AI's Dark Horse?
Issue 12: Trump's Autonomous Scientists // OpenAI's Struggles
Hello Futurists,
We haven’t always known what to make of Anthropic… while they’re certainly always part of the top AI model conversation, they’re rarely leading it.
But their quiet dominance in one niche field continues to be setting them up to win out over better-resourced competitors.
It’s all about AI coding.
Read on for Ejaaz’s bold take.
Also—catch up on this week’s biggest AI headlines:
Google Nears $4T as TPUs Go Stratospheric
Anthropic Ships the Heat
The Thinking Game
Trump Wants Government AI Agents
OpenAI Facing “Economic Headwinds”
Sponsor: Unichain — Faster swaps. Lower fees. Deeper liquidity. Explore Unichain on web and wallet.
Google Nears $4T as TPUs Go Stratospheric
There’s never been a better time to hold $GOOGL stock.
The Information reports that Meta is in talks for a multi-billion dollar deal to bring in Google’s specialized TPUs to train and enhance their AI models as supply constraints continue to hit Nvidia.
If confirmed, this would be the third major TPU deal after Anthropic’s 1 million TPU deal and their major agreement with NATO.
These partnerships further fuel the narrative that Google now presents the ultimate threat to Nvidia’s $5T market cap. $NVIDIA fell 5% on the news as $GOOGL soared to a new 52-week all-time high.
Anthropic Ships the Heat
This week Anthropic released Claude Opus 4.5, a brand new LLM that excels in coding. The model blew past analysts’ expectations achieving a record 80% on the SWE verified benchmark and the #1 spot on the ARC AGI test (notoriously the hardest test for an AI model).
The launch marks an important milestone in Anthropic’s journey to be the top AI lab. The young startup had been struggling to attain compute resources to train their model but this new release now firmly cements their position as the #1 provider of coding AI.
The release added fuel to the fire as they reportedly near closing a deal that values them at a staggering $400 billion.
Thanks for reading! Subscribe for free to support and plug in to Limitless.
The Thinking Game
As a certified Google fanboy I feel obligated to share this film that was released privately last year but just made available to the public this week. I’m not going to spoil anything but if you’re looking to feel informed and inspired by the people building this next paradigm that is AI I cannot recommend it enough. After you’ve caught up on Limitless, give this a shot and we can all chat about it on the show next week.
Trump Wants Government AI Agents
In a landmark partnership involving OpenAI, Google and Anthropic, the U.S. Government will allow select AI labs to access petabytes of proprietary national scientific data to train autonomous scientific AI agents that will advance the USA’s scientific discovery.
Using data from 17 national labs, the labs will engage in “closed loop” discovery where their models will analyze the data, create a novel scientific hypothesis and then test it in a fully autonomous robotic lab.
The system will apply across all scientific sectors including material and medical sciences. The partnership’s vision is to discover novel cures for diseases and create new materials for advanced physical infrastructure.
OpenAI Facing “Economic Headwinds”
In a leaked internal memo to the OpenAI team, CEO Sam Altman claims the company is facing several economic headwinds after Google’s recent Gemini 3 Pro release which smashed benchmark records setting a new precedent for the leading AI model.
He claimed the company is facing “rough vibes” as the standout quality of their GPT models is called into question. Despite this, Sam claimed that OpenAI is “catching up fast” and that he “wouldn’t trade positions with any other company.”
Unichain offers the most liquid Uniswap v4 deployment on any L2 – giving you better prices, less slippage, and smoother swaps on top trading pairs. All on a fast, low-cost, and fully transparent network. Start swapping on Unichain today.
Anthropic: AI’s Dark Horse?
Ejaaz on why only one benchmark matters right now.
As excitement over Google’s Gemini 3 Pro launch was still peaking last week, a surprise contender stepped out of the shadows: Anthropic, and for reasons no one saw coming.
Their debut of the Claude Opus 4.5 model is a standout threat to their industry competitors for one simple reason: it wins at coding.
I know plenty of you are growing tired of model benchmark jockeying, but as the industry matures, it’s becoming increasingly clear that the coding market is big business — honestly it could be the only thing that matters at this stage.
Imagine spending years building and iterating your product until it’s perfect… and then an AI spends seconds creating an exact replica from scratch enabling competitors to dismantle all your hard work in seconds. Pretty depressing, right? Makes you wonder what the moat behind a software product is.
An Ephemeral Future
Well here’s my theory on where that end state is: Ephemeral AI.
If you assume that the best AI will copy your product in seconds then the inevitable result is every experience online becomes ephemeral (short-lived) and transient.
Imagine wanting to buy something online — but instead of hunting down the right brand and clicking through their website, your AI just builds a custom shopping interface on the spot. It spins up a site in your favorite subtle blue tones, defaults to your preferred payment methods, filters for your preference for high-quality materials, and surfaces only the brands you actually care about. And when you’re done, the whole thing vanishes, and you’re right back to asking your AI for whatever’s next.
The only way this kind of experience becomes real is by building vastly better coding AI.
Anthropic gets that — which is why they’ve poured essentially every dollar they’ve raised into pushing frontier-level coding intelligence. They see the future of digital experiences as ephemeral, and they want to be the ones powering it.
So now that the vision’s clear, here’s why Anthropic actually has a convincing lead.
AI Coding Supremacy
First off, they’ve had the best coding model for a while. Claude Code, Anthropic’s mass consumer coding product, is the most widely used AI coding tool in the world. Whether you’re using the desktop client, the chrome browser plug-in, the browser agent or even via a 3rd-party like Cursor… Claude is king.
Oh and it’s not just retail consumers, enterprises love it, too. In fact it’s so popular that Anthropic has taken the lead in enterprise customer market share beating out OpenAI. This alone has put them on a path to profitability in 2028, projecting a massive $70B in revenue.
They beat Google despite having fewer resources.
Last week, Google Gemini 3 Pro beat every single model on every benchmark (by a wide margin)… except one, the SWE coding test, where Claude Opus 4.5 scored an incredible 80%. To put this into perspective, this model can do pretty much any of the hard grunt-work required of a talented entry level software engineer, in some cases even an advanced one.
Don’t believe me? Check out this testimony
The point is, even with less funding, fewer resources, and nowhere near the hype surrounding the other major AI labs, Anthropic has still pulled ahead on coding. It’s clear they’ve tapped into a training formula no one else has cracked — and so far, it’s held up against every competitor. That’s a real moat.
An Unlikely Alliance
Next, did you know Google owns 14% of Anthropic?
Did you also know that Anthropic just signed a deal with Google to access 1 million of their custom TPUs to train the next version of Claude? With access to the same chips that trained Gemini 3 Pro, it’s not too difficult to imagine Claude 5 will be leagues ahead of any other model at coding.
For Google, their TPUs are specifically designed for niche use-cases, a perfect fit for Anthropic’s obsession with coding. Their most recent TPU, Ironwood, is rumored to have unlocked a 1.5-2x performance improvement per dollar spent on training — which is insane.
Conclusion
I’ll be honest, if you had told me three months ago that Anthropic was going to be a real contender for best AI lab, I would’ve laughed at you, but it turns out their team is having the last laugh.
The combination of being hyper-focused on coding, capturing the enterprise market while everyone else obsessed over retail and securing the next-gen GPUs from Google makes these guys a force to be reckoned with imo.
The rumor that they’re about to announce a $400B valuation (which puts them only $100B less than OpenAI) is now starting to make a lot of sense.
Thanks for joining us for our twelfth issue. Now go listen to our podcast :)