The Apple Tax Is Dead, and With It the Thesis That Made Apple a $3T Company
On 5 June 2026 the Ninth Circuit Court of Appeals struck down Apple's restructured commission on external payment links — the 27% intellectual-property levy plus 3% payment-processing fee that Cupertino adopted in 2021 to preserve the substance of its original 30% App Store commission, after losing the underlying Epic Games v. Apple injunction. The court's reasoning, as reported on 5 June 2026, was that the fees could not be tied to the actual cost Apple incurred in facilitating external links. Tim Sweeney, the chief executive of Epic, declared on the same day that "the Apple tax is dead in the United States." That claim is half right. The tax is wounded. The thesis that justified the multiple is now in open litigation.
For a company whose stock trades at roughly 37x trailing earnings, the ruling is more than a regulatory nuisance. It strikes at the narrative scaffolding that, between 2011 and the present, re-rated Apple's price-to-earnings multiple from 9.7x to 37x. That scaffolding is the services-aggregation thesis, and the man most responsible for laying it down was Luca Maestri, then Apple's chief financial officer. The same week the court ruled, OpenAI abandoned its six-month vesting cliff, Anthropic placed a $21B order for Broadcom's Ironwood TPUs, the Office of Personnel Management announced a 1,000-engineer federal placement programme, and the chief executives of Snap and Rivian used a long-form interview to defend hardware bets that, in different ways, also depend on the assumption that scale pays. The Apple story is the spine. The rest is the muscle around it.
The Maestri pivot, ten years on
In fiscal 2015, Maestri told investors that services tied to Apple's install base represented $31B in purchase value, up 23% year-on-year. The figure did the work of an entire strategy deck. It implied that a billion-strong iPhone fleet could be monetised the way Visa monetises transactions: a small, predictable cut of every digital dollar that crossed Apple's rails. The market extrapolated, and the multiple expanded accordingly. By the time Epic v. Apple reached trial in 2020, Apple was no longer being valued as a hardware company with a side business in services. It was being valued as a platform with a side business in hardware. As one observer put it on 5 June 2026, summarising the aggregation logic: "the value of the earnings is so much higher… because they've developed this services monopoly, essentially. It's the Ben Thompson aggregation. It's a toll road for your life."
Thompson himself, in 2018, was the early sceptic. The aggregation model, he said then, "seems incredibly worrisome to me any time a company that predicates its growth story on rent seeking… the pursuit is corrosive on whatever it was that made the company great in the first place." Ten years on, with iPhone unit growth flat and services carrying the growth story, the rent-seeking critique has migrated from the analyst note to the appellate record.
What the Ninth Circuit actually cracked
The appellate ruling, covered in detail on 5 June 2026, addressed the post-injunction fee structure Apple adopted after Judge Yvonne Gonzalez Rogers ordered the company to allow developers to direct users to external payment methods. Apple's response was to keep its commission intact in substance by splitting the original 30% rate into a 27% "IP licensing" fee on external transactions and a 3% "payment processing" fee. Epic's Pepper lawsuit argued the structure was a sham. The Ninth Circuit agreed, in the language of the decision, that the fees could not be justified by reference to the actual costs Apple incurred in facilitating external links.
The immediate consequence is that Apple's commission on external links will fall. The structural consequence is harder. It forces a re-examination of the 30% rate Apple charges on in-app purchases, a rate that has never been independently justified on a cost basis. If the same standard is applied to that rate, the services thesis as currently understood stops working. The aggregation rent is no longer assumed. It must be earned.
Meanwhile, the AI talent market is being repriced
While Apple absorbed the ruling, OpenAI made a quieter move that reveals how the labour market at the frontier of artificial intelligence is being repriced. The company's apps chief, Fidji Simo, announced on TBPN the prior week that OpenAI had ended its policy requiring employees to work six months before their equity vests. xAI had shortened its own cliff from twelve to six months earlier. Meta, Google, and Anthropic have reportedly put packages above $100M on the table for senior researchers.
Connor Sen, writing in Bloomberg Opinion, has noted that OpenAI is on track to spend $6B on stock-based compensation in the current year — roughly half of projected revenue. That figure is striking only to those who have not been paying attention. A frontier AI lab whose market capitalisation is now reported at $500B can amortise the cost of its workforce across a much larger base than its annual revenue. The market is paying for optionality, not current earnings.
The deeper signal in the no-cliff move is structural. A researcher who leaves Anthropic on Friday and joins OpenAI on Monday is fully vested from day one. As one observer noted on 5 June 2026: "the fact that [AI researchers] are adding value on day one — basically, you just left Anthropic, we're gonna plug your brain — that seems like a win for the AI researcher." It is also a permanent restructuring of equity compensation in the industry. The cliff was a retention tool. Its disappearance is an admission that retention is no longer the binding constraint. OpenAI is raising $40B in the private market — more than the $30B SpaceX is targeting in its IPO bank pitches. Private capital has become the new Aramco-scale pool for the top AI companies, and the labour market is adjusting accordingly.
Anthropic's $21B bet, and what it says about Claude's next eighteen months
If the AI talent market is being repriced on the demand side, the supply side is signalling equally loudly. Broadcom's most recent earnings call disclosed that Anthropic has placed cumulative orders worth $21B for the company's custom TPUs — $10B in the third quarter for Ironwood racks, and a further $11B in the current quarter for delivery in late 2026. Anthropic is separately reported to be in talks with Google for tens of billions of dollars more in cloud capacity.
For context, $21B is roughly the gross domestic product of Jordan, committed to a single supplier relationship over a defined delivery window. The implication is that Anthropic intends to train and serve models at a scale that requires its own silicon, not rented capacity. It also implies a great deal of confidence in the trajectory of Claude's commercial demand into 2027. Compute orders of this magnitude are not placed by companies worried about a soft landing. They are placed by companies that have decided, for better or worse, that the rent at the application layer will be settled by who controls the most flops.
Hardware bets from the second tier: Snap and Rivian
Two of the more interesting hardware stories of the past week came from companies that are not, by conventional measures, on the frontier of the AI capex cycle. Evan Spiegel, the chief executive of Snap, used his appearance on 5 June 2026 to make the case for augmented reality as a new computing platform. Snap has spent more than $3B on AR glasses over eleven years, and the company's first consumer version ships this year. The company has nearly a billion monthly users globally and is one of the largest messaging platforms in the United States — larger than WhatsApp domestically. Spiegel drew the line at virtual reality: "We've invested zero dollars into VR. We essentially think that it takes people out of the real world… and so our bet has been exclusively on augmented reality." He added that for AR glasses to displace the phone, the experience needs to be "10 times better than what's possible on your phone." Snap, with its existing distribution, has the audience to find out.
RJ Scaringe, Rivian's chief executive, drew a similar boundary on the software-defined vehicle. Rivian, he said, holds roughly 35% of the US market for electric vehicles priced above $70,000, and in California is the best-selling premium SUV regardless of powertrain. The company has signed a $5.8B software-licensing deal with the Volkswagen Group, and has built an in-house self-driving inference processor with 800 TOPS of compute, 35 billion transistors, and the ability to process 5 billion pixels per second. The current generation runs at roughly 200 TOPS on NVIDIA silicon. Scaringe was blunt about the competitive landscape: "There are only two companies in the West that have [a true software-defined architecture]: Rivian and Tesla. And then in China there's more than five, less than ten." The R2 SUV ships later this year at a $45,000 starting price, roughly half the price of the R1S. "If you want a Model Y, buy the Model Y, not some other company's version of it," he said — a rare piece of competitor endorsement in a category built on differentiation.
The state, finally, decides to hire
In Washington, the Office of Personnel Management announced a programme that, on its face, looks modest: 1,000 engineers, product managers, data scientists, and AI specialists placed across federal agencies for two-year tours, with 25 industry partners including OpenAI, NVIDIA, xAI, Coinbase, Databricks, and Snowflake. The head of OPM, Scott Kupor, framed the gap the programme is meant to close: only 7% of the federal workforce is early-career, versus 25-30% in the technology industry. The mismatch, in his telling, is structural. "Drop into the blob" is the phrase he used to describe what happens to a talented technologist who joins a federal agency unprepared to deploy them.
A thousand engineers will not close a workforce gap measured in the hundreds of thousands. But the symbolism matters. The same week that a court reined in the largest platform rentier in the US economy, the federal government was trying to re-enter the talent market on the same terms as the private sector. The state, after a generation of standing aside, is reaching for the levers that the Maestri generation used to re-rate Apple.
The lesson travels
The platform thesis, the AI capex thesis, and the state-capacity thesis all have one thing in common: they are bets on aggregation. They assume that the player who controls the most users, the most compute, or the most institutional reach will collect the rent. The Ninth Circuit's Apple ruling suggests that, in the US legal environment, the rent is no longer assumed. It must be defended on its merits, in court, against cost-based scrutiny. The lesson will not be lost on the AI labs that have committed $21B to a single chip supplier, nor on the federal agencies now trying to hire the engineers those labs are paying nine figures to retain. Aggregation is still the game. But the era of collecting it without an argument is closing.
Wire provenance
This editorial synthesis draws on the following public wire/social posts:
- https://www.youtube.com/watch?v=eAhTvDjDyhk