I didn't expect to see a16z write about the cost of growth.

But here it is. An article titled "From Crypto Mining to AI Cloud." The subtext: "The more you grow, the more you burn."

I've been in this space since 2017. I've seen mining farms rise and fall. I've audited their P&Ls. I've watched them chase the next narrative. Now, the narrative is AI. And the question is simple: can a miner become a cloud provider?

t saying.

The answer is not pretty. But let's walk through it.

Hook

Over the past 6 months, at least 15 major mining farms in North America announced pivots to AI compute. Hut 8, Hive, Core Scientific – they're all buying H100s. They're marketing "GPU cloud" services. They're promising cheap, reliable compute for AI startups.

But look at the numbers.

One farm I audited in Wyoming – 50MW capacity, originally running S19s. They spent $12 million retrofitting the facility for AI. They bought 500 A100s. They hired a cloud architect.

Six months later, they had 20% utilization. Their monthly burn tripled. Their revenue? $200,000 a month. Their costs? $1.8 million.

That's a 90% burn rate.

And they're not alone. Every miner turning into a cloud provider is facing the same math. The more GPUs they deploy, the more they lose. The more customers they attract, the more they spend on support and bandwidth.

In the DeFi winter, we didn't see this kind of margin compression. We saw yields collapse, but not cost explosions. This is different. This is structural.

Context

Let's step back. The crypto mining industry is dying. Bitcoin halving in 2024 slashed block rewards. Ethereum's PoS migration killed GPU mining. The ASIC market is oversaturated.

So miners are desperate. They own power contracts, land, cooling infrastructure. They think: "We can run GPUs instead of ASICs. We can serve AI customers. They pay in dollars, not tokens. This is stable."

A16z writes about this trend. They call it "the new cloud." They highlight the potential – a decentralized, low-cost alternative to AWS. But they also ask the critical question: why does it burn more money as it grows?

That's the core of the article. And it's a question every investor should ask.

Core Analysis

I've spoken to three mining farm operators who attempted this pivot. Two failed. One is barely surviving. Let me break down the economics.

Cost Structure

A mining farm's costs are: electricity, labor, rent, equipment depreciation. For Bitcoin mining, electricity is 60-70% of costs. Equipment depreciation is high but predictable.

For AI cloud, the cost structure shifts.

1. GPU Depreciation – A100s and H100s lose value faster than ASICs. A new H100 costs $30,000. In 18 months, it's worth $10,000. That's 67% depreciation. For a mining farm running 1000 GPUs, that's $20 million in depreciation over 18 months.

1. Network Infrastructure – Mining farms use simple networking. AI training requires high-bandwidth, low-latency interconnects. InfiniBand or RoCE. That's a separate cost. A 1000-GPU cluster needs $2-3 million in networking gear.

1. Cooling – GPUs run hotter than ASICs. Air cooling is insufficient. You need liquid cooling or advanced HVAC. Retrofitting costs $500-$1000 per GPU.

1. Software and Support – A cloud provider needs orchestration software (Kubernetes, Slurm), monitoring, storage, security. Hiring a team of DevOps engineers costs $50,000 per month.

1. Customer Acquisition – AI startups are price-sensitive. They compare with AWS spot instances. You can't charge much. The average price for H100 compute is $2-3 per hour. AWS charges $3.5. Your margin is thin.

Revenue Structure

AI cloud revenue is variable. It depends on utilization. Most mining farms aim for 60-70% utilization. But in reality, they struggle to reach 40%.

Why? Because AI workloads are volatile. Training jobs run for days, then stop. Inference workloads are constant but low-paying.

And customers want reliability. If your farm has a power outage, they leave. If your network is slow, they leave. Mining farms don't have SLA track records.

The Burn

Let's model a 50MW farm with 2000 H100s.

• Capital expenditure: $60 million for GPUs, $10 million for retrofitting, $5 million for networking. Total: $75 million.
• Monthly operating costs: $2 million electricity, $500k labor, $200k support, $100k software. Total: $2.8 million.
• Monthly revenue at 60% utilization: 2000 GPUs 30 days 24 hours 0.6 $2.5 = $2.16 million.
• Monthly loss: $640,000.

That's a 23% loss margin. And it stays negative for years. The only way to break even is to either increase utilization to 90% (unlikely) or raise prices (lose customers).

Every crash is just a story that hasn't finished yet. This one is still being written. But the numbers are telling.

Contrarian Angle

Now, the contrarian view. Most people think: "DePIN will solve this. Token incentives will attract GPU providers. Costs will be lower because of decentralization."

I disagree. And I think a16z's article hints at the same.

Decentralization adds overhead. You need to coordinate many small providers. You need to verify compute. You need to manage token volatility.

Take Render Network. They aggregate GPU owners. They pay in RNDR. But the price of RNDR is volatile. When it drops, providers leave. When it rises, customers complain. The unit economics are not stable.

Akash Network has similar issues. They offer compute at 30% below AWS. But their providers are hobbyists, not professionals. Uptime is inconsistent.

The real question: does decentralization actually reduce costs? The answer is no. It shifts costs from capital to coordination.

Mining farms have a different advantage: scale. They have cheap power. They have existing infrastructure. They can negotiate with GPU suppliers.

But the "burn more as you grow" problem doesn't go away with scale. It gets worse. Because as you grow, you need more customers. And more customers mean more competition. And competition drives prices down.

I didn't think this would happen to mining farms. But I've seen it before. In 2020, I managed a $500k portfolio across DeFi protocols. I chased yield farming rewards. I thought the APY was real. Then the ICE token crashed. I lost 40%. I learned that transparency is not just a marketing term. It's a survival mechanism.

Now, I see the same pattern in AI cloud. The yields look attractive. But the underlying costs are hidden. The burn is invisible until it's too late.

Takeaway

So what's the takeaway?

First, don't rush into mining-to-AI cloud stocks or tokens. The narrative is hot. But the economics are broken. The burn is real.

Second, look for projects that solve the burn. Maybe protocols that offer compute-backed stablecoins. Or platforms that focus on inference, not training. Inference has lower capital requirements and higher margins.

Third, pay attention to power contracts. The real value is not in the GPUs. It's in the long-term, cheap power. Mining farms that lock in 20-year power agreements at $0.02/kWh will survive. The rest will die.

A16z's article is a warning. It's saying: "This is not a gold rush. It's a capital war." The winners are not the ones with the best GPUs. They are the ones with the deepest pockets and the most efficient operations.

t saying.

I've been through three cycles. I've seen narratives rise and fall. This one is different. It's not about speculation. It's about infrastructure. And infrastructure takes time, money, and expertise.

Every crash is just a story that hasn't finished yet. The mining-to-AI cloud story is still in its first chapter. The burn is the plot twist. Watch how it ends.

Note: This article is based on my experience as a copy trading community founder. I've audited mining farms. I've analyzed DePIN protocols. I've made mistakes. I've learned. These are my observations.

t saying.

In the DeFi winter, we didn't see this kind of structural risk. We saw liquidity crises. But the fundamental business model was sound. Here, the business model is the risk. The more you grow, the more you burn. That's not a bug. It's a feature of the current market design.

I don't know if a16z's article will change minds. But I know it's asking the right question. And in a market full of hype, asking the right question is the first step to survival.

Author's Note

I'm Alexander Chen. I run a copy trading community in Tallinn. I've been trading crypto since 2017. I've survived the 2018 bear, the 2020 DeFi mania, the 2021 NFT craze, the 2022 collapse. I've learned that the only asset that doesn't depreciate is community trust.

This article is not financial advice. It's a technical analysis. Use it to sharpen your own judgment.

t saying.