MARA × BlackRock × Kaspa: An Interesting Architecture Is Emerging If we connect three completely different developments, an interesting architectural picture emerges. But it is important to separate from the beginning: what is already happening, what is confirmed by documents, and what is still a hypothesis. Let’s start with MARA. MARA Holdings is no longer positioning itself only as a Bitcoin mining company. In 2026, the company explicitly describes its strategy around: power + land + compute. MARA is developing AI and HPC infrastructure at its power-rich sites, and through its partnership with Starwood plans initially around 1 GW of IT capacity, with the potential to scale to more than 2.5 GW. mara.com/posts/mara-announce… In its reporting, MARA also says that it is reallocating part of its infrastructure toward AI/HPC workloads, while maintaining Bitcoin mining as part of the model and using mining as a flexible load where economically appropriate. This is an important transition. Not: Bitcoin mining → disappears But rather: Energy → Compute → different high-intensity workloads. Bitcoin mining becomes one of the workloads that can be used to monetize energy while higher-value compute opportunities are being developed. MARA itself describes its model around the ability to allocate capacity between workloads depending on economics and demand. And this is where Kaspa enters the picture. Because MARA already has a direct history with Kaspa. In June 2024, Marathon Digital officially announced that it had begun mining KAS to diversify its digital-asset compute business. The company reported purchasing approximately 60 PH/s of Kaspa ASICs, of which 30 PH/s were already operating at its own facilities in Texas, with the remainder planned to be brought online later. At that point, MARA had already mined 93 million KAS. ir.mara.com/news-events/pres… And this is an interesting historical fact: MARA doesn’t just know about Kaspa. MARA has already actually used its mining infrastructure for Kaspa. But here it is very important not to make the wrong conclusion. MARA has not stated that it plans to turn Kaspa into an L2 for its AI compute. No such architecture has been publicly announced today. So from this point onward, we are talking about an architectural hypothesis, not a confirmed fact.

Now let’s add BlackRock. In its The Machine-Native Economy research, BlackRock describes another piece of the same potential system: AI agents → payments → digital assets → compute → programmable blockchain settlement. blackrock.com/us/individual/… This creates a separation of roles. AI can be an economic agent. Compute becomes an economic resource. Digital assets become machine-native money. And blockchain can perform the function of programmable settlement. And now we can look at MARA and Kaspa through this lens. MARA Power + land + data centers + compute Kaspa PoW + BlockDAG + fast settlement + UTXO + programmable L1 AI layer Models + agents + APIs + compute Payment layer Stablecoins / tokenized assets / KAS / machine-payment protocols This no longer looks like one company or one blockchain. It looks like a stack. And here an interesting question emerges: what if physical compute infrastructure and programmable settlement infrastructure develop as different layers of the same machine economy? Conceptually: AI AGENTS ↓ AI / HPC ↓ GPU / DATA / COMPUTE ↓ MARA POWER + LAND + DATA CENTERS ↓ APPLICATION / COORDINATION LAYER ↓ KASPA L1 PoW + BlockDAG + UTXO + COVENANTS ↓ VERIFIABLE SETTLEMENT This is not a declared MARA/Kaspa architecture. It is a model of how already-existing directions could theoretically connect. And this is exactly where it is important not to confuse the layers. MARA is not becoming a “GPU.” MARA is developing infrastructure on which GPU/HPC/AI workloads can run. Kaspa is not becoming an L2. Kaspa is an L1. If specialized L2s or application protocols for the machine economy ever emerge, they could potentially be built on top of the L1, rather than turning Kaspa L1 itself into an AI execution environment. Now it becomes clearer why the previous BlackRock analysis was important. BlackRock says: AI → machine-native intelligence Digital assets → machine-native money Compute → economic resource Blockchain → programmable settlement MARA is developing: Power → data centers → AI/HPC compute Kaspa already has: PoW → BlockDAG → fast settlement → programmable UTXO And this is where a potential architectural connection emerges: Physical infrastructure ↓ Compute ↓ AI agents ↓ Machine payments ↓ Programmable settlement ↓ Kaspa L1 But there is another interesting fact. BlackRock is a major institutional holder of MARA shares. According to data as of March 31, 2026, BlackRock held approximately 59.8 million MARA shares, or 15.47% of the company. But here we cannot make the logical leap: BlackRock owns MARA → MARA mined Kaspa → BlackRock invested in Kaspa. That would be incorrect. BlackRock owns MARA shares, meaning a stake in a public company. That does not mean ownership of KAS. And it does not mean that BlackRock has made a bet on Kaspa. Therefore, it is more accurate to talk about three intersecting directions, rather than one confirmed deal: BlackRock → machine-native economy MARA → power + digital infrastructure + AI/HPC Kaspa → PoW BlockDAG + programmable settlement

And this is where it gets really interesting. If BlackRock describes an economy in which: AI agents buy data + compute + services and pay for them through machine-native payment rails then the physical infrastructure still has to exist somewhere in the real world. You need: electricity land data centers GPU networking cooling compute And this is exactly the layer MARA is actively developing. MARA has already secured access to a site in Texas with potential capacity of up to 2 GW, intended for large-scale digital infrastructure, including HPC and flexible compute workloads. So: BlackRock is describing a new economic model. MARA is building part of the physical infrastructure for the compute economy. Kaspa is developing a programmable PoW settlement layer. These are three completely different levels. And that is exactly why it is interesting to look at them together. Now the most important part: we should not claim that this system already exists. It does not yet exist in this complete form. There is no public confirmation that: MARA is building AI infrastructure specifically for Kaspa. There is no public confirmation that Kaspa will become MARA’s settlement layer. There is no public confirmation that BlackRock considers Kaspa part of this architecture. And there is no public confirmation that future MARA AI/HPC workloads will be settled through KAS. All of those claims would be speculation. But the individual components already exist. BlackRock → machine-native economy MARA → energy-backed compute infrastructure Kaspa → high-throughput PoW settlement And that is exactly why the architectural hypothesis is more interesting than simply: BlackRock + MARA + Kaspa. In its cleanest form, the potential stack can be represented like this: AI AGENTS ↓ MCP / A2A ↓ AI / HPC ↓ GPU / COMPUTE ↓ MARA POWER + LAND + DATA CENTERS ↓ APPLICATION / PAYMENT PROTOCOLS ↓ STABLECOIN / TOKENIZED ASSET / KAS ↓ KASPA L1 ↓ UTXO + COVENANTS ↓ BLOCKDAG + GHOSTDAG ↓ PoW ↓ VERIFIABLE SETTLEMENT And this is where the principle of separation itself becomes particularly interesting. MARA does not need to be a blockchain. Kaspa does not need to be an AI model. AI does not need to live inside the L1. Each layer performs its own job. Physical infrastructure provides compute. AI makes decisions. Payment protocols transfer economic value. The L1 verifies rules and settlement. This kind of modularity may ultimately be much more important than trying to make one blockchain do everything. So my careful conclusion would be: BlackRock describes the Machine-Native Economy. MARA is building physical energy + compute infrastructure that can serve AI/HPC workloads. Kaspa is building fast, permissionless, PoW-based programmable L1 settlement infrastructure. Today, there is no announced joint project between them. But architecturally, they could occupy completely different levels of the same future stack: MARA → physical compute / energy AI → intelligence Protocols → coordination & payments Kaspa → settlement And if real integration ever emerges between these layers, it would no longer be simply a story about cryptocurrency. It would be the connection of: ENERGY → COMPUTE → INTELLIGENCE → MONEY → SETTLEMENT And this is exactly where the intersection of BlackRock × MARA × Kaspa becomes genuinely interesting. Not because they have already announced one common system. But because three different directions are already developing around different parts of one potential machine-native economy.