📰 Welcome back!

Hello AI fans and welcome to 2026.

As a newsletter, we bring the noise. Or try to. Or try to cut through it. Whatever.

Point is, as we start the new year, we want to do things better. Over the past few weeks we’ve been plotting how to do this better; to be quicker and more consistent. Perhaps even funnier.

Don’t worry though, some things will remain the same – same irreverence, same typos, same bullshit industry takes and contempt for the dick-swingers in it.

So this week we wanted to say hello and goodbye in our current form – we can’t wait to show you what we’ve been cooking soon.

🚀 What we’re covering today…

• 🤫 Grok’s sexual deepfake scandal… again

• 🫧 AI snot bubble speculation

• 🗣️ Yan LeCun dishes the dirt on Meta exit

• 💪 TII’s small but mighty model

• 💣 Blowing up the AI job apocalypse narrative

• 🤸🏻‍♀️ Break-dancing robots at CES

• 🤖 AI’s job hunt

• 💰 xAI’s $20bn funding bump

• 🍎 Apple’s Gemini bet

• ☢️ Meta pushes the nuclear option

🔴 Quick Note: We like to cover loads of AI news in our newsletter, so for a better reading experience, we suggest opening this in your browser for the full experience! 

Head to the ‘READ ONLINE’ tab at the top of this email.

👁️ 👁️ What you might have missed

• There’s nothing like starting the year with a massive grubby scandal. And of course, Elon Musk is at the heart of it. A new trend saw Musk's Grok chatbot predictably plunge into a shady, nonconsensual porn generator, with users asking the model to undress submitted images.
  
  As you can imagine, the requests turned lurid quick, shuttling against Grok’s famously loose safety parameters. xAI's already tiny safety team has lost key members while CNN reported that Musk had ordered staff at xAI to loosen the guardrails on Grok last year, with a source saying he was “unhappy about over-censoring”. Three xAI safety team members had left the business soon after. In light of ‘bikini-gate’, the developer responded:
  
  “There are isolated cases where users prompted for and received AI images depicting minors in minimal clothing,” Grok said in a post on X in response to a user. “xAI has safeguards, but improvements are ongoing to block such requests entirely.” Another post followed: “As noted, we’ve identified lapses in safeguards and are urgently fixing them—CSAM is illegal and prohibited,” referring to child sexual abuse material.
  
  In response, xAI has confined Grok’s image-generating capability to paid users, but the uproar has been deafening. Literal countries are banning the model. Indonesia and Malaysia became the first countries to ban Grok entirely, while the UK government and its European counterparts are now investigating its legality – and ability to “self-regulate” in the country. Musk, initially and somewhat predictably responded with apathy; dismissive emojis to pics of him with tits, but plenty of noise about how his platform has reached record-high engagement levels.
  
  Since xAI’s pedogedden started Musk’s X team seemed to have relented, saying they’ll block Grok from engaging in any form of digital undressing, but as of this weekend, it appears you can still use the app to create sexual deepfakes. Life comes at you fast.

  

  Bleach > Eyes. Source: X/@elonmusk

  
  

• Anyway, we’re starting as we mean to go on: Lots of bullshit analysis about the AI snot bubble and when it may burst. As it was when closing out 2025, this still remains to be seen. Record valuations, IP acquisitions and the rapidly advancing state of tech make predictions slippery, but we do know that in just the second week of 2026 the AI hypemachine has already gone haywire. Gartner called it first last year: AI spending will smash through $2 trillion this year – and up to $2.5tn by current estimates – up from $1.5tn in 2025. That's a lot of GPUs, folks. And in the latter stages of last year their analysts said it’s only going to get bigger as investment into the sector creates a new supercycle. Whether this represents genuine value creation or the biggest tech bubble since the Dot Com glory days remains to be seen. So where’s the money headed? The team at The Motley Fool have made some good insight – check their stock predictions for 2026 to get a better idea where the smart money is headed.
  

• Yann LeCun, has finally dished the dirt on why he quit Meta as its Chief AI Scientist back in November 2025. Pressure to rush development, Llama 4’s spectacular flop and Zuck’s $14bn bet on Scale AI all were motivators, but it feels like the introduction of 29-year-old former Scale AI CEO Alexandr Wang to run its “superintelligence division” – and by virtue become LeCun’s boss – was the tipping point. Needless to say, “the godfather of AI” was not impressed, calling Wang “young” and “inexperienced” before describing the indignation of being told what to do. LeCun has since stepped out on his own, launching his own world-model startup, and poaching Meta and OpenAI talent in the process.

• Is it time AI finally grows up and gets a proper job? The obsession with building ever-bigger language models may well be giving way to smaller, practical deployments that actually solve real problems, according to the folks at TechCrunch. I think we can all agree our focus is less "AGI by Tuesday" and more "can this thing automate my shitshow of a CRM before I junk my laptop out a window?" etc. TC says ‘world models’ are becoming a major focus in AI development for 2026; systems that understand 3D spaces and physical interactions, representing a shift from language-focused models to AI that can comprehend spatial relationships and how objects behave in the real world – pertinent for essential infrastructure such as robotics, autonomous systems, and applications requiring physical world understanding.
  

• Technology Innovation Institute just dropped Falcon-H1R 7B, and we think it's a bit of a David vs. Goliath moment. This compact model matches the performance of systems seven times its size, which is genuinely impressive. Smaller, more efficient, and probably cheaper to run – if we want to get into it: “Falcon-H1R scored 88.1% on the AIME-24 math benchmark, surpassing the 15-billion-parameter Apriel 1.5 model, which scored 86.2%. It also achieved 68.6% on LCB v6 coding tasks, outperforming the 32-billion-parameter Qwen3 by about 7 percentage points. Impressively, it processes around 1,500 tokens per second per GPU at a batch size of 64.” This fits perfectly into 2026's trend toward practical, deployable SML/AI rather than massive, expensive models that require their own power station. Size matters, nudge nudge, wink wink.
  

• Continuing our theme of being pragmatic: Oxford Economics thinks the AI job apocalypse narrative may be a bit trumpy. Turns out, AI-related job losses might be relatively limited compared to broader market layoffs, and instead, a more cunning kind of corporate shithousery is afoot. Oxford Economic believe firms might be using AI as convenient cover to engage in more traditional cost-cutting and restructuring practices, which is neither very new age, very cool nor very surprising. “We suspect some firms are trying to dress up layoffs as a good news story rather than bad news, such as past over-hiring,” the think tank said. Their research also addressed fears that AI is eroding entry-level white-collar jobs, citing graduate unemployment levels peaking at 5.5% in March 2025 but stating this is likely “cyclical rather than structural,” pointing to a “supply glut” of degree-holders as a more probable culprit.
  

• What better ways to allay fears of an AI bubble blowing than break-dancing robots at this year’s edition of CES? I honestly can’t stop laughing at these little humanoid chaps, cutting shapes and fulfilling whatever utility they’re meant to fulfil. And by the looks of it, this included playing chess (boring), stock a refrigerator (lame) or perform open heart surgery (terrifying). Anyway, the real story is about industry giants being doorstepped over topics like energy infrastructure and hardware manufacturing, and why exponential acceleration gets some feeling spooky. But this is CES and it served as a bubble for positive industry sentiment in itself: “We’re in the earliest stage of what’s possible. So when I hear we’re in a bubble, I’m like… This isn’t a fad,” deflected Panos Panay, Amazon’s devices and services chief, artfully. “It’s not going to pass.”
  

  

  Never not gets me. Source: China News Service

  
  

• xAI has raised $20bn in new funding, running clear of its original target of $15bn. xAI said the funding round values the firm at $230bn – more than double it posted last spring. Strategic investors include Nvidia and Cisco. Anthropic, meanwhile, is planning to raise $10bn at a current valuation of $350bn, again nearly double its valuation from just four months ago. This has fuelled expectations Anthropic will go public this year. New capital will top the $15bn Nvidia and Microsoft will invest in the Claude creator, as tech’s big boys and girls hedge their investments in the big frontier developers. 
  

• New year, the same maladies for Apple’s own AI IP – although things could be getting significantly better after the iPhone developer struck a multi-year deal to power Siri and other services using Google's Gemini AI. IDC analyst Francisco Jeronimo didn't mince words: Apple's “effectively admitting that its internal efforts couldn't compete with Google's Gemini,” marking a massive deviation from Apple's usual protocol for keeping software dev in-house. The company already partnered with OpenAI for ChatGPT integration last June – something that has largely improved the customer experience – but this could cause raised eyebrows among regulators given Google and Apple's existing "effective duopoly" in UK markets.

🧩 Other Bits

• Can’t get enough of in-app advertising disrupting your experience? No me neither. OpenAI has said it is trialling ads for some US ChatGPT users based on their chat conversations. OpenAI is urgently trying to maximise revenue from its 800 million monthly users in order to afford the $1.4tn it has committed to spending on AI infrastructure over the next eight years. CEO Sam Altman said the company expected to end 2025 with around $20bn in annual revenue, but would need to multiply this by a factor of many to achieve its ambitions.

• Talking of bridging the AI infrastructure gap, Meta has struck deals with three companies to supply nuclear power to its data centres. According to TechCrunch, Oklo and TerraPower, to developers small modular reactors (SMR), each signed agreements with Meta to build multiple reactors, while Vistra, a nuclear operator with a portfolio of reactors across the US, is selling 2.1 GW of capacity from two power plants in Ohio.
  

• Siemens and NVIDIA have announced an Industrial AI OS partnership, with the first reference site launching at Siemens' Erlangen facility in 2026. This is AI moving from chatbots to actual factory floors – a trend we’ve covered towards the tail-end of 2025. The collaboration aims to bring NVIDIA's AI chops into industrial settings, potentially transforming manufacturing – something Siemens calls an “AI Brain” that enables real-time adaptation over periodic optimisation.
  
  

💸 Financials

• AI evaluation platform LMArena raised $150 million in Series A funding at a $1.7 billion valuation, nearly tripling its value from a $600m valuation just seven months earlier. The round was led by Felicis and UC Investments, with participation from Andreessen Horowitz, Kleiner Perkins, Lightspeed, and others.
  

• Robotics startup Skild AI secured $1.4bn in funding that values the company at more than $14bn – more than triple its $4.5bn valuation from just seven months ago. The Series C round was led by SoftBank Group Corp., with participation from Nvidia Corp., Macquarie Group, and Jeff Bezos' Bezos Expeditions.
  

• Chipmaking stocks rallied to kick off 2026, with Micron Technology and ASML jumping 10% and 9% respectively, while Lam Research and Intel rallied more than 6% each. The rally lifted the VanEck Semiconductor ETF about 4% to build on a nearly 49% rally in 2025.
  

• SpaceX is reportedly looking at a value of around $1.5tn when it hits the public market, which would surpass the prior record set by Saudi Aramco's IPO in 2019. Media reports have suggested OpenAI could debut on public markets at a $1tn valuation, which would make it the highest valued IPO of all time.
  

• Capital spending on AI infrastructure is expected to grow from $600bn in 2026 to at least $3tn by 2030, according to Edge AI + Vision Alliance, while total global funding to VC-backed financial technology startups totalled $51.8bn for 2025, a 27% increase from 2024's total of $40.8bn
  
  

🕵️‍♂️ FREE ENTRY TO OUR INVITE-ONLY AI CHAT ON TELEGRAM…

If you share this newsletter with a friend and they actively sign up for the Big Machines newsletter, we will send you access to our invite-only Big Machines Telegram group, which is full of builders, investors, founders, and creators.

Access is now only granted to those who refer our newsletter to active subscribers, which means if you sign up on your work email, we will know, you sneaky bastards.

This would kill our open rate, so please don't do that, we beg.

👋 Bon voyage, farewell and we’ll see you soon

Sam, Grant, Matt, Mike and the Big Machines team.