Firmus raised $2 billion at a $10.5 billion valuation. That is a lot of zeros. But ask me what they actually own. Nothing you can verify. No customer contracts. No GPU purchase orders. No team bios. Just a press release and a narrative.
That narrative is simple: Bitcoin miner turns into AI infrastructure company. It is a hot story in 2025. Every miner with a substation and a cooling tower wants to be the next CoreWeave. The market is paying a premium for the pivot. But the premium is a bet on execution, not on reality.
I have been in this industry long enough to smell the difference between a signal and a noise. In 2017, I audited Uniswap v1 smart contracts on testnet. I found an integer overflow in the liquidity pool logic before mainnet launch. The code was hiding a critical flaw. The same pattern is repeating here. The narrative is the code. The flaw is hidden in the details.
—
Context: The Miner-to-AI Playbook
The transition from Bitcoin mining to AI data centers is not new. Hut 8, Core Scientific, Iris Energy – they all walked this path. The logic is simple: miners own power infrastructure. Substations, transformers, cooling systems, land. AI data centers need the same. The difference is the hardware. Instead of ASIC miners, you need GPU clusters. NVIDIA H100s, H200s, or the new Blackwell. The electricity consumption is similar. The network topology is not.
Mining is a simple compute model. ASICs hash. That is it. No networking between nodes. No low-latency interconnects. AI training requires RDMA, InfiniBand, or at least high-bandwidth Ethernet. The cooling requirements are different. Air cooling might work for a mining farm. For a GPU cluster processing massive datasets, you need liquid cooling, direct-to-chip or immersion.
The capital expenditure is orders of magnitude higher. A single H100 GPU costs around $30,000. A cluster of 10,000 GPUs is $300 million just for the silicon. The data center buildout adds another $100-200 million. The power connection agreement can take years to secure. The lead time for NVIDIA GPUs is currently 12-18 months for new orders.
So when Firmus announces a $2 billion raise and a $10.5 billion valuation, the first question is not "what is the vision?" It is "what is the timeline?" Because the capital is large, but the execution window is tight.
—
Core: The Technical Anatomy of the Transition
Let me break down what a $2 billion raise actually buys. If the entire amount is used for GPU procurement, at today's prices, you get roughly 66,000 H100 GPUs. That is a 66 petaflops cluster at FP16. Respectable but not revolutionary. The largest AI clusters are now pushing 100,000 GPUs. CoreWeave and Microsoft have clusters in the 300,000 range.
But the raise is probably not all equity. The article does not specify the debt vs equity split. If it is a convertible note or high-yield debt, the cost of capital eats into margins. At current interest rates, a $2 billion debt at 8% carries $160 million in annual interest payments. That is before you power a single GPU.
And the power costs. The article emphasizes sustainable energy. That is a buzzword, but the real metric is the PPA rate. A miner with a fixed 10-year power purchase agreement at $0.03/kWh has a massive advantage over a miner paying spot market rates. But the article does not disclose the PPA. So we assume nothing.
During the Terra/LUNA collapse in 2022, I manually executed a liquidity exit from Curve Finance pools. I saved $2.4 million in capital before the bridge hack. The lesson was simple: verify the assumptions. The root cause of the collapse was a stale price feed, not a bad code. The same applies here. The key assumption is that Firmus can secure the GPU supply and the data center buildout within a reasonable time frame. That assumption is unverified.
The code does not lie, but it does hide. The hiding here is the lack of disclosure. No customer contracts. No GPU orders. No team background. The only thing we have is a valuation. That valuation is a narrative. And narratives are fragile.
—
Contrarian: The Smart Money Is Not Buying the Story
The market is euphoric about miner-to-AI. But the smart money is already shorting the pivot. Look at the stock performance of Hut 8 and Core Scientific. They rallied on AI announcements, then corrected when the revenue didn't materialize. The market is treating the transformation as a fait accompli. It is not.
The real value of a miner is the power contract and the land. The transition to AI requires a completely different operational skill set. You need people who understand high-performance computing, network architecture, and enterprise sales. A mining team knows how to run ASICs. That is a different discipline.
Alpha hides in the friction of liquidity. The friction here is the GPU supply chain. The market is pricing in a smooth transition. But the reality is that GPU supply is constrained. NVIDIA allocates to the highest bidders and the most reliable partners. A miner without a track record in AI will not get priority. The lead time is 18 months. That is a lot of time for a competitor to build a better facility.
And the sustainable energy angle? It is a nice story, but it does not solve the core problem. The marginal cost of electricity for a miner is often the second highest expense after capital. For an AI data center, the power cost is still high, but the bigger cost is the GPU depreciation. A GPU loses value fast. The economics of AI computing are a race against time. The faster you fill the cluster, the better the return. If the cluster is idle for even three months, the IRR drops significantly.
Volatility is the tax on uncertainty. The uncertainty here is high. The valuation is based on a future that has not yet been built. The contrarian view is that this is a bubble within a bubble. The miner-to-AI narrative is a way to raise capital before the music stops. The music might stop when the first major miner fails to deliver on its AI promise.
—
Takeaway: The Only Signal That Matters
In the next 18 months, the critical signal is the first customer announcement. If Firmus signs a contract with a hyperscaler like Microsoft, Google, or a major AI lab, the valuation is real. If they announce a partnership with a cloud provider, it is a start. But if the only news is more funding rounds, the narrative is running on empty.
Check the gas, then check the truth. The gas here is the energy cost. The truth is the PPA rate. If the power is cheap and locked, the miner has a moat. If the power is spot market, the moat is a mirage.
Precision is the only hedge against chaos. The market is chaotic right now. The miner-to-AI narrative is a wave. The question is whether Firmus can ride it or will be wiped out by the undertow. I will be watching the GPU supply chain and the customer contracts. That is where the real signal is.
Yield is never free; it is rented. The AI yield is rented from the power infrastructure and the GPU supply chain. The rent is due every month. The market is paying the rent now, but the renewal is not guaranteed.
Backtest the assumption, not just the data. The assumption is that a miner can become an AI infrastructure company. The data is the $2 billion raise. The backtest is the execution. I have seen this pattern before. The ICO boom was full of narratives and limited execution. The ones that survived were the ones that delivered. The rest are ghost towns.
Firmus has the capital. Now it needs the plan. The plan is not in the press release. It is in the details. And the details are missing.
Watch the next 12 months. If you see a large GPU order from Firmus, the story is real. If you see a partnership with a data center builder, the story is moving. If you see nothing, the narrative is a trick.
That is the code. And the code does not lie. It hides.
