We cover the fifty most interesting companies in technology. Last week, we broke down major reports from Marvell (MRVL: NASDAQ) and Hewlett Packard Enterprise (HPE: NYSE).

This week, the focus shifts to Adobe (ADBE: NASDAQ), Oracle (ORCL: NYSE), and Rubrik, with big implications for software, AI, and cybersecurity. Our preview is below.

Subscribe to our YouTube channel for quick-hit video summaries of our research. Less reading, more watching. (And if you play it at 2x speed, you can enjoy the Alvin and the Chipmunks take on tech analysis. We won’t be offended.)

As always, we’re focused on three things:

1. Technology-driven change

2. The latest in innovation and startup trends

3. Stock fraud

Companies Discussed

In This Note:

• Companies Discussed

• Oracle’s AI: Bigger, Harder, Faster

• ASICs: Broadcom’s AI Crown

• SEC FOIA Logs: Hidden Clues?

• Epistrophy In The News

• 📆 of Epistrophy Events

• Availability This Week

Oracle’s AI: Bigger, Harder, Faster

Oracle (ORCL: NYSE) has spent decades as an enterprise software company, a dominant but unremarkable presence in database technology. That reputation is changing, and it’s been hard. The company is emerging as one of the most aggressive builders of artificial intelligence infrastructure, challenging Amazon Web Services (AMZN: NASDAQ), Microsoft Azure (MSFT: NASDAQ), and Google Cloud (GOOG: NASDAQ) in a market they once controlled outright. Oracle Cloud Infrastructure (OCI) is becoming the preferred platform for high-performance AI workloads, particularly for companies that need large-scale GPU clusters at lower costs than its competitors offer.

Oracle’s technical advantages in AI computing are rooted in its cloud design. Unlike AWS and Azure, which operate general-purpose cloud services with legacy constraints, OCI was engineered from the ground up for high-performance computing. This allows Oracle to offer better networking speeds, lower latency, and more efficient use of GPUs for training AI models. NVIDIA has partnered with Oracle to deploy the largest GPU clusters on any cloud platform, including a 65,000-GPU NVIDIA H200 supercomputer built on OCI. The cost of this supercomputer is substantial – given that a single NVIDIA H200 GPU can cost between $31,000 and $32,000 it could exceed $2 B for the chips alone. That scale gives Oracle an edge over Microsoft’s AI infrastructure, which is built on Azure but has been criticized for network bottlenecks that slow AI model training. AWS, meanwhile, has bet heavily on its own Trainium and Inferentia chips rather than NVIDIA hardware, a move that limits its appeal to AI firms that rely on NVIDIA’s CUDA ecosystem.

Oracle’s AI push is already reflected in its financial performance. In fiscal Q2 2025, the company reported $14.1 billion in revenue, up 9% from the previous year. Cloud infrastructure revenue grew 52%, driven by AI-related workloads. OCI’s GPU consumption revenue surged 336%, and Oracle’s remaining performance obligations—a measure of contracted future revenue—grew 50% to $97 billion. The company expects cloud revenue to exceed $25 billion for the fiscal year.

Beyond raw computing power, Oracle is integrating AI into its core products. The Oracle 23ai database, released in 2024, includes built-in vector search capabilities for AI-driven applications, competing directly with Microsoft’s AI-enhanced SQL Server and Google’s AlloyDB. Oracle is also embedding AI across its enterprise applications, including its Fusion Cloud ERP and NetSuite platforms, which now use generative AI for financial forecasting and supply chain optimization. Unlike Microsoft’s Copilot, which relies on third-party AI models running on Azure, Oracle’s AI applications run natively on OCI, offering customers lower costs and better integration with existing workflows.

Oracle has engineered major cost advantages into its cloud infrastructure offerings, setting OCI apart from the pricing practices of rival providers. One key differentiator is OCI’s “uniform global pricing” – Oracle does not impose the regional price uplifts that AWS, Azure, and GCP routinely do. OCI charges the same rates worldwide for compute, memory, and storage​, whereas the big three often add regional premiums of up to ~41% for identical services outside their cheapest US regions​. Oracle’s compute pricing is also dramatically lower on a per-instance basis. For example, independent analysis of comparable AMD EPYC-based instances found AWS and Azure to be over 2.3× more expensive than OCI (and GCP about 2× more) for the same performance. Even when competitors’ long-term discounts are considered, Oracle still holds a lead – OCI’s pay-as-you-go rates undercut AWS’s 1-year committed prices by ~46% (and GCP’s by ~30%) for equivalent new CPU. Additionally, OCI reduces hidden fees by including generous data egress allowances (the first 10 TB/month of outbound data is free)​, easing a notorious pain point where AWS and others levy high charges to move data out of their cloud. Together, these factors give Oracle a clear pricing edge in cloud compute and network services, allowing customers to run workloads on OCI at significantly lower total cost compared to the same usage on AWS, Azure, or GCP.

These cost advantages directly bolster Oracle’s momentum in the cloud and AI markets. Oracle’s leadership openly emphasizes that its cloud is not only high-performance but also faster and is less expensive than other clouds.

By offering superior price-performance and easing data transfer costs (OCI’s free 10 TB egress effectively removes a tax on moving large datasets, Oracle enables AI developers and enterprises to train models and run data-intensive applications more economically than on other platforms. This compelling value proposition is translating into tangible business gains for Oracle. The company reported that record AI-driven demand helped drive a 52% surge in Oracle Cloud Infrastructure revenue​, reflecting how cost savings are attracting big AI and high-performance workloads to OCI. Oracle even projects its total cloud revenue to reach $25 billion this year, underscoring that its aggressive pricing strategy – coupled with technical strengths – is resonating with customers and fueling OCI’s growth in both AI-specific and general cloud services..

At the center of this shift is Oracle’s $500 billion AI infrastructure project, Stargate. Announced in January 2025 with OpenAI, SoftBank, and MGX, the project will fund the construction of AI data centers across the United States, starting with ten locations in Texas. The initiative, introduced at the White House alongside President Donald Trump, is designed to accelerate AI adoption while reducing reliance on AWS, Azure, and Google Cloud. Larry Ellison, Oracle’s chairman and chief technology officer, has positioned Stargate as a national AI infrastructure effort, emphasizing Oracle’s ability to deliver better performance at lower cost than its rivals.

The impact of Oracle’s AI investments extends beyond its own business. The Meta deal to train its Llama model on OCI was a turning point, signaling that even the largest AI firms are willing to move workloads away from Microsoft and Google. Oracle’s partnership with OpenAI further demonstrates its ability to offer a viable alternative to AWS and Azure. The Stargate project, if executed as planned, will reshape AI infrastructure, giving customers a credible third option in a cloud market long dominated by two players.

Oracle is set to report fiscal Q3 earnings on March 10. The results will provide more clarity on whether its AI expansion is sustainable. For now, Oracle’s bet on specialized cloud infrastructure is paying off, and its competitors—Microsoft, AWS, and Google—are being forced to respond.

ASICs: Broadcom’s AI Crown

With Q1 2025 results, we learned that Broadcom (AVGO: Nasdaq)’s AI business isn’t just growing—it’s exploding. The company reported total revenue of $11.96 billion, up 34% year-over-year, driven by surging AI demand. AI revenue surged 77% year-over-year to $4.1 billion in Q1 2025, easily topping its $3.8 billion forecast. The company’s Q2 guidance—$4.4 billion in AI revenue, up another 44%—cements its position at the center of the AI infrastructure boom. Two new AI customers—likely Apple (AAPL: Nasdaq) and OpenAI—joined Broadcom’s roster of hyperscalers, which already includes Google, Meta (META: Nasdaq) and ByteDance. It’s an undeniable show of strength.

“Broadcom’s largest technology customers continue to ‘invest aggressively’ in their next-generation AI models… each racing toward building a 1 million AI chip cluster by the end of 2027,” Hock Tan said on the earnings call. 

That’s an amazing number. To put this in perspective, Nvidia’s (NVDA: NASDAQ)’s H100 GPUs, which power many of today’s AI data centers, cost around $25,000 per unit. Broadcom’s custom ASICs is likely half that cost (given their relative simplicity and higher yields) so outfitting a single million-unit cluster would exceed $12.5 billion in hardware costs alone. And that’s before adding networking, storage and cooling. Networking, driven by the shift to high-bandwidth Ethernet, could tack on billions, especially with Broadcom’s Tomahawk 5 and Jericho3-AI switching silicon which already dominate cloud data centers, with the latter optimized for congestion-free AI workloads and large-scale fabric connectivity. Running a cluster of this scale would also demand dedicated data centers equipped with Broadcom’s PAM4 SerDes technology for high-speed interconnects and a power draw on par with a small city. If Tan’s claim holds, AI infrastructure spending is about to hit an entirely new stratosphere.

The AI hardware market is heating up and Broadcom sits at the heart of it. Companies racing to build massive AI clusters need high-performance custom silicon. That’s where Broadcom thrives. Its application-specific integrated circuits (ASICs) are tailor-made for hyperscalers looking for efficiency at scale. Unlike off-the-shelf chips from NVIDIA or AMD (AMD: NASDAQ), Broadcom’s AI accelerators, including its NetXtreme Ethernet controllers and Stingray SmartNICs, are designed for specific workloads, giving customers a competitive edge in power efficiency and throughput.

Tan’s confidence is unwavering. “Broadcom’s record first-quarter revenue and adjusted EBITDA were driven by both AI semiconductor solutions and infrastructure software,” he said on the March 6 conference call. “We expect continued strength in AI semiconductor revenue of $4.4 billion in Q2, as hyperscale partners continue to invest in AI XPUs [accelerators] and connectivity solutions for AI data centers.” The term ‘XPU’ is key. While CPUs and GPUs have dominated computing for decades, XPUs represent a broader category of heterogeneous accelerators—including ASICs, FPGAs, and custom AI chips—optimized for AI workloads. Broadcom is betting that hyperscalers will favor highly customized silicon over generic GPUs as AI models scale in size and complexity, leveraging its custom-built AI ASICs and Broadcom PEX PCIe switches for high-performance interconnects.

The competitive landscape is another factor. NVIDIA still dominates the AI chip market and companies like AMD and Intel (INTC: NASDAQ) are vying for a bigger slice. Even as hyperscalers favor custom ASICs, Broadcom’s foothold isn’t guaranteed. Internal chip development efforts at Google (TPUs) and Amazon (Trainium, Inferentia) highlight a trend toward vertical integration that could eventually squeeze third-party suppliers.

Broadcom’s future in AI hinges on execution. If it can maintain technological leadership in AI-specific ASICs and networking, while gradually diversifying its customer base, it will remain an AI powerhouse.

SEC FOIA Logs: Hidden Clues?

The Securities and Exchange Commission released its January 2025 Freedom of Information Act (FOIA) logs, detailing companies subject to B7A exemption denials. While the SEC aims to keep formal investigations confidential, these denials—issued when disclosure could interfere with enforcement proceedings—serve as subtle indicators. 

Academic research, such as “Does FOIA Foil the SEC's Intent to Keep Investigations Confidential?”, demonstrates that these denials often precede significant regulatory actions. Another study, “Private Information Acquisition via Freedom of Information Act Requests: Evidence from the Securities and Exchange Commission”, supports these findings, suggesting that exemption denials can signal ongoing investigations.

Historically stocks such as Nikola (NKLA: NASDAQ) and Lordstown Motors (RIDE: NASDAQ) appeared in SEC exemption denials before plummeting amid fraud allegations and regulatory investigations. The aforementioned studies found that exemption denials were significant predictors of future SEC actions, implying that astute investors analyze these lists for early warnings. In a jungle full of opaque financials and ambitious growth claims – this being the Golden Age of Fraud – this could be a useful list. Because it’s a ju

Epistrophy In The News

On NewsNation with Connell McShane, I discussed President Trump's recent executive order establishing a Strategic Bitcoin Reserve and a U.S. Digital Asset Stockpile. While the administration presents this as a move to legitimize and stabilize the cryptocurrency market, critics argue that it primarily benefits existing crypto holders by potentially inflating asset values. This initiative raises concerns about market manipulation and the ethical implications of government intervention in favor of specific financial assets.

📆 of Epistrophy Events

Availability This Week

I’ll be in San Francisco all week and available anytime via email or text (and I’m headed to Los Angeles, Washington D.C. and New York the following week). Oracle’s results might well be the biggest story next week, but with the news cycle as chaotic as ever, surprises are inevitable.

You can access our analysis in multiple ways:

• Full written reports for clients

• Video breakdowns on YouTube

• Executive summaries on Instagram and TikTok (@drilldownpod)

If you have questions, insights, or a sharp take on economic indicators and enterprise tech spending, I’d love to hear it. And if you know others who’d benefit from this analysis, send them my way.

Looking forward to the discussion.