The Platform Tax Arrives

If You Only Read One Thing

The most important platform story today is not a product launch; it is rent extraction. California's proposed cloud and software tax and OpenAI's Apple fight are the same argument from opposite sides. In California Taxes the Stack and OpenAI Meets the Landlord, software becomes infrastructure, and whoever controls the choke point starts charging for access.

California Taxes the Stack

California's budget problem has turned into a software-pricing experiment. That is more revealing than another fight over "taxing tech."

Gov. Gavin Newsom's revised budget proposes a 7.25% tax on digital software, data processing, and cloud services used by businesses, with the administration estimating more than $1 billion of annual revenue, according to Business Insider. The state frames the change as a modernization of sales-tax treatment for a services economy, and the budget summary puts it inside a broader deficit package. Bloomberg also covered the proposal as part of California's attempt to close a budget hole, but the structural story is bigger than the fiscal line item.

Why it matters: the state is not really taxing "software" in the consumer-app sense. It is taxing the operating substrate of modern companies: cloud compute, SaaS subscriptions, data-processing services, and the back-office systems that make a firm legible to itself. That matters because B2B software has historically benefited from a strange political invisibility. It is mission-critical enough to have pricing power, but abstract enough to avoid the kind of explicit tax treatment that hits physical goods.

The familiar version is a sales tax on office equipment. The new version asks why a company should pay tax on a server rack but not the AWS bill that replaced it, or on boxed accounting software but not the cloud ledger that replaced the box. Once the state makes that substitution explicit, enterprise software loses some of its aura as a weightless productivity layer and starts looking like taxable infrastructure.

That creates a second-order pricing problem. Software vendors can absorb the tax, pass it through as a California line item, or rework contracting so the taxed service is separated from untaxed support, implementation, and consulting. The path they choose will reveal where pricing power actually sits. If customers accept the pass-through, the tax becomes a public-sector claim on SaaS gross margin. If vendors eat it, the proposal becomes another pressure point on a sector already dealing with AI-driven seat compression.

Room for disagreement: the strongest counterargument is that this is just a bargaining chip in a state budget fight, not a durable policy shift. California business groups will fight it, and the final budget could narrow the base, phase in the charge, or exempt smaller firms. The politics are also awkward: taxing software used by businesses can raise costs for companies that are not obviously "tech companies" at all.

What to watch: whether large SaaS and cloud vendors start disclosing California-specific tax pass-through language in contracts. That would show the tax is moving from Sacramento budget math into enterprise software pricing architecture.

OpenAI Meets the Landlord

OpenAI is learning the oldest platform lesson in the consumer internet: a partner with distribution is not the same thing as a channel you control.

Bloomberg reported that OpenAI is preparing legal action against Apple over the way Apple has handled its AI partnerships, and TechCrunch's account puts the dispute in the context of earlier Siri and ChatGPT integration promises. Apple announced ChatGPT integration in 2024, but Siri's broader AI rebuild has repeatedly slipped, and Apple has reportedly explored other partners while keeping the user-facing relationship inside its own operating system. Daring Fireball's skeptical read is useful because it asks the right legal question: what exactly did Apple promise that OpenAI can enforce?

Why it matters: OpenAI has model demand, brand recognition, and consumer usage. Apple has invocation rights. That is the more important asset for an AI assistant because the user does not wake up wanting "a model"; the user highlights text, asks Siri, opens a camera, replies to a message, or needs help inside an app. Whoever owns those moments can route the request to one model, many models, or no external model at all.

This is why the dispute is not just a bruised partnership. It is a distribution conflict over where AI value accrues. OpenAI wants Apple to be a high-volume consumer channel. Apple wants AI suppliers to be modular inputs beneath the Apple interface. If Apple succeeds, OpenAI becomes a supplier inside the operating system. If OpenAI succeeds, Apple's AI delay becomes a legal and reputational vulnerability that lets the model company claim the consumer relationship more directly.

There is a parallel to the old search distribution market, but with a twist. Google paid Apple for default placement because search queries had direct monetization. AI assistants do not yet have an equivalent consumer monetization engine, so the fight is happening before the revenue model is settled. That makes the legal threat as much a negotiating tactic as a lawsuit preview: OpenAI is trying to raise the cost of being treated as interchangeable.

Room for disagreement: Apple may simply be moving slowly because its AI stack is late, not because it is excluding OpenAI. The company can also argue that platform integration requires privacy, reliability, and product-control decisions that no model vendor has a right to dictate. The fact that ChatGPT remains a leading iOS app weakens the simple version of the "Apple is burying us" claim.

What to watch: whether OpenAI's next move is an antitrust-style complaint about Apple distribution or a narrower contract dispute. The first would signal a platform-war escalation; the second would signal hard bargaining over a failed integration.

The Contrarian Take

Everyone says: the OpenAI-Apple fight is about Apple falling behind in AI.

Here's why that's wrong, or at least incomplete: Apple being late is the visible problem; Apple still controlling invocation is the structural problem. OpenAI can have the better model and still lose the consumer economics if Apple owns the moments where users ask for help. California's tax proposal makes the same point from the public side: once software is embedded deeply enough in daily operations, the owner of a chokepoint starts charging rent. The question is not whether AI or cloud software is important. The question is who is allowed to meter the importance.

Under the Radar

Data-center politics has numbers now - Gallup found that 71% of Americans would oppose an AI data center in their area, including 48% who would strongly oppose it. Semafor's useful addition is that AI advocates still have time to change the story, but the baseline is brutal: local politics now treats compute as a land, water, power, and noise project, not a cloud abstraction. (Gallup)
The Apple fight has a remedies problem - The easiest headline is "OpenAI may sue Apple." The harder question, raised by Daring Fireball, is what remedy would make sense if Apple did not promise exclusivity, default placement, or a timeline it can be forced to honor. That matters because a weak legal hook would turn the threat into public negotiation rather than real platform litigation. (Daring Fireball)

Quick Takes

Anthropic turned export controls into a 2028 scenario. Anthropic published a policy paper warning that advanced AI systems could accelerate Chinese military and cyber capabilities by 2028 unless U.S. controls harden around chips, model weights, and distillation. The useful read is not that Anthropic has become hawkish; it is that frontier labs are writing geopolitical threat models that double as industrial-policy asks. (Source)
TanStack's blast radius reached OpenAI devices. OpenAI said two employee devices were affected through the TanStack supply-chain attack, while user data and production systems were not compromised. That is exactly the uncomfortable middle zone: trusted open-source build chains can fail below the level of a production breach but above the level of harmless maintainer drama. (Source)
The Supreme Court kept the mail-order drug channel open. The Court declined to disrupt mail access to mifepristone, preserving the distribution layer that telehealth providers and pharmacies now depend on. The structural point is that healthcare access fights increasingly turn on logistics, identity, and prescribing channels, not only on the underlying drug approval. (Source)

The Thread

The connective tissue is not "tech regulation." It is metering. California wants to meter cloud software because business software has become infrastructure. Apple wants to meter AI assistance because operating systems still own user intent. Data-center opponents want local communities to meter compute expansion through land and power politics. The 2020s software stack is losing its old abstraction privilege: when it becomes physical, fiscal, or behavioral infrastructure, someone else claims the right to price it.

Predictions

New predictions:

I predict: By 2026-06-30, California's final budget deal will preserve the digital-software tax concept but narrow the base or phase in implementation rather than passing the 7.25% proposal unchanged. (Confidence: medium; Check by: 2026-06-30)

Coming Next Week

Next week, we're going deep on the SaaS seat-compression problem: whether AI agents are really breaking per-seat pricing or merely forcing public software companies to admit the post-2020 hiring base was too large.

Generated 2026-05-15 03:21 EDT