The pain point they raised is this is too complicated for people who just want to get started, that is not true anymore.

It is certainly fun to fine-tune and setup if you like do something like that, however the need to do it hardly is a barrier for those who don't want complexity as OP imagines.

Lower level API/interfaces should not be a barrier for people if they are apply framing that way. More and more people are thinking agent native so this is not really a issue.

They could ask your current agent to a) search for this type of content online for the optimal setup for their hardware b) have the current agent/harness spin it up have it verify the config run few experiments.

Sure AI may make mistakes, or won't get the best possible config probably, but it certainly do a good enough setup, this is a task with feedback on whether the server crashed or poor performance easily measured so the agent can do a pretty good job.

It is not once and done you have some continuous costs in either direction.

1. you will not be able to still move all your workloads only your base loads .

2. There will be residual workloads not only if they are spot/flexible bursts but also they are too deeply integrated to Paas offerings .

3. you have to pay for power cooling and network (and its backup) for 5 years .

4. Finally you also have pay for infra team to maintain, this is increasingly difficult (and expensive) to hire and retain .

Don’t get me wrong I would love to be able buy something like this, but the cost economy is pretty steep and 1-2M spends is too early (it is not on oxide, serious hardware costs a ton these days )

My understanding is the full rack is about $1.2M and smallest half rack is about $600k and this was before the RAM and other price hikes now.

You are still going to have some residual cloud workloads(so all AWS won’t migrate ) and HA and DR regions etc plus the maintenance and incidentals (power , connectivity ).

I don’t expect anyone less than 5M spends and high base loads is a good fit for their offering.

It only bit of effort to write in your own voice , anyone with high school level of writing skills should be able to muster if they were inclined to do so .

[1]used as aporetic or perhaps even epiplectic device

Cost of per kilo to GEO is high. That is not going to change by enough even post starship level of 2 stage resuse - as GEO launches would need a third kicker stage as the second stage will not go so high to be reusable.

Also nobody wants to clutter GEO with many new sats. There are limited slots and dead satellites need to move graveyard orbits and cannot be de orbited[1]. The slots + frequencies are so limited that small pacific country of Tonga[2] makes millions every year renting this out.

Every final stage/booster that launched a GEO satellite is also in orbit now. Including the stage that launched this mission.

All this method basically reduces launched orbital objects from two (stage + sat) to just stage. The new sat/repair unit just attaches to older satellite .

---

[1]In reasonable time frame that is, unlike in LEO, GEO reentry times can range between hundreds to thousands of years

[2] Due to quirk of international law they have rights to it. Like IP address allocation it was basically first come first serve for countries and one guy figured he coule exploit the loophole via small sovereign country .

[1 ]https://openknowledge.worldbank.org/entities/publication/130...

K3 may be bit smaller/ similar in total parameter count than Opus, but Opus (and GPT) definitely has become a lot more efficient in the last 6 months or so, hence more or less forced upgrades. I expect the active parameter count and cache performance is quite different.

--

Opus is still almost twice as expensive (if tokens were equal) at $5/$15 compared to K3 at $3/$15. Tokens are not equal though, Anthropic's tokenizer is much less dense than other frontier lab's so the actual price difference is like 3x.

Kimi uses more reasoning tokens and is generally more inefficient with its usage, that doesn't impact token cost economics for the provider though . It does for us as buyers thus the need to evaluating by Cost per task rather than unit pricing.

On pure tokens/$ - there is definitely room for a price war if operators start going by pure unit costs. Both probably want(ed) to have good enough numbers in preparing for the IPO.

---

[1] the pricing kind of reflect this already - 10x diff for uncached input.

[2] Moonshot does not have access latest gen GPUs so their unit economics is likely hampered a bit for high parameter models.

It would be illegal under the currently proposed /implemented laws and also open up social media to liability, which wouldn’t be true for other products like Alcohol or fire arms that require minimum age to buy but not give to children

There is an enormous open short position in SpaceX now https://www.bloomberg.com/news/articles/2026-07-15/short-sel...

The float is going to triple in several steps with 900M employee stocks getting added to 550M issued in the IPO.

The thesis that larger pool will depress prices even if all other things remain same is quite sound, just like a small float inflates the price .

These events are merely triggers / focal points as would the quarterly results next month and various unlocking dates .

the traders are savvy they are just looking at different things and using company news to focus the price actions on .

Valuations are not based on business models or financial performance for the highly visible public companies, that has always been the case since earliest days of the modern market .

it is just vibes or to put it differently a strong belief/faith discounting every risk . Without that faith TSLA won’t trade at 350 PE with middling growth no new products currently scheduled . SPCX at $2T+, even for an AI company a lot - it is larger than valuations of both market leaders combined while having <5% share/revenue .

Maybe the faith will prove justified or maybe not, either way the price moves by sentiment alone not any actual business logic

There was an another around after that . Moonshot raised $2B on a $20B valuation in May - https://techcrunch.com/2026/05/07/chinas-moonshot-ai-raises-...

It is closer to corporate sponsorships, plenty of projects[1] can pull in few hundred thousand a year.

The post itself talks about a $5k/month donation from Bun to Zig Foundation. ZSF generates about a half million in donations https://ziglang.org/news/2025-financials/

The "crowdfunding" term is here to contrast direct VC funding (that Bun took) from a single source to more diffuse(albeit smaller) funding that ZSF style projects operate under. Without VC funding Bun wouldn't need to , and if non profit 501(x) then simply not even be to consider an acquisition from Anthropic.

Of course Zig (and other projects) benefit from indirect VC funding too - Bun could only afford to donate $5k/ month on Zig because they took VC money after all.

We can always keep going down this secondary/tertiary effects for any tech funding it would unavoidable even with direct small donations - people who contribute are likely drawing salary from VC funded companies, and would contribute less if market goes south and funding plus jobs dry up. . The idea being the level of influence from corporate goal pressure reduces with degrees of separation.

There is a good amount history of this model. Linus Torvalds work itself is by funded by Linux Foundation through a fellowship. LSF charges $500,000 from their 12 Platinum members that includes Microsoft. Nvidia is a silver member, he has famously used some colorful language against both in the past, which in theory would be lot more difficult without the diffuse structure LSF generates funding ( compared to direct employment/VC funding).

[1]That isn't to say that a lot of core projects the entire internet economy depends on is maintained by one guy in Nebraska voluntarily without any support.

The dynamics are not any different from content creators translating their following to income on Patreon -larger the following better the chances.