← Track
Analysis · Crypto assets

AKT: cheap compute
in the middle of a shortage

The world is starved of GPUs. Akash sells compute far cheaper than the hyperscalers. Its token trades ~90% below its high. Somewhere in that gap sits either an opportunity or a trap.

Open By Ismail Rustamov AKT Published · 28 May 2026 Price ~$0.75 · spot Original post on Telegram →

The world is starved of GPUs. Akash sells compute up to ~85% cheaper than AWS. And its token trades about 90% below its all-time high. Somewhere in that gap is either an opportunity or a trap — and the whole question is whether a decentralized network can take share from the giants who bought up all the cards in the first place.

Where we are now

AKT trades around $0.75 after a volatile May1. The context is sobering: that is roughly 90% below its all-time high of $8.08, set back in April 20211. Market cap is only about $220M — somewhere around 170th among crypto assets1. This is a small, by-market-standards illiquid asset that jumped on the AI narrative and then pulled back.

So this is not a fashionable story at its peak. It is a deeply drawn-down infrastructure token the market has nearly forgotten.

The bull case

Akash's strength is that the product is real and working — it does not live on a roadmap.

Akash is a decentralized cloud-compute marketplace: owners of servers, datacenters and GPU rigs rent capacity to developers through a reverse auction, at prices up to ~85% below AWS, Google Cloud and Azure — an H100 runs about $1.33 per hour on Akash against roughly $3.93 on AWS2. The positioning fits the moment exactly: the network casts itself as a decentralized GPU cloud for training and inference of AI models, right when GPUs are in acute global shortage2.

There is genuine development underneath. The Burn-Mint Equilibrium (BME) model, live since the March 2026 hard fork, turns every dollar of compute paid through Akash into a market buy and a permanent burn of AKT — by the end of March some 53,520 AKT had already been burned3. And AKT is not an empty governance token: it is used for staking, governance and paying for compute inside the network2.

In one line: if the GPU shortage is here to stay, cheap decentralized compute is a logical pressure valve — and Akash is one of the few networks where it already works.

The bear case

Now the uncomfortable question — and it is stronger than any bull argument.

Can a decentralized network actually take share from the hyperscalers? GPUs are scarce precisely because AWS, Microsoft and the cloud giants bought them — the very players Akash is up against. Demand for cheap compute is enormous, but large AI teams want reliability, SLAs and enterprise-grade support that a decentralized marketplace struggles to provide. This is the perennial DePIN problem: a great story, but market share that does not compound.

And the metrics are mixed, not accelerating. Akash posted a record ~$5M in compute spend in Q1 2026 on enterprise AI demand — but in the same quarter lease (compute) revenue fell about 45% quarter-on-quarter and average GPU usage dropped sharply, leaving utilization near 34%56. The headline grows while the engine underneath sputters.

Then size. A ~$220M cap makes the token volatile and hostage to AI sentiment: when the theme cools, assets like this fall faster than the market. Competition in the niche is fierce — Render alone is several times larger7. And a meaningful slice of the max supply is still to come: about 294M of 388.5M AKT are in circulation, so dilution lies ahead1.

Akash has to prove not the idea but the traction: real growth in network load, not just a scarcity narrative.

Thesis

Akash has the rare thing in DePIN — a working product — and is still missing the rarer one: proof that the GPU-scarcity narrative turns into network traction.

The cheap-compute story is real, and the BME burn gives AKT genuine utility for the first time. But the numbers underneath are mixed, the token is small, illiquid and still diluting, and it is fighting the very hyperscalers that hoard the cards. So over a Q3–Q4 2026 horizon we expect AKT to stay range-bound — roughly $0.50–0.90 — and we lean to the lower half of that band. This is a "show me traction" call, not a bet on a re-rating.

Our bet runs one way: narrative without proven traction keeps AKT rangebound and heavy. We are not betting that decentralized compute fails, and we are not betting on a breakout. Conviction is low — about 50% — and we say why: the product is real, the traction is not, yet.

What would prove us wrong

One condition, not two. Our thesis is that the narrative does not yet convert into traction — so there is exactly one thing that would prove us wrong.

The condition that closes the thesis
A breakout backed by real usage

A reclaim and hold above ~$0.90 — but only if it comes with rising real network usage: compute spend, GPU utilization and provider revenue accelerating together, not price alone. That would mean Akash is finally converting the scarcity narrative into demand, and we were too cautious.

What does not change it — and what still counts as a miss

A loss of the $0.65 support without any pickup in those metrics does not contradict the thesis — it confirms it. But the claim we published is the $0.50–0.90 corridor: a sustained break below $0.50 breaks that corridor, and we book it as a miss, not as a win for the bearish lean. Either way, the tell is the network, not the candle.

We watch the real metrics — GPU utilization and provider revenue — over the price. For an asset like this, they are the honest signal4.

When a condition triggers, we book the result and publish the piece. As always.

Interest disclosure No position · as of 28 May 2026 Policy →

This is analysis, not investment advice. Do your own research. The thesis is not changed after the fact.

Updates & Outcome

Status: Open  — no updates yet. This thesis stands exactly as published. When a condition triggers, a dated entry is added below; nothing above is edited or removed.

Sources

Figures are as of publication, 28 May 2026, and may differ from current market data. Links point to the primary data sources.

1 CoinGecko — Akash Network (AKT) · price (~$0.75), all-time high $8.08 (April 2021), market cap ~$220M, circulating ~294.4M of a 388.5M max supply
2 Akash Network — Decentralized Compute Marketplace · reverse-auction marketplace, pricing up to ~85% below AWS / Google Cloud / Azure (H100 ~$1.33/hr vs ~$3.93 on AWS), GPU cloud for AI training and inference, AKT used for staking, governance and payments
3 Akash Network — Burn-Mint Equilibrium (AEP-76) · BME hard fork (live 23 March 2026): every dollar of compute triggers a market buy and permanent burn of AKT; 53,520 AKT burned by 31 March
4 Akash Stats — Network Metrics · live network metrics: GPU utilization, provider revenue, active leases and compute spend
5 CryptoDaily — AKT After the AI Rally: Can Decentralized Compute Prove Real Utilization? · Q1 2026 mixed picture: record compute spend alongside lease revenue down ~45% QoQ and GPU utilization near 34%
6 Messari — State of Akash · network revenue, compute spend and utilization context, including the record ~$5M quarterly compute spend
7 Messari — Render Network · DePIN GPU competition: Render is several times larger than Akash by market cap

The thesis is open. The record is set.

This piece stays in the archive unchanged — with its publication date. If the thesis fails, the result will be booked here too.

Open the ledger Telegram