It’s time to rethink compute — thoughts from our Co-Founder Alessandro De Carli
As we surpassed 65'000 devices in over 130 countries on the Acurast network, it has become very clear to me that our vision of a real decentralised compute network is no longer just a dream — it’s becoming reality. Further, it confirms my thesis that the only way to actually create a sustainable global compute network is by sourcing the compute of phones.
Does security matter?
Coming from an extensive history in the mobile cyber security field, we started working on Acurast in 2022. We mostly started the project because we were very well aware of the superiority in security provided by the Hardware Security Modules (HSM) of mobile devices. While security often feels mundane (until something goes wrong) in the context of a decentralized compute protocol, it means:
- Verifiability of the hardware and execution
- Confidentiality of the execution
Both are properties which are imperative for a decentralised compute network.
Without verifiability you will end up with fake hardware, aka the compute is not real. The reason is simple: in an adversarial environment like the internet actors will always cheat if it’s beneficial to them. This phenomena became very visible with incidents like io.net (https://cointelegraph.com/news/io-net-responds-to-gpu-metadata-attack), but this is not an isolated case, attacking these self-reporting mechanisms is very straight forward.
Without confidentiality of the execution you end up in a situation where you require trust in whoever is providing you with compute, which again, in an adversarial environment like the internet leads to all kind of attack vectors hindering real adoption.
By leveraging the HSM’s in mobile devices Acurast provides verifiability of hardware and execution as well as confidentiality of the execution. And it shows, traction becomes visible: since no fake or virtualized infrastructure can be on boarded onto Acurast you cannot fake the hardware and the community is very proud in showing their setups, a real DePIN:
What about enterprise grade performance?
A common misconception (and only reason why some software vendors are still able to sell hardware) is that if compute is not provided in the form factor of a pizza box server in a data center it won’t be performant or “enterprise grade”. To highlight this misconception Jan von der Assen, Researcher at the University of Zurich conducted an empirical performance analysis of serverless compute provided by Acurast vs. centralised cloud providers AWS, GCP and Azure (read the full paper here; https://arxiv.org/pdf/2404.08306).
To our own surprise the study showed that already the unoptimized version of the Acurast runtime was providing compute cores that were on average 34% faster than the next best which was AWS.
Another way to look at this is by comparing the benchmark scores of server CPUs (https://www.cpubenchmark.net/singleThread.html#server-thread) vs. mobile CPUs (https://www.cpubenchmark.net/singleThread.html#mobile-thread), you will notice mobile CPUs on the top of the list with similar scores compared to server CPUs. However what these benchmarks ignore is all the additional acceleration hardware packed in these devices.
So why are these devices equipped with the best hardware available? Well, because they are big business: Apple invests 33B in RD every year, the iPhone makes up for 2/3 of the entire company’s revenue (https://www.investopedia.com/articles/investing/022316/economics-iphone-aapl.asp).
Long-story short: Acurast leverages the hardware with the largest RD budget without needing to invest itself into hardware R&D.
What about Reliability?
Performance itself is of course not the only thing, reliability plays a big role as well. The reality of things is that just like servers, phones can and will have hardware failures, but when it comes to the reliability of the compute offered for both it’s seldom the hardware that will fail you, but more often than not the human factor that plays the bigger role and needs to be accounted for.
When running a decentralised compute network, which is permissionless and allows anyone to participate, you will have to run with the assumption that the compute may come and go, this is just as true if the compute is coming from a server as if it is coming from a phone.
How Acurast solved reliability and introduced super resilience is by applying two concepts: staked compute combined with a reputation system resulted from academic research (Whitepaper) and working with replicas (aka redundancy). As Acurast uses serverless computing for developers to deploy workloads, they can scale and specify how many processors should be running the workload making it super resilient and scalable.
What about costs?
Capital expenses
Bootstrapping compute is incredibly expensive. Finding exact numbers on the costs is tricky but sources like https://thundersaidenergy.com/downloads/data-centers-the-economics/#:~:text=Data give a 10M$-20M$ USD per MW. Using the average power consumption for a U42 rack of 7 kW (https://www.nlyte.com/blog/how-much-does-it-cost-to-power-one-rack-in-a-data-center/) we can infer about 6'000 U1 servers per MW (at a 100% efficiency).
If we take Hetzner for example: Their average U1 server offers 16 cores. Mobile devices on the other hand offer on average 8 cores. Making a simple calculation we’d need 2 phones to cover the CPU Cores requirements an average server offers. Now looking at the current state of the Acurast testnet, where the community managed to bootstrap 65'000 phones, one could infer we have the compute equivalent of 5.4 MW, which would’ve cost up to 108M$ USD in capital expenses. Obviously the numbers are debatable and data centers never run at 100% usage/efficiency, but it gives an idea of how scalable the Acurast model is.
What I found even more impressive is the fact that Acurast processors are already running in 130 countries, allowing to bootstrap compute in areas where infrastructure is not ready for data centers at an unprecedented speed.
Operating Expenses
Another important factor is obviously the operating expense of running this infrastructure. Because Acurast leverages phones, a device that never had the privilege of being constantly connected to a power supply, the power efficiency of the compute outperforms servers. Taking our 7kW rack example, per server we get a 170W consumption, a mobile device on the other hand consumes 5W under load, if we take 2 phones to have the same number of cores, that’s an impressive 17x more energy efficient compute, while idle the numbers look even more dramatic.
Again for both the capital and operating expenses you should take it with a grain of salt, as these are debatable, but it’s fair to say on average if you source your compute from phones rather than server you will end up with a lower cost structure and unprecedented scale, translating to more value for everyone!
Sustainability
One aspect that we haven’t highlighted in our messaging but became very clear was the fact that we give phones which are either unused or with malfunctions a second life. All those phones sitting in drawers housing hyper-engineered hardware can now be made productive and be used for useful compute. Not only is this a way more sustainable way to go about these devices compared to melting them for Gold, but it can be a true game changer in times of compute shortage.
Final Remarks
My goal with this article was to share what deeply excites me about Acurast — not just as its Co-Founder, but as someone who truly believes that we must rethink what compute means in a world demanding more decentralization, transparency, sustainability, and inclusion.
The journey we’ve been on since 2022 is nothing short of remarkable. What started as a bold vision — that real, verifiable, confidential, and high-performance compute could come from mobile phones — is now proven reality. With over 65,000 devices across 130+ countries, we are no longer imagining what a decentralized compute network could look like — we’re running it.
The very phones in our pockets — powered by the world’s most advanced hardware and backed by billions in R&D — are quietly outperforming traditional infrastructure, offering security guarantees that most cloud providers can only dream of. They are cost-efficient, energy-efficient, and globally distributed by nature. And perhaps most inspiring of all, they are being reused, repurposed, and given new life — transforming a global drawer of idle devices into a living, breathing compute network.
This is not just innovation — this is democratization.
Acurast shows that we don’t need more centralized data centers, more massive capital investments, or more complex trust assumptions to scale meaningful compute. What we need is to look around — to see the untapped potential already in our hands, and to enable it with the right technology.
So yes — it’s time to rethink compute. Not just as a technical challenge, but as a chance to build a future that is secure, inclusive, resilient, and sustainable by design.
Let’s continue to challenge the status quo. Let’s empower more people. Let’s turn potential into progress — together.
— Alessandro De Carli, Co-Founder of Acurast
Website | Twitter | Discord | Telegram | Documentation
