I believe that not using AI when modeling should be considered negligence. In the near future 1-2 years, this opinion will be considered obvious. like FSD for investment decision making. You can absolutely model by hand if you so choose, but it’s not the smart decision. Every institution will have to grapple with this truth in the coming months.

Something feels off with this chorus of “India’s internet is great and super cheap” … the data shows that Indian market is absolutely dominated by mobile. When they say it’s cheap and well connected factually this only applies to mobile usage. This makes sense because a tower connects thousands of subs instantly after construction and running fiber is a building by building or resident by resident job. This is also why “fixed wireless” has started growing fast. Follow the incentives. It’s the cheapest and shortest route to revenue. Due to this, India is amongst the highest mobile bandwidth usage populations on the planet. 30-40GB/month. (20-50% above US/UK usage) However, home Fiber lines in India pull closer to 300GB/month. 10x gap For context, that gap in the US and UK is 20-30x. It’s pretty ubiquitous that if you have access to a lot of more bandwidth you will use it. Bandwidth abundance does not favor mobile devices. I’d argue that it’s not very hard to believe that these paltry sub 5% fixed customer penetration numbers would be much higher if a cost effective alternative existed. Is that alternative existed it wouldn’t be crazy to expect to see the fixed customer numbers jump from 6% to maybe 10%, 15% or 20%. 100M-200M customers. And if the rest of the world is any precedent that would lead to a dramatic increase in overall bandwidth used. However, that would mean that mobile bandwidth usage would likely drop or stagnate with in home mobile usage.

Excited to announce that all super intelligence organizations have now jointly agreed to the ultimate in AI/SI safety: Moving all testing to Delta Airlines flights, where accessing the Internet is utterly impossible!

Remember that phones coming in 2027/2028 will be equipped with chipsets to handle Starlink mobile’s other higher throughput spectrum natively. Another part of the spectrum puzzle that’s already sorted but part of the conversation before SpaceX was public so has gone under the radar.

building out the full suit of spectrum needed for starlink mobile. This low band is good for in building, My hunch is there may be some more spectrum purchases in the future, but almost certainly much less than the big MNOs believe is required to compete head to head.

I use Starlink everyday and Starlink mobile on all my road trips. Switched from spectrum over a year ago, far less connection disruptions. Easy recommendation to make. If you’re considering the switch and $100 makes the decision easier you can use my link. My referral credits go toward service credits. Or don’t use the link out of spite and share a screen shot in the comments just to rub my face in it.

If you’re trying to model Starlink over the next 2 years, I think this thesis has only gotten stronger with the release of grok bot. Top tier Starlink plans are only going up from here. $/Gbps may go down, but bundling one of modern humanity’s most basic utilities with grok bot is consumer bundling dream. This matters because even modest reduction in churn has enormous customer lifetime value gains at scale.