Ask HN: What's the Deal with Vast.ai?

I'm intrigued by vast.ai from a client and host perspective.

As a client (renting GPUs) I've found it to be pretty unreliable. I've tried renting GPUs a handful of times and hosts, bandwidth, and performance are all over the place compared to what's listed in the console. I suspect people are gaming the host benchmarks somehow. Additionally, even when using on-demand instances the reliability is nowhere near what's reflected by the reliability/uptime metrics reported in the console.

While they've done a pretty good job considering with their software stack it's still kind of a pain to use with significant limitations and multiple hoops to jump through compared to say a GCP/AWS GPU instance.

That said, there are stories of people using vast.ai successfully so maybe I've just had incredibly bad luck?

From a hosting standpoint I can't figure out the economics/opportunity:

Median pricing in the US for a verified RTX 3090 is $0.38/hr. Let's assume total capex is $2000 (likely much higher looking at reported hardware configs).

Assuming US national average electricity costs of $0.15/kWh and hardware being under fairly high utilization you're probably looking at 500 watt load. Let's assume costs like bandwidth, etc are baked in because it's hosted on a circuit you'd be paying for anyway.

So ongoing operational cost (power) is $0.07 (rounding down). Vast.ai takes 25% of listed pricing. Basic math for this theoretical host:

Rental fee (median): $0.38/hr

Less vast.ai take: $0.29/hr

Less power cost: $0.22/hr

Less taxes (they ask for SSN/EIN so I assume income reporting) at effective (low) tax rate of 10%: $0.20/hr

Absolute best case monthly income assuming full usage with 730 hours/mo: $146

Yearly income: $1752

ROI for hardware capex: 14 months

The numbers for A100 PCIe rental are much worse - median rental cost of $0.79/hr with a $15k card. You're never making money on that...

I assume most of the hosts are fairly technical and their time is relatively valuable (imagine a random tech consultant that makes $75/hr). If you consider this side gig "work" and spend an hour a month dealing it (even controlled maintenance, etc) your monthly income drops in half relative to your ability for billable time (opportunity cost).

In the case of sharing an internet connection for a remote worker/home host without fairly advanced QoS bandwidth bursts could absolutely wreck the usability of your internet connection. Your fans will be screaming anytime you're actually "making" money, dumping hot air in your environment, potentially increasing air conditioning power costs, etc.

PLUS now you have random internet users on your internet connection (and potentially LAN!) which opens up all kinds of security/legal issues.

How can this possibly be worth it?

Is this just desperation because of the collapse in cryptocurrency mining? What about the A100s, etc listed that no one in their right mind would have bought for cryptocurrency mining? Is this another Uber/Doordash/gig economy case of people not fully understanding the economics, risks, etc? Maybe some kind of arbitrage?