The week ahead pivots on risk. The Fed meets Wednesday with rate cuts off the table but market positioning stretched thin. Micron (MU: NASDAQ) reports Thursday, with investors watching for the faintest hint of AI overbuild—or undersupply. And in the background: the growing possibility of a widening war between Iran and Israel that could reorder energy and tech markets alike.

Last week, our coverage of Adobe (ADBE: NASDAQ) focused on the quiet strength in Digital Media, where Creative Cloud bookings and Firefly adoption continue to drive forward momentum. The slowdown in enterprise budgets is real—but Adobe remains one of the few scaled software firms still growing across users and upsells. Meanwhile, Oracle’s (ORCL: NYSE) strong close to fiscal 2025 highlighted accelerating demand for AI infrastructure—especially in sovereign and distributed cloud. One software giant is adapting to its buyers; the other is building the data centers behind them.

As always, I’m focused on three things:
1) Technology-driven change;
2) the latest in innovation and startup trends, and;
3) stock fraud.

Companies Discussed

In This Note:

• Companies Discussed

• The New Hyperscale King

• Tweet O’ The Week

• Epistrophy In The News

• 📆 of Epistrophy Events

• Availability This Week

The New Hyperscale King

Oracle’s (ORCL: NYSE) transformation into a cloud infrastructure heavyweight has arrived without fanfare but not without force. The company’s fourth-quarter results weren’t just strong — they were clarifying. Amid a frenzied AI arms race, Oracle has quietly become a critical supplier of the factories, software, and compute needed to train the next generation of reasoning models. While the rest of the market marveled at NVIDIA (NVDA: NASDAQ)’s quarterly spectacle, Oracle did something more useful: it began monetizing Blackwell at scale.

That matters now because the AI infrastructure race has entered its second phase. It’s no longer about which cloud vendor can say “AI” the loudest — it’s about who can build and deliver the scale, specialization, and price performance required for actual deployment. Oracle’s pitch, once dismissed as an also-ran cloud, now leans on differentiated architecture, a sprawling distributed cloud footprint, and aggressive partnerships with chipmakers. It’s starting to pay off.

The technical foundation of Oracle’s recent success rests on two pillars: hyperscale GPU availability and distributed cloud orchestration. In the June 12 release, Oracle announced the general availability of NVIDIA GB200 NVL72 systems on its Supercluster infrastructure, with the ability to scale up to 131,072 GPUs. That’s not just a bragging right — it’s a signal to enterprises that Oracle can meet or exceed the capacity constraints limiting its competitors. It’s also a milestone in liquid-cooled rack-scale design, an area Oracle has invested in heavily to optimize power and density for training large models.

Unlike its peers, Oracle isn’t just listing NVIDIA hardware through a marketplace. It has made NVIDIA AI Enterprise software available natively through the Oracle Cloud Infrastructure (OCI) Console. That means customers can use existing Oracle Universal Credits, integrate AI workloads across dedicated regions, sovereign clouds, and public regions, and receive direct billing and support — all of which reduce deployment friction. The move also unlocks 160+ AI tools and microservices, including NVIDIA NIM, a family of cloud-native inference services designed for rapid GenAI deployments.

At the same time, Oracle is positioning itself as a launch partner for AMD (AMD: NASDAQ)’s Instinct MI355X, announcing a zettascale cluster with up to 131,072 of the new GPUs — again, in OCI Supercluster. The platform promises 2.8x higher throughput over previous generations, with support for FP4 precision and 288 GB of HBM3 per GPU. That’s built for production-scale inference. Oracle will also be the first cloud provider to deploy AMD’s new Pollara NICs, using the Ultra Ethernet Consortium’s RoCE standards to offer lower latency networking at high bandwidth.

Taken together, Oracle’s GPU offerings make a powerful claim: whether customers want NVIDIA or AMD, inference or training, public or private deployment, Oracle can deliver. And increasingly, Oracle is pitching its infrastructure not only as general-purpose cloud compute, but as a specialized foundation for AI “factories”—production-scale inference environments customized to regional or vertical use cases.

The database business, long Oracle’s profit engine, is starting to benefit from this AI infrastructure credibility. Oracle said its database consumption revenue grew 50% year over year, driven by customers deploying Oracle Autonomous Database and Exadata Database Service on OCI. The company is leaning into the idea that AI infrastructure and data services are a flywheel — GPU clusters require fast, secure, and co-located data access, and Oracle’s database stack sits naturally atop that architecture.

Oracle ended the quarter with $138  billion in remaining performance obligations (RPO), up 29% year over year. These figures underscore Oracle’s visibility and booking strength. RPO is not a guarantee of future revenue, but it is a strong proxy for customer commitment to long-term infrastructure projects, especially among regulated industries where deployment timelines stretch across years.

In terms of financial results, Oracle delivered $14.29 billion in total revenue for the quarter, up 3% year over year (8% in constant currency). Cloud services and license support revenue grew 9% to $10.2 billion, while hardware fell 6% to $700 million. Non-GAAP operating margin was 41%, flat from the prior year. Earnings per share were $1.63 on a non-GAAP basis and $1.51 GAAP. The company generated $5.5 billion in free cash flow and spent $2.7 billion on capital expenditures — a clear signal that Oracle is still building.

But the story here isn’t quarter-to-quarter comparisons. It’s trajectory. Oracle’s multiyear cloud pivot is starting to look credible even to skeptics, not because it matches AWS or Azure on scale, but because it offers a coherent alternative. The company’s pitch — lower-cost, high-performance GPU infrastructure, backed by software integrations and regional compliance flexibility — resonates in a world where AI demand is global but governance is local. Governments, hospitals, and sovereign cloud clients want AI factories that don’t compromise compliance. Oracle now speaks their language.

That positioning has consequences for the competitive landscape. Amazon's (AMZN: NASDAQ) AWS and Microsoft (MSFT: NASDAQ) are still dominant, but their scale has begun to cut both ways. Both companies operate vast, general-purpose infrastructure, but their AI offerings are largely gated behind their own models (SageMaker, Azure OpenAI). Oracle, by contrast, is pitching neutrality — come train your model here, whatever it may be. That posture echoes Nvidia’s approach to silicon: fuel the ecosystem, not just your own stack.

The bet seems to be working. Oracle highlighted new AI customer wins across Europe, including Almawave in Italy and Cerebriu in Denmark, both using OCI for domain-specific models in language and radiology, respectively. These aren’t headline-grabbing logos — but they are the kinds of localized, high-intensity use cases where OCI’s distributed design and cost structure shine.

Looking forward, Oracle’s challenge will be scale. Demand is not the issue — supply is. Meeting the computational and network needs of 100,000-GPU clusters requires not just chips, but cooling, interconnects, land, power, and capital. Oracle’s success hinges on whether it can keep building at this pace, particularly as the U.S. government tightens restrictions on chip exports and geopolitical constraints ripple through the supply chain.

Yet in a market increasingly defined by bifurcation — between general-purpose cloud and AI-specialized cloud — Oracle seems to have found its lane. It may not win every hyperscaler contract. It may never catch AWS in market share. But if it continues to assemble infrastructure at this clip, and pair it with tools that developers actually use, it won’t have to. OCI doesn’t need to be for everyone. It just needs to be where serious AI work gets done.

And in this quarter, it was.

Tweet O’ The Week

Epistrophy In The News

On Schwab Network with Nicole Petallides, I walked through the limitations of the current “Magnificent Seven” and proposed a fresh basket for this moment. Oracle (ORCL: NYSE), Netflix (NFLX: NASDAQ), Palantir (PLTR: NYSE), and IBM (IBM: NYSE)—not Nvidia, Meta, or Apple—may offer better signals for what’s next in AI infrastructure, services, and software. A quick screen of trailing and forward metrics supports the case. It's time to revise the canon.

📆 of Epistrophy Events

Availability This Week

I'm attending KLA Investor Day midweek and off Wednesday for Juneteenth, but otherwise reachable. I’ll be working on several pieces related to sovereign cloud, memory cycles, and our Magnificent Seven rethink. If you're exploring any of those angles—or want to challenge them—I’m all ears.

Written reports are available to clients, with video summaries on YouTube, and of course our popular summaries of the summaries on Instagram, TikTok, and YouTubeShorts.