They Turned Down Bezos's $2 Billion. Then Built a Bigger AI.

The next Apple may already exist. Insider sources say Elon has spent two years building a secret device inside Tesla's facilities — one he claims will be "10x bigger than the largest product in history."

There's just one problem: the company is private, and unless you know Elon personally, you can't buy a single share. That was true until Guardian's research team found three public ticker symbols sitting in the launch supply chain.

Click here to see all 3 tickers, free of charge.

You won't hear these names on CNBC — Wall Street hasn't published a word on the connection. But when the launch hits September 21, that quiet ends.

Some are already calling this the biggest opportunity since AI. For anyone who missed Apple before the iPhone, this may be a second look at that kind of setup.

Get the 3 ticker symbols here.

Unlock the free briefing now.

1.  Two researchers reject a Bezos-backed fortune, then unveil a physics AI that dwarfs every language model.

Caltech professor Anima Anandkumar and co-founder Benedikt Jenik launched Accelerated Understanding this week with a model that handled 5 trillion pieces of data in a single prompt, roughly 5 million times what flagship models from Anthropic or Google typically take in. The pair had turned down an offer to lead Jeff Bezos-backed Project Prometheus that included a 35% stake and more than $2 billion in committed financing. Their model skips the Transformer architecture entirely and predicts physical phenomena instead of words. Read the full Reuters exclusive.

2.  Nvidia posts a $96.2 billion quarter, up 106% in a year.

The chip giant beat Wall Street on both revenue and earnings, with its data center business alone bringing in roughly $89 billion. CEO Jensen Huang told investors that AI has hit its inflection point and that compute is now revenue, and shares climbed more than 5% in after-hours trading. See the full numbers at Fortune.

3.  Salesforce and Anthropic launch Claudeforce, putting the world's biggest CRM inside Claude.

The expanded partnership ships a Salesforce plugin for Claude with 37 prebuilt sales skills, covering meeting prep, deal health checks, and pipeline updates, and makes Claude a default reasoning model across Salesforce's agent products. Salesforce shares jumped 12% in extended trading as the news landed alongside earnings that beat expectations. CNBC has the details, or read the official announcement.

Bonus read:  The story behind the physics duo, including the dinner meeting that started it all. The Japan Times carries the full feature.

The defection.  When Anandkumar and Jenik walked away from Project Prometheus, they were betting that the most valuable AI of the next decade will not chat. It will simulate. Their neural operator model is built to predict how heat moves through a chip, how air moves through a turbine, and how storms move through an atmosphere. Reuters tells the full story.

The crowded frontier.  They are not alone. Startups overseen by Yann LeCun and Fei-Fei Li are chasing world models, systems meant to understand spatial reality better than anything trained on text. Prometheus itself raised a $12 billion Series B in June to automate the manufacturing of complex physical systems. The shared premise: language alone caps what machines can understand.

The money.  Investors have noticed. Physical AI startups raised $47.4 billion globally in the first half of 2026, nearly four times the previous six months, according to Crunchbase data. London-based PhysicsX, founded by two former Formula 1 engineers, raised $300M at a $2.4 billion valuation to replace engineering simulations that take days with AI that answers in seconds.

Why size matters here.  In physical simulation, input size is not a vanity metric. Chip design, power grids, and weather all involve enormous numbers of interacting variables. A model that cannot hold them all in working memory has to approximate, and approximations compound into errors. That is why a 5 trillion data point prompt is the headline, not the model name.

The reframe.  For three years the AI race has been scored on how well machines write. The next phase may be scored on how well they predict the physical world, and the winners will sell to factories, utilities, and chipmakers rather than to you. If your industry touches anything physical, this is the corner of AI to start watching now.

Save our voices.  More than 80 UK actors, including Nicola Coughlan, Hugh Bonneville, and Matt Lucas, signed an open letter demanding a legal right to own your own voice as AI cloning spreads. ITV News has the campaign details.

Loophole, closed.  Meta is patching its smart glasses so the camera shuts off if you cover the recording light mid-video, killing the most shared privacy bypass trick online. Engadget explains the fix.

The canary count.  Stanford's updated jobs study finds employment for workers aged 22 to 25 in AI-exposed roles now sits 19% below their less-exposed peers, up from 13% a year ago. Read the researchers' summary.

Headless software.  The spiciest take on Claudeforce: enterprise software may stop having screens at all, with sellers working entirely from inside Claude. VentureBeat makes the case.

Stress-test any decision like a physicist

This week's lead story is about AI that thinks in physics instead of words, so here is a prompt that makes your AI assistant do the same for your decisions. Paste it into Claude, ChatGPT, or Gemini and it will strip a hiring call, a pricing change, or a big purchase down to the few variables that actually matter. It takes under a minute and works surprisingly well on decisions you have been circling for weeks.

Swap [DESCRIBE YOUR DECISION] with something specific, like "raising my freelance rates by 20% in October", "switching our team from monthly to annual billing", or "hiring a second support rep before the holiday rush". For sharper output, add one line of real numbers, such as current revenue or budget, right after the placeholder.