Firebird plans to deploy 250 MW of NVIDIA AI infrastructure in Armenia and Kazakhstan over the next 12 months. That validates expanding demand for compute. It does not prove that the capital funding this expansion will earn attractive returns.
National AI Capacity Broadens the Buildout
Firebird intends to serve researchers, startups, industries, and governments. A speaker discussing the plan described AI factories as infrastructure that turns energy into intelligence.
The implication extends beyond technology companies selling software. Countries may not want all of their intelligence capacity imported through global cloud services. If domestic compute becomes strategically important, governments and industries could join private companies in funding AI infrastructure.
Firebird’s plan suggests that this transition is already reaching Armenia and Kazakhstan. But strategically useful infrastructure is not necessarily a good investment at every price.
The Market Is Pricing Growth Across the Stack
A screen attributed to @KoyfinCharts projects 2027 revenue growth of 243.1% for IREN, 190.8% for Nebius, and 96.5% for CoreWeave.
The broader group spans Nvidia, AMD, Broadcom, Micron, TSMC, ASML, Applied Materials, Arista Networks, Seagate, Western Digital, Arm, and Oracle. These companies cover compute, chips, semiconductor equipment, networking, storage, and supporting infrastructure.
That concentration suggests investors are valuing AI as a capital buildout across the technology stack, not merely as a software category.
Yet revenue projections show only where spending may go. They do not establish what that growth is worth, how much capital it will require, or whether the resulting cash returns will justify current infrastructure valuations.
Capacity Validates Demand, Not Returns
A speaker at @value-investing called the US market the biggest financial bubble in history, alleged circular AI financing, and argued that “there is no return on cash flows.” No supporting figures accompanied those sweeping claims, but they point toward the right test.
Capacity announcements validate equipment demand. They do not establish attractive returns on the capital funding that capacity.
Rapid revenue growth can still produce weak economics if it requires equally aggressive spending on chips, power, data centers, and financing. The buildout thesis and the valuation thesis are not the same. Firebird’s 250 MW plan supports the first. The second still requires evidence of durable cash generation.
Open Source Could Separate Usage From Value Capture
A speaker at @geopoliticaleconomyreport placed AI competition within a broader argument about China’s technological rise. The speaker cited China’s roughly 20% share of the global economy versus less than 15% for the United States, referred to an ASPI study finding China leads in 66 of 74 advanced technologies, and claimed that Moonshot’s open-source Kimi K3 outperforms OpenAI and Anthropic models on several benchmarks.
Those are the speaker’s claims. The investment risk they raise is that rapid open-source progress could simultaneously increase AI adoption and weaken platform pricing power.
The world can consume more AI without returns being distributed evenly across the stack. The buildout is increasingly difficult to dispute. The unresolved investment question is who converts expanding compute capacity into durable cash flow.