Yuval Avidani
Author
Elon Musk calls him "Scam Altman," and Apple just filed a 41-page lawsuit against his company. That's the kind of week Sam Altman, the CEO of OpenAI, just had. On July 10, 2026, Apple filed a federal lawsuit against OpenAI claiming it stole trade secrets to build a hardware device, and a few days later Musk celebrated it on X with the nickname he himself coined for Altman. But behind the noise and the insults hides a much bigger story: a war worth hundreds of billions of dollars over who controls artificial intelligence. Let's break it all down - the drama, and also the real numbers of who's actually leading.
The week the AI war left the labs
Turns out this whole story crammed itself into just ten days. In under two weeks we saw a new voice model launch, Meta's first paid model, a physical keyboard, a massive lawsuit, the cancellation of a controversial feature, and a war of words between the two richest people on earth. The fun part is that all of this happened at almost the same time, and together they paint one picture: the AI race is no longer being fought just over model quality - it's being fought in courtrooms, in ad units, and in hardware. Here's the whole week, day by day.
The lawsuit: Apple vs. OpenAI
Let's start with a fact that's important to get right, because a lot of headlines missed it. This is not an antitrust lawsuit, and it's also not an "insider information" case in the stock-trading sense. This is a trade secret theft lawsuit. A trade secret is confidential proprietary information - product plans, manufacturing methods, technical specs - that gives a company a competitive edge as long as it stays secret. The moment it leaks to a competitor, the damage is done.
The suit was filed on July 10, 2026 in the Federal District Court for the Northern District of California, titled Apple Inc. v. Liu et al., on grounds of trade secret theft (Defend Trade Secrets Act) and breach of contract. The defendants are OpenAI itself, hardware company io Products (the startup of legendary designer Jony Ive, which OpenAI acquired for about $6.5 billion in 2025), plus two individuals: Tang Tan, OpenAI's chief hardware officer who previously served as a VP at Apple, and Chang Liu, a senior electrical engineer who worked at Apple for eight years before moving to OpenAI in 2026.
Apple's claim is that the theft happened "at every level, from team engineers to the chief hardware officer." According to the complaint, over 400 former Apple employees now work at OpenAI, and some of them, Apple alleges, took secrets with them. The details read like a spy movie: per the lawsuit, Liu exploited a security bug to access Apple's network storage and wrote to himself "LOL, found out I can access this, so funny," never returned his company laptop, and said "I have another computer." Tang Tan, meanwhile, allegedly asked job interview candidates to bring "actual parts" from Apple products with them, and coached departing employees on how to dodge the "immediate walkout" Apple gives anyone who resigns.
Important to say: these are allegations, not proof. OpenAI denies all of it, stating on July 14 that it is "not aware of any evidence that this complaint has merit," and that it has "no interest in other companies' trade secrets." Think of it like a divorce fight between two companies that used to be partners - just two years ago, Apple and OpenAI integrated ChatGPT right into the iPhone. Now Apple is building its own AI, OpenAI is building its own hardware, and suddenly each one sees the other as a competitor poaching its people.
"Scam Altman": The Musk-Altman war
Enter Elon Musk, and here again we need precision. Musk isn't a party to Apple's lawsuit - he's just watching from the sidelines and cheering. But the nickname "Scam Altman" (a play on the name Sam and the word scam) is entirely his: he coined it back in February 2025, after Altman and OpenAI's board rejected Musk's $97.4 billion buyout offer. Now, with Apple's lawsuit, Musk has gone back to the nickname full force: "Scam Altman strikes again," "he's taking the scam to a new level," and even "after stealing the open-source AI nonprofit, you stole all of Apple's phone technology too."
Altman, true to form, didn't let it slide. He fired back at Musk: "There are a lot of benchmarks showing 5.6 sol is the best model in the world right now, but the most reliable way to know it is that Elon is obsessed with me again" (5.6 sol refers to GPT-5.6 Sol, OpenAI's flagship model). In another jab, he hinted that Musk is the one selling investors dreams about data centers in space.
Behind the faces, there's a fact that's easy to miss: Musk, who co-founded OpenAI back in 2015 and then left, has already lost two of his own lawsuits against the company this year. His personal lawsuit was dismissed in May 2026 (an advisory jury ruled it was time-barred), and a trade secret lawsuit filed by his company xAI against OpenAI was dismissed "with prejudice" (meaning it can't be refiled) in June 2026. The only lawsuit Musk still has alive is an antitrust suit by xAI and X against Apple and OpenAI jointly, currently in the document discovery phase. So when Musk celebrates Apple's lawsuit, he's really riding on someone else's win, after losing his own battles.
OpenAI is chasing revenue
So here's the question: why all this pressure? The answer is simple, and it's money. OpenAI posted an operating loss of about $21 billion in 2025, and it's now trying almost every possible way to grow revenue. Look at what happened in just the last six months:
- A cheap $8 subscription. In January 2026 OpenAI launched ChatGPT Go globally at $8/month (below the $20 Plus tier), with 10x more messages and uploads than the free version. The goal is clear: pull people who won't pay $20 into the paying network.
- Ads in the chat. Starting February 2026, OpenAI is testing "sponsored" boxes at the bottom of answers in the US, but only on the free and Go tiers. To be fair: the company frames this as "accessibility" rather than an escape from losses, and its head of apps said advertising will remain "a minority of revenue, probably forever."
- Adult content - that never launched. Here's an important correction to a common mistake. Altman did announce in October 2025 that by December OpenAI would allow "even more, like erotica for verified adults." But this never actually shipped, and was shelved indefinitely in March 2026. What did launch is an age-prediction system that identifies minors and increases protections for them - a safety filter, not an unlocking of content.
- A physical keyboard for Codex. On July 15, OpenAI launched Codex Micro, a $230 desktop keyboard for controlling its coding agents (Codex is the agent that writes and runs code for us). It has 13 mechanical switches, a dial for setting how much it should "think," and keys that light up in color based on the agent's status (blue = thinking, green = done, red = error). The irony: it's launching a hardware product exactly while being sued over hardware secrets.
- A new voice model. On July 8, OpenAI launched GPT-Live, a family of voice models that listens and talks simultaneously (full-duplex, meaning you can interrupt it mid-sentence like in a real conversation), translates in real time, and hands off hard questions to GPT-5.5 running in the background.
To me, this whole list tells one story: a company burning cash at a massive rate is trying to turn attention into money from every possible direction, and fast.
And with the competitors too: voice, Hebrew, and a Meta misstep
While OpenAI was making headlines, the competitors weren't resting either. And here there are two clarifications worth making, because I've read a lot of inaccuracies about these too.
The first concerns SpaceX's Grok. After Musk's xAI merged into SpaceX (a roughly $1.25 trillion merger, the largest private merger in history), the company got a new public brand: SpaceXAI, as of July 6. They have a great voice model, but contrary to what's being claimed, it's not "new this week" and it's not actually "faster" than OpenAI's. The model, grok-voice-think-fast, actually launched back in April 2026 as an xAI product. It genuinely is the world's first-place finisher on an independent benchmark for complex voice tasks (tau-Voice), but on independent measurement of response speed it's actually a bit slower than OpenAI's voice model (about 1.25 seconds to first sound vs. about 1.14). Grok's real edge is task accuracy and price, not speed.
The second concerns Meta, and here there are two pieces of news that directly touch us here in Israel. The good one: on June 30, 2026, Meta launched Meta AI in Israel with Hebrew support, right inside WhatsApp and the app, powered by its Muse Spark model (not Llama, as was mistakenly reported). The less good one: on July 7, Meta launched its image generator Muse Image, along with an Instagram feature that let anyone tag a public account and have Meta AI pull that account's public photos to generate new images from them - without asking, with every public account automatically opted in. After public backlash, including opposition from the actors' union SAG-AFTRA (about 160,000 members), Meta pulled the feature within three days and said it "missed the mark." But notice the detail: only the tagging feature was removed, not the image generator itself. The critics' demand is clear: consent should be opt-in, meaning we approve in advance, not something we have to cancel after the fact.
The real ranking: who's leading the AI race
Now for the part that actually matters, the part where we talk data. You asked to see who's leading, so let's put only verified numbers on the table - users, paying subscribers, revenue and profit. One warning before we start: these are different currencies, and you can't just add them up. Companies like Meta and Google also count "embedded" usage (someone bumping into AI inside search or WhatsApp), while ChatGPT and Claude are assistants people go to on purpose. So let's compare carefully.
Let's start with monthly users. Here Google, Meta and OpenAI are in a league of their own, each around a billion users, while Claude and Grok are far smaller.
Now for paying subscribers, and this is the number that separates the giants. OpenAI clearly has the most paying users: over 50 million individual subscribers plus over 9 million business users. Grok, via X, also reports about 6.3 million subscribers. Meta, on the other hand, doesn't charge anything for Meta AI (it's completely free), Google doesn't break out how many of its 350 million subscribers belong to Gemini, and Anthropic simply doesn't publish consumer numbers. So in the arena of "who convinces the most people to pay for a chatbot," OpenAI wins.
And here's where the surprise comes in. If you look at company valuation and profitability, it's actually Anthropic, the maker of Claude, that jumped to the top. Per its official announcement from May 2026, its annual run rate (ARR, meaning monthly revenue times 12) crossed $47 billion, and it raised at a $965 billion valuation - higher than OpenAI's $852 billion. And more importantly: according to a CNBC report, Anthropic was the first among the AI labs to post a quarterly operating profit (about $559 million), though the company itself warns this could reverse.
On profit, bottom line: almost everyone is losing a ton of money. OpenAI's operating loss was about $21 billion in 2025, and xAI's was about $6.4 billion. This race is running on massive fundraising, not on profits, and that's exactly why everyone is chasing revenue so hard. Here's the full scorecard, company against company, so you can see the whole picture in one place.
My take
In my view, the "Scam Altman" drama is great entertainment, but it's also smoke hiding the real fire. And the fire is this: the AI race has become a war worth trillions of dollars, and it's no longer being fought only over who builds the smartest model - it's being fought over people (who's stealing whom), over money (who finds revenue before the cash runs out), and over hardware (who builds the next device). And when you look at the rankings, there's no single "winner": OpenAI has the most users and payers, Anthropic has the highest valuation and the first profit, Google and Meta have the widest and cheapest reach, and Grok remains a well-funded player but still trailing behind.
I also need to be honest about the limitations: some of these numbers are annualized run-rates (an estimate extrapolated from a single month) rather than audited annual revenue, some were measured on different dates, and a private funding valuation is not the same as profit. And of course, none of this is investment advice or financial guidance - it's a factual snapshot, not advice on what to do with your money.
What is clear: while the giants are fighting it out in courtrooms and on X, the tools in our hands keep getting better. Meta AI already speaks Hebrew, OpenAI already talks to us in a live voice, and even the feature that tried to scrape our photos without permission got pulled within three days. So now that we've broken it all down: who do you think is really leading this race - the one with the most users, the one worth the most, or the one who starts turning a profit first?