Inklings #002 📧
Everybody already assumed that after Xitter merged with xAI and xAI with SpaceX that Tesla would be next. But Cursor – and possibly Mistral – throw some interesting new wrenches into these bonkers equations to build this Voltron of a company...
M.G. Siegler
Thoughts On...
🍁 Cohere + Aleph Alpha – Begun, the great AI consolidation has. The newly combined $20B AI startup seemingly has a better shot at government and business contracts across the EU now, given Aleph's German roots and Cohere's bigger scale and more well-known pedigree. Still, beyond OpenAI, Anthropic, Google, Microsoft, and everyone else (all the "open" Chinese models and players?), they'll presumably be running head into Europe's bigger AI champion: Mistral. That is, unless Elon Musk can successfully buy them as well, which, while perhaps good for Mistral's tech and fundraising capabilities would immediately shove them into the US 'Big AI' bucket. And that plus the Musk factor would likely push more business Cohere's way. We'll see. Regardless, this is clearly the start of more consolidation in the space, following the news that SpaceX has a call option to buy Cursor later this year. Cohere – by my count, backed by AMD, Cisco, NVIDIA, Oracle, and Salesforce clearly needed an answer beyond being Canada's AI champion, and they found one, but is even that enough? [FT 🔒]
🥔 GPT 5.5 – "Spud" seems nice and fast thus far. Greg Brockman, when not constantly saying "I see things a little differently..." to every single (good) question put forth by Alex Kantrowitz still gave a pretty good overview of the state of the model – generally how it can do more with less guidance. And it puts OpenAI back at the top of most (but not all) of the key AI benchmarks, beating out last week's Anthropic Opus 4.7 and even the mythical Mythos here and there (but Mythos still seems like it will be ahead if and when it's released broadly). Of course, that performance comes at a cost, quite literally, a higher cost on the API (though OpenAI is saying it's far more "token efficient" which should negate that). Mainly, I want to see how/if this alters the perception of OpenAI vs. Anthropic. Does this (plus the impressive new image model) help them regain momentum? 5.5 will apparently power the forthcoming "Super App". More importantly for them, does it help power Codex versus Claude Code? It also sure feels like there's an effort to change the tone and tenor when talking about such releases by OpenAI's outward facing folks from Sam Altman on down? Though that could be just as much about the fact that Altman (and Brockman) are about to be thrown head-first into a very public trial against Elon Musk next week... [Axios]
🐳 DeepSeek V4 – Do you hear that? No you don't hear anything? That is the sound of everybody talking about the latest version of DeepSeek. "No one cares." That's not fair, of course – it may end up another step for open source models with regard to coding and agentic usage – but relative to the "DeepSeek Moment" last year, it's sort of wild how few shits anyone gives now. The stock market is not only not crashing, it's up at the moment. All of this is undoubtedly because of what ended up being the actual fallout from the famous "Sputnik Moment" – which was... not a whole lot. Some cheaper models, sure. But it's not like the top model makers were displaced, let alone world economies. If anything, it just opened up Big Tech's eyes to the potential problems of distillation and the US government to looking into the situation, including, potentially, with unauthorized usage of NVIDIA chips. Anyway, a couple new DeepSeek models are here now in preview. One seems to be the largest open model ever and the other is perhaps the cheapest. But they're also not on par with the top frontier models – which unlike DeepSeek, are getting updated left and right at the moment (see: above). So... the question remains if open models truly reach "good enough" status for their cost advantages to eat into frontier's take. Meanwhile, some key models in China are following Meta and going the other way, locking down the cutting edge... [Bloomberg 🔒]
📈 Intel's Bounce Back – With the stock up – checks ticker – 25% today, Intel share price is now past its Dot Com era peak. This is wild any way you slice it – that was 26 years ago – but especially given how downtrodden Intel has been over the past many years after famously failing to capitalize on the shift to mobile and then, at first, to AI. But the broader chip boom is raising all boats. Intel may not make GPUs, but their CPUs are still vital for data centers (and yes, some PCs) and now it seems like Pat Gelsinger's big foundry bet may be starting to pay off, with Tesla signed on to use their advanced '14A' node (eventually). Sadly, Gelsinger didn't survive the Board and Lip-Bu Tan was smart in playing the game on the field, as it were – which meant giving the government a taste of the upside here. That $8.9B for a 10% stake is now worth over $40B. Might Apple get back on board with using Intel for at least some of their chip needs – maybe modems, the business they bought from Intel? Even just packaging? Seems like Amazon and Microsoft are hanging around the rim. Intel isn't quite back at their peak market cap, which was just over $500B in August 2000, but they're close. The only problem? NVIDIA's market cap just shot past $5T again today by taking advantage of an opportunity Intel missed at first. [NYT]
I Quote...
"When I got the call I said, wow, it’s Tim Apple (Cook!) calling, how big is that? I was very impressed with myself to have the head of Apple calling to 'kiss my ass'."
– The President of the United States of America, giving his honest assessment of the power dynamic between himself and the CEO of one of the biggest companies in the world who just announced he's stepping down from the role – but not fully leaving Apple as he'll still be in charge of interacting with said President.
Asides
- Threads is copying Xitter's live conversations feature – itself copied from Clubhouse, RIP – to try to make the network feel more alive and real time. Not sure that will work, but appreciate the continued effort! [TechCrunch]
- Instagram is copying Snap for the 200th or so time, trying to make Snapchat itself happen yet again with a new 'Instants' app. [9to5Google]
- Miami Vice '85 is official with Michael B. Jordan (fresh off his Best Actor Oscar) and Austin Butler locked and loaded. The fact that they refer to this as a "period piece" hurts my soul. But while I actually quite liked Michael Mann's 2006 version, it sort of has to be set in the 80s? Joseph Kosinski is going to film it in IMAX but will it be shot entirely on IMAX cameras (the second movie after Christopher Nolan's upcoming Odyssey)? [THR]
I Spy...
With new leadership in place, Microsoft is very clearly trying to make Xbox great again. "Microsoft Gaming"? Goodbye. Game Pass prices? Cut. Call of Duty? Back to being sold to start – perhaps for a year or more, which just makes sense. Exclusive titles? May be back too. I would still put some of the games on other systems after an exclusive window as it could entice players to switch to your platform for sequels, etc. This is all so obvious, it makes you wonder what the hell Microsoft was going for these past few years. The answer, I fear, is nobody knows.
Below, members of The Inner Ring will find thoughts on:
• OpenClaw Struggles
• Big Tech Layoffs Amidst AI Boom
• and more...