Two of the most powerful firms in enterprise tech take center stage this week—and they’re more alike than you think. Apple (AAPL: NASDAQ) holds its WWDC on Monday, unveiling software that ties together the company’s ecosystem with some new found AI functionality. Meanwhile, Oracle (ORCL: NYSE) will report earnings that show how aggressively it’s turning into an infrastructure company for the very same AI boom.

Our recent research cuts through the earnings choreography. "The $2,093 iPhone: Six Reasons Apple Won’t Be Made in America" lays out the cold math behind domestic production fantasies, while "Oracle’s AI: Bigger, Harder, Faster" details how Ellison’s cloud is built to sell compute to the same startups driving Nvidia’s order book. If these two firms move markets this week, you’ll already know why.

The Epistrophy website—epistrophy.beehiiv.com—is temporarily open to all readers, no password required. Check it out: it’s pretty!

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

Table of Contents

• Companies Discussed

• Broadcom’s Big Turn

• Trump’s Big Beautiful Ban of AI Regulation

  • Why the US needs Federalism, not this Federal ban

• Tweet O’ The Week

• Epistrophy In The News

• Epistrophy In The News

• Availability This Week

• 📆 of Epistrophy Events

• Availability This Week

Broadcom’s Big Turn

Broadcom (AVGO: NASDAQ)’s powerful Q2 results were headlined by a record $15 billion in revenue and an eyebrow-lifting 67 percent adjusted EBITDA margin. The numbers matter less than the velocity: Q2 AI semiconductor sales surged 46 percent year on year to $4.4 billion—nine straight quarters of growth that no longer looks like a streak but a regime.

That regime matters now because the industry is climbing out of the brief but bruising semiconductor downturn. Non-AI chip revenue fell only 5 percent from last year and, at $4 billion, appears to have found bottom; broadband and enterprise networking already ticked higher. Free cash flow, meanwhile, hit $6.4 billion—43 percent of revenue—giving Hock Tan room to lean in while leaner rivals are still picking themselves off the mat.

Broadcom’s technical hand is stronger than even the headline growth suggests. Custom XPUs now anchor three hyperscalers and four prospects, while AI networking—built on the decidedly unfashionable Ethernet—supplied 40 percent of AI sales. The new Tomahawk-5 switch moves 102.4 terabits per second and flattens clusters to two tiers, slicing latency and power draw inside racks that can exceed 100,000 accelerators. Tan’s early bet on Ethernet for scale-up as well as scale-out, once heresy in GPU land, is suddenly orthodoxy.

That bet is one of three moves that define Tan’s last half-decade. First, he opened Broadcom’s ASIC tool-chain to customers, letting cloud giants hard-wire their own silicon instead of renting Moore’s Law from Nvidia (NVDA: NASDAQ) or AMD (AMD: NASDAQ). Second, he doubled down on Ethernet as the lingua franca of AI fabrics, avoiding the proprietary cul-de-sacs that trap competitors. Third, he bought VMware and flipped it from perpetual licenses to full-suite subscriptions; 87 percent of the top 10,000 customers now run VCF, positioning Broadcom’s software arm as a private-cloud on-ramp for AI workloads that never touch the public cloud.

Scale begets leverage. By running semiconductors and software under one roof, Broadcom can tune silicon, switches and virtualization layers for the same customers—then pocket margin twice. The approach also inoculates the company against the next cycle: if AI demand cools, infrastructure software’s 93 percent gross margin and sticky recurring revenue soften the blow. Competitors focused on single markets will feel the swing more acutely; suppliers of proprietary interconnects may wake up to find the data center has quietly standardized around open protocols they do not control.

“We anticipate our fiscal 2025 growth rate of AI semiconductor revenue to sustain into fiscal 2026,” Tan said on the June 5 earnings call. If that holds, Broadcom’s AI engine lands north of $20 billion next year, backed by a software annuity and a balance sheet that can retire debt or fund the next acquisition. The company is still mortal—export rules and hyperscaler capex discipline lurk—but momentum has a habit of rewriting risk.

Trump’s Big Beautiful Ban of AI Regulation

Why the US needs Federalism, not this Federal ban

A decade-long freeze on state AI rules, nested inside the House’s “One Big, Beautiful Bill,” (which is, of course, an “Act” not a bill, but I guess everyone isn’t so versed in School House Rock)  yanks the regulatory brake just as artificial intelligence moves from lab demo to critical infrastructure. States that pioneered guardrails—California on driverless trucks, Colorado on algorithmic bias, Tennessee on deepfake impersonations—would be benched until 2036, while cloud providers, chipmakers, and platform networks sprint ahead under a single federal shield. Investors and engineers celebrate the simplicity; litigators and local officials see an unfillable vacuum.

The legal frame now slated for mothballs dates to Rule 240 of 1938 – an FCC order shielding phone companies from what callers said. The Telecommunications Act of 1996 carried that logic online: Section 230 declared that networks and websites could not be sued for user speech, leaving content policing to market forces. Over the next quarter-century those forces produced transformer models that draft contracts, fleets of autonomous cars and generative systems that can design proteins. Liability doctrine never kept pace, and today’s patchwork of state proposals—mandatory algorithmic audits, autonomous-vehicle kill switches, disclosure rules for synthetic media—represents local attempts to update the bargain. The moratorium would halt that process cold.

There are massive financial implications. On one hand, unfettered by restrictive laws, big AI companies could grow at a 5% faster annual rate—a wild estimate, but directionally useful. Given their current $9 trillion combined market cap, that implies an additional $5.7 trillion in value over the next decade.

On the other hand, liability could reprice that gain overnight. A fatal robo-taxi pile-up or a deepfake that craters a municipal-bond sale might restore the old risk premium, wipe out the uplift, and stack on higher insurance costs, bigger contingent-liability reserves, and fresh class actions. The “AI-law holiday” looks less like a free gift than a knife-edge cost curve.

What is there to worry about? These problems are easy to imagine. 

“Shadow-tech” medical advice – A popular chatbot, masquerading as a wellness coach, quietly pulls treatment regimens from fringe subreddits. Patients follow unvetted dosage schedules that interact fatally with prescription drugs; no medical-device clearance is triggered because the bot claims only to offer “lifestyle tips.”

Autonomous-fleet hijack – A single spoofed over-the-air update seeds rogue code across thousands of robo-taxis. Vehicles receive a synchronized “hard-left” command during rush hour, blocking emergency routes and causing chain-reaction collisions city-wide.

Hyper-personal coercion – Campaign operatives fine-tune a language model on private therapy-forum dumps. The bot micro-targets swing-state voters with messages engineered to exploit individual traumas (“Your anxiety will spike if polling lines are long—better stay home”).

Synthetic blackmail 2.0 – An image generator stitches innocuous vacation photos into convincing CSAM composites. The blackmailer auto-emails employers and local police, giving the target fifteen minutes to pay before “evidence” goes viral. Clearing one’s name takes months; reputational damage is immediate.

Deep-fake market crash – A hyper-realistic live-stream shows the Fed chair announcing an emergency 200-basis-point hike. High-frequency trading algorithms dump Treasuries, erasing $2 trillion in minutes before the video is exposed as fake.

Synthetic-voice swatting – An AI system clones the exact cadence of a local teenager and phones 911, sobbing about a “hostage situation” at her address. Police dispatch a tactical unit; the confused family’s panic leads to a lethal misfire.

We see these companies in the crosshairs an AI regulation ban, with both short-term upside, a boost in market cap and yet some residual land-mines that could result.

Global context helps calibrate the stakes. The European Union’s AI Act, approved in March, imposes product-safety duties—conformity assessments, model cards, recall powers—on every system that reaches its market; China’s rules mandate real-name registration, dataset vetting, and prompt takedowns of “illegal content.” A U.S. moratorium would invert that trend: rather than tightening, it would suspend sub-national oversight and leave any future guardrails to a Congress that has not passed major tech legislation since the Obama era. Multinationals would face a split-screen: rigorous ex-ante licensing abroad, ex-post tort exposure at home. The compliance delta, not the algorithm, could become the competitive edge.

The insurance market is already pricing the gap. Cyber carriers report loss ratios above one-hundred percent on policies that bundle AI incidents with conventional data breaches. Lloyd’s underwriters now treat autonomous-vehicle fleets and foundation-model APIs as “emerging perils” subject to hard sub-limits or outright exclusions. If state regulators exit the field, primary insurers will demand broader representations and warranties, shift more risk to captives, and push premiums toward the high-hazard bands normally reserved for offshore drilling and commercial aviation. Capital-intensive start-ups may find that the cost of insuring latent algorithmic defects eclipses any savings from preemption.

Political risk completes the triangle. In Malwarebytes v. Enigma (2020), Twitter v. Taamneh (2023) and Biden v. Knight First Amendment Institute (2021), several Justices—especially Justice Clarence Thomas—have written separate opinions or statements suggesting that the Court is open to re-examining the scope of Section 230 (the law that now immunizes platforms from most state-law liability for user content).. Should an AI regulatory vacuum coincide with a headline disaster—a fatal robo-taxi crash on I-95 or a viral deepfake that moves a presidential poll—the Court could confront dueling petitions: companies invoking federal preemption, states invoking traditional police powers. The resulting jurisprudence might write federalism back in through litigation rather than statute, producing the very patchwork the moratorium is meant to prevent. Investors, insurers, and engineers would then navigate rules drafted not in one Capitol but in fifty courthouses.

Congress could avert an all-or-nothing clash between national uniformity and state experimentation by pairing a federal floor with cooperative federalism. States would be the legal laboratories to probe appropriate boundaries for AI laws. Washington would set baseline duties—accident reporting, safety cases for high-risk models, algorithmic-audit trails—then permit states to layer stricter requirements so long as they do not undercut the federal minimum. Reciprocal recognition of certified systems would prevent redundant testing, while a five-year sunset and GAO scorecard would force evidence-based renewal or revision. The result preserves innovation’s scale economies yet honors the states’ historic role as early-warning sensors for emerging risks.

Tweet O’ The Week

Epistrophy In The News

Epistrophy In The News

On Schwab Network, we walked through Broadcom’s earnings with Alex Coffey—focusing on the company’s surprising software performance and its dominant position in AI infrastructure. The key stat: custom AI accelerators accounted for more than 25% of Broadcom’s semiconductor revenue, a sharp jump in just two quarters. We also discussed how Broadcom’s networking hardware plays a hidden but essential role in hyperscaler data centers. You can watch the segment here.

Availability This Week

I’m in San Francisco all week. Happy to meet, walk or talk (we have two new people starting this month so I might actually be able to get away from the desk for a change!).

Written reports are available to clients (and friends who ask nicely), with video summaries on YouTube, and of course our popular summaries of the summaries on Instagram, TikTok, and YouTube Shorts.

📆 of Epistrophy Events

Availability This Week

I’m available all week so email or even text if you don’t hear back right away. I’d love to expand on the thoughts I’ve share, and I’ll be right on top of all the earnings reports outlined above.  Written reports are available to clients, with video summaries on YouTube and, of course our popular summaries of the summaries (yes, the second derivative) on Instagram and Tiktok.

I hope these notes are helpful to you. I’d love to discuss them further and, as always, comments, questions and ideas are appreciated. If you have a friend or even a frenemy whom you think might benefit from this note, have them reach out and I’ll put them on the list.