclosed how it plans to fund these expanded chip purchases. During its earnings call last month, the company said customers are increasingly footing the bill for chip costs directly, a shift that changes the capital intensity calculus for Oracle's own balance sheet. Oracle maintained its capital spending target of roughly $90 billion to $95 billion for fiscal 2027, a figure that underscores just how much capital is now flowing through the AI infrastructure buildout industry-wide.

The timing is notable. Oracle is set to hold an investor day this week, and this announcement reads as a preview of the narrative management wants to set: aggressive scale with Nvidia, paired with strategic independence from any single supplier. That's a reasonable story to tell shareholders nervous about capital intensity and vendor concentration risk in the AI infrastructure race.

Key Risks

The risks here are not trivial. Multi-vendor chip strategies introduce real engineering and integration complexity. Running Nvidia GPUs, Google TPUs, Cerebras wafer-scale chips, and SambaNova inference hardware across a unified cloud platform is operationally harder than standardizing on one architecture. Customers may also find the "flexibility" pitch less compelling in practice if performance tuning across heterogeneous hardware proves costly or slow.

There's also the unresolved financing question. Oracle's capital spending guidance of $90 billion to $95 billion for fiscal 2027 is enormous relative to the company's historical capex base, and shifting more of the cost burden onto customers, as Oracle indicated last month, changes the revenue recognition and margin profile of these deals in ways that aren't yet fully transparent to investors.

Nvidia's position, while still dominant in training, faces a slow erosion of its networking revenue as Oracle and presumably other large customers build or source in-house alternatives like Acceleron. That's a small crack, not a collapse, but worth monitoring over successive quarters.

The Bottom Line

Oracle is doing something sensible: expanding with the market leader while quietly reducing single-vendor exposure. The Nvidia relationship remains the core of Oracle's AI infrastructure story, and the Vera Rubin expansion to multi-gigawatt scale is a genuine vote of confidence in Nvidia's roadmap. But the addition of Google TPUs, Cerebras, and SambaNova signals that Oracle sees concentration risk as a real threat to long-term margins and negotiating leverage.

For investors, the signal to watch isn't the Nvidia headline. It's whether Oracle's diversified chip strategy actually translates into better unit economics and customer retention over the next several quarters, particularly as the company's investor day this week should offer more detail on financing structure and capital allocation. Until then, the capex guidance of $90 billion to $95 billion for fiscal 2027 remains the single most important number to track against actual cloud infrastructure revenue growth.