M.G. Siegler β€’ β€’

Inklings #004 πŸ“§

OpenAI Missing Their Numbers β€’ Amazon's 'Quick' Superagent β€’ Cost-Per-Click AI Ads β€’ Stargate Shift β€’ Meta's Reverse Hackquisition β€’ Microsoft & OpenAI's New Deal

Happy Earningspocalypse to those who celebrate. Amazon, Google, Meta, and Microsoft all report today after the market closes. Obviously, all eyes will be on AI spend – and any signal of actual returns from spend to date.

How high will the CapEx numbers go this time?

Thoughts On...

πŸ’« OpenAI Missing Projections – On one hand, you could read a startup missing internal projections as the company either not being great at extrapolation (more understandable during fast growth) or setting targets that are too optimistic (perhaps to help woo investors). These are internal numbers for a reason; they aren't held to the same level of scrutiny as, say, those of a public company. On the other hand, OpenAI is an awfully mature company at this point having raised more money than any other in history, with an experienced team, and one that is obviously trying to go public sometime soon. Plus, as far as we know, they haven't missed on such projections in the past, so... It's not a great look, but it's made worse by their response which includes the highly scientific measurements such as "firing on all cylinders" while calling the report "prime clickbait" – both of which, of course, are non-denial denials to a report, which has now been backed-up. Even the partners, whose stocks are getting hit by proxy at the moment can't seem to align on the right rebuttals here. The actual actions of the company perhaps remain more indicative here – killing "side quests", altering deals to expand their market, and yes, bringing on some help in the marketing/PR front. [WSJ πŸ”’]

πŸ’¨ Amazon's 'Quick' Superagent – The branding for their desktop-based AI assistant isn't bad, but confusing. 'Amazon Quick' definitely sounds like a fast delivery offering, not their version of Claude Cowork. Then again, 'Q' didn't really work for them – and was weird for other reasons – so I guess just add some letters? More importantly, Amazon doesn't have the best track record when it comes to software. Almost all of it whether on the web, desktop, or mobile, quite frankly, sucks. So the hope will have to be that their approach – multiple models, obviously their own 'Nova' models and presumably Claude, but presumably not GPT, at least not yet – and their AWS connections will win the day. For $20/month, it better be fast. [Information πŸ”’]

πŸͺŽ Cost-Per-Click AI Ads – It feels like OpenAI has little choice but to go down this path already for ChatGPT – they need to start showing results to advertisers and this is the way you traditionally show results. But I'm still quite skeptical this model will work all that well in AI. There's a very fundamental problem: such AI services, at least to date, have removed most of the need to click. That's how they've been disrupting Google, of course. With web search, users are trained to click. With AI, users are trained to not need to click. So, yeah. Good luck getting to that $100B+ advertising business – which would be the overall biggest driver of the business at that point – in a few years this way. I just think they'll have to find a new model and methods for advertising in AI. Just like Google had to do for Search and Meta had to do for the feed. [Digiday]

I Quote...

"Stargate has now had three or four permutations, I don’t know what it is right now. I can’t tell you. Maybe it never really existed in the first place."

– An unnamed "someone familiar with Microsoft's thinking" which, of course, sounds a lot like someone who works at Microsoft and is undoubtedly chuckling right now over "Project Stargate". As it, of course, actually started life as an initiative between OpenAI and Microsoft, before "The Blip" derailed, well, pretty much everything between the two.

Soon, Stargate was reborn/repackaged as a $500B mega-project unveiled at The White House between OpenAI, SoftBank, and Oracle. Why? Because obviously OpenAI as a massively unprofitable startup couldn't raise the debt required to finance such a project on their own. Yet they probably needed to in order to ultimately compete with Google.

But, well, the best laid plans and all that – which, to be clear, this was not. Now we have Microsoft picking off Stargate scraps and "Stargate" itself morphed into just a catch-all PR term for OpenAI data center deals. "I don't know what 'Stargate' means at this point," says another "involved" buy unnamed source. So say we all.

Asides...

  • While I'm not sure about 'Quick', an AI-powered voice-based shopping assistant sounds like a smart thing for Amazon to try. [TechCrunch]
  • John Giannandrea has his first post-Apple role, albeit part-time, helping out a UK startup, CuspAI, scale in the US. [Upstarts πŸ”’]
  • More 'iPhone Ultra' smoke as the name for Apple's upcoming foldable. And, like the 'iPhone Air', no number in the name, it seems. Plus, the touchscreen OLED MacBook, should be the 'MacBook Ultra' – in 2027. [MacWorld]
  • Tired: data centers in Space. Wired: powering data centers from space, as Meta is seeking to do with a new partnership. [Bloomberg πŸ”’]
  • Much like Soylent Green, IAC is now 'People'. [NYT]
  • Nic Cage as 1930s-era Spider-Man? Yes. Also interesting: you'll be able to watch the Prime show in black & white or in "true hue" color. [THR]

This post is for paying subscribers only