xAI is looking more like a datacentre REIT than a frontier lab

xAI’s shifting role: frontier lab vs. datacenter landlord

• Many commenters say Grok is no longer frontier‑leading; xAI looks more like a compute lessor than a cutting‑edge AI lab.
• Renting out Colossus GPU capacity to Google and Anthropic may quickly recoup capex (18 months is cited), turning stranded capacity into revenue.
• Some see this as rational: hard to catch OpenAI/Anthropic’s revenue‑>compute‑>better‑model flywheel, so sell shovels instead of digging for gold.
• Others argue this is effectively abandoning the AGI race and should be valued like low‑multiple infrastructure, not like a frontier lab.

Circular deals, valuations, and bubble risk

• Strong concern about “circular” or opaque financing: big tech investing in AI labs and datacenters, then booking huge paper gains and revenue from each other.
• Comparisons are drawn to dot‑com and Enron‑style dynamics: real underlying tech but potentially unsustainable stock valuations and write‑off risk.
• Several worry pensions and index funds will end up holding the bag if/when AI valuations re‑rate; expectations range from “dot‑com‑scale bust” to “bigger than 2008” to “music never stops.”

GPU scarcity, economics, and depreciation

• Consensus: massive GPU and power shortages; “no dark GPUs” once deployed. Older chips (A100, H100) still rent at high prices and may even appreciate temporarily.
• Debate over how fast compute actually depreciates: technologically fast vs. market‑driven scarcity keeping prices high for years.
• Some claim inference APIs are clearly profitable at current prices; others counter that full costs (training, capex) are not reflected and profitability claims are unproven until post‑IPO filings.

Orbital datacenters and long‑term strategy

• A faction sees xAI/SpaceX positioning for LEO datacenters: Starship plus Starlink plus Colossus‑like clusters as a unique, long‑term moat.
• Skeptics question the economics (launch cost, radiators, hardware lifetime) and note this only works under rosy assumptions about continued AI demand and terrestrial permitting delays.

Environmental, regulatory, and social issues

• Colossus 1’s rapid build is credited to aggressive rule‑bending, especially on on‑site gas turbines; critics see heavy local pollution and precedent for anti‑datacenter backlash.
• Some flag lawsuits and potential injunction risk; others view this as industry‑wide behavior (on‑site generation, water use, secrecy) rather than uniquely xAI.

Model quality and usage patterns

• Users comparing Grok, ChatGPT, Claude, Gemini report Grok feels 1–2 generations behind in general reasoning and coding, but:
  • Strong on current events via X integration.
  • Less sycophantic and more willing to handle sensitive or legally tricky topics.
• Broader concern that all LLMs are still somewhat sycophantic or politically biased, and that “therapy” or life decisions via LLMs may be risky.