The $1.77 trillion question: is this the most important company on earth?

Yesterday in AI | Friday, June 5, 2026

The $1.77 trillion question: is this the most important company on earth?

The biggest IPO in history just set its price, and trading starts next week. Researchers found a way to turn your WhatsApp notifications into a weapon against your own AI assistant, no clicks required. Two image labs went head-to-head Wednesday and flipped the leaderboard. Martin Scorsese, at 83, just publicly backed an AI tool, and what he's using it for is not what you'd expect. Plus, DeepSeek is actually closing its funding round, and the numbers confirm the pricing pressure on Western labs is here to stay.

Feedback? Email mike@yesterdayinai.news or connect on LinkedIn, X, Bluesky, or Substack. If you like the show, please take a minute to rate and review it so others can find it!

Links:
Website: https://www.yesterdayinai.news/
LinkedIn: https://www.linkedin.com/in/mrmichaelrobinson/
X: https://x.com/Yesterday_In_AI
Bluesky: https://bsky.app/profile/yesterdayinai.bsky.social
Substack: https://yesterdayinaipodcast.substack.com/

Transcript

Copy Transcript Copied

SPEAKER_00 0:05

Hi folks, this is Yesterday in AI, your daily digest of everything happening in the world of AI in 10 minutes or less. I'm Mike Robinson. It's Friday, June 5th, and Elon Musk's rocket company just set the price on the biggest IPO in history. Someone figured out how to weaponize a WhatsApp message against your AI assistant, and two image labs went head to head for the top spot on the leaderboard. Let's get into it. We covered the SpaceX S1 filing two weeks ago, so the setup isn't new. What's new is the price. SpaceX set it at $135 a share on Wednesday, confirming a $1.77 trillion valuation and a $74.4 billion raise. If those hold, it's the largest public offering in history, roughly three times Saudi Aramco. Shares start trading on Nasdaq next Thursday, June 12th, under the ticker SPCX. The question on the table, is the bundle of rockets, satellites, and AI tooling worth $1.77 trillion? The market answers next week. But before we move too far away from the news cycle, here is something more personal that dropped this week. Security researchers at Safebreach Labs found a way to hijack Google Gemini, and every user should hear how simple the attack actually is. Gemini is Google's AI assistant, built into Android phones, and the attack is simpler than you'd expect. Someone sends you a WhatsApp message. You don't click anything, you don't type any commands. Gemini reads your incoming notification, sees hidden instructions embedded in the message, and quietly follows them. The technique has a name, Fake Context Alignment. It makes malicious instructions look like a natural part of your ongoing conversation with Gemini, bypassing Google's existing defenses. The attack doesn't stop at WhatsApp either. Slack, Signal, SMS, Instagram, Messenger. Any notification Gemini reads can carry hidden attack instructions. Five threat categories were demonstrated Data Theft, Unauthorized Actions, Phishing Relay, Account Takeover Prep, and Silent Surveillance. This is Safebreach's second time breaking Gemini in this way. Their previous research used Google Calendar invites. They disclosed to Google before publishing, so Google is aware and working on it. But there's no clean patch that fully solves this at the model level, because the problem is structural. The model reads text. Attackers can write to that text. The more capable your AI assistant gets at reading your context, the more attack services that creates. And the people most exposed are the ones getting the most out of Gemini, power users with deep app integrations. The exact audience this kind of attack is designed to hit. Worth knowing. Sticking with AI assistants for a moment, ChatGPT crossed 1 billion monthly active users in May, according to Sensor Tower. That makes ChatGPT the fastest app in history to reach 1 billion users, faster than TikTok, Instagram, YouTube, and Google Maps. OpenAI launched in late 2022, so that's roughly three years from launch to a billion users. TikTok took five years. Instagram took six. The telephone took 75 years to reach a billion users. There's a real debate about what monthly active users means for an AI app versus a social network. But as a directional signal of how fast AI tools embedded themselves into daily life, this one lands. OpenAI's stated goal is to become the platform every employee routes their workday through. A billion monthly users suggest they're closer to that than most people assumed. And it all happened while OpenAI was simultaneously managing a trial, an IPO filing, and a robotics division launch. Not a slow spring. Now over to Meta, who had two very different stories running in parallel this week, and together they say something about where the company actually is. On the distribution side, Meta launched its WhatsApp business AI agent globally. Businesses can now deploy this agent on WhatsApp to handle customer queries, recommend products, book appointments, qualify leads, and escalate to a human when needed. Meta spent two years testing it in India and Mexico before rolling it out worldwide. It connects to Shopify, Zendesk, and other enterprise tools, and it's landing in Instagram DMs as well. Large companies pay per token. It's bundled into WhatsApp business premium for smaller ones. Meta has 3 billion WhatsApp users. Putting an AI agent in front of all those business conversations is a serious distribution play. On the model side, their most ambitious AI model, Muse Spark, has no developer release date. Meta said it'd release it to developers this month. It hasn't. Outside evaluations haven't happened. The model is rated by Meta's own internal assessments as competitive with Claude and GPT 5.5, which makes the delay more notable. Meta is spending $115 to $145 billion on AI infrastructure in 2026. The distribution side is live. The model they're spending that money to build is running late. While Meta is sorting out their model timelines, Congress is working on something that would affect all of them. House members Jay Obernolte and Lori Trehan released a discussion draft this week that would block states from regulating AI model development for three years. The pitch is a single federal standard instead of the current state-by-state patchwork. It explicitly leaves room for states to regulate what happens after deployment. Consumer impacts, downstream harms, that kind of thing. But during development, states would be frozen out. Critics are already pointing out that the draft has no child protection provisions. That's a real gap. Most of the state AI laws that actually passed were responding to specific documented harms, not just creating compliance burdens for labs. This is a discussion draft, not a bill yet, but it signals where Congress is leaning. If it becomes law, Frontier Labs get a three-year window to develop models without worrying about 50 different compliance regimes. The people who've been doing the state-level legislative work on safety so far lose their main lever. Stepping away from policy, two image generation labs shipped competing models Wednesday and flipped the leaderboard. Reeve 2.0 topped the Arena AI Leaderboard, an independent site that ranks image models head-to-head based on user votes, taking the second highest overall spot and bumping Nano Banana 2 down. What makes Reeve different is the interface, a code-based layout that lets you fine-tune specific elements of a generated image with unusual precision. At the same time, Ideagram 4.0 launched and took the top spot among open weight image models, meaning models that are freely available for anyone to download and run on their own hardware. Both dropped on the same day, which wasn't a coincidence. The image generation rankings move fast, neither of these is a household name yet, and they're both beating the incumbents on quality benchmarks right now. And staying in the AI image space for one more. Martin Scorsese publicly backed an AI image generation tool called Flux2 from Black Forest Labs and joined as an advisor. He's 83. Six decades of filmmaking. Goodfellas, Raging Bull, The Departed. The way he described his use of it is what's worth paying attention to. He's using Flux 2 as a visualization engine during pre-production. He prompts it, and within seconds, his cinematographer and production designer can see exactly what he means. There's always been this problem of how do you communicate what you see in your head to your cast and crew, he said. He's treating it as a compression tool for creative communication, not a replacement for any part of the filmmaking craft. When someone with that much standing in the field says AI is a useful tool for the work he cares about, the cultural resistance argument gets harder to sustain. Finally, Deep Seek. We covered the early fundraising discussions back in May. Now the round is actually closing. About 50 billion yuan, or about $7 billion US, from fewer than 10 investors. DeepSeek's founder is personally putting in 20 billion yuan. The round closes within weeks and there's no IPO in the picture. Why does the capital raise matter? Because DeepSeek V4 Pro charges $4 per million tokens while Claude runs around $25 and GPT-5.5 around $30. They made that discount permanent about two weeks ago. So they're taking in $7 billion while pricing at roughly one-seventh the cost of the major Western AI companies. The intellectual property concerns and geopolitical risk of building on Chinese infrastructure are real constraints that stop many Western enterprises from touching it. But the pricing pressure is structural regardless. The major U.S. labs built their business models on premium pricing. DeepSeek is betting on volume at drastically lower cost, and this capital raise confirms that they can hold that position longer than anyone expected. Just a couple of more items. If you have any feedback about this show, you can email Mike at yesterday andai.news, or you can find me on LinkedIn, X or Blue Sky. And if you like this podcast, please take a minute to rate and review it so others can find it. Thanks. That's all for this edition of Yesterday and AI. Stay curious, and I'll see you tomorrow.