MACRO — knowledge base

Overview

Jim Liu frames the AI buildout as an extreme productivity revolution exceeding industrialization, financialization, and digitalization. It channels liquidity into assets via concentrated profits in frontier labs, hyperscalers, neoclouds, power, and silicon layers, enabling sustained expansion with limited broad consumer inflation. AI diffuses rapidly on existing digital infrastructure, extending the cycle beyond the dot-com boom. Complementary views treat AI, Bitcoin, and power/data centers as core megatrends: equities at all-time highs while Bitcoin undergoes a mid-cycle correction now ranging since the ~58k June/July 2026 sweep (mid-to-high $70ks as of mid-Sep 2026) with heavy 82k/50-week MA resistance; last-cycle (2017-era) alts failed to reclaim ATHs despite QE/ZIRP so even a Trump money printer/ZIRP would not put recycled tickers into price discovery—prefer multi-market wealth-creation accounts over single-asset bull-market heroes; Fed hikes/cuts viewed as a nothing-burger versus historically more negative BoJ hikes (caveated by June 2026 exception); financialization and institutional product builds (vs. pure fundamentalism) are positioned to cross Bitcoin’s adoption chasm; investor edge centers on long duration, equanimity, and resisting short-term instruments/margin that increase exploitability. Within power/data-center layers, settled baseload and multi-year Batch Zero inclusions create scarce moats; operators holding contract duration open can capture higher 2027 GPU/ARR pricing amid physical-build constraints, while long-end yields and oil act as near-term momentum dampeners.

Key facts & figures

A100 GPUs launched May 2020; by 2026 entering seventh year of use ▶ 00:14:46.
Only three companies (TSMC, Intel, Samsung) operate fabs for most advanced semiconductor nodes ▶ 00:10:51.
HBM production dominated by SK Hynix, Samsung, and Micron ▶ 00:10:05.
Intel lost semiconductor manufacturing lead under CEO Brian Krzanich (2013-2018) due to diversification into non-core areas ▶ 00:11:06.
AI expected to drive deflationary pressure allowing greater government liquidity expansion ▶ 00:02:59.
OpenRouter represents significant share of non-hyperscaler, non-OpenAI/Anthropic open-source inference tokens ▶ 00:06:09.
Equity market at all-time highs while Bitcoin in a mid-cycle correction (drawdown from Oct 2025 ATH) ▶ 00:21:40.
Nvidia Q2 FY2027 earnings scheduled ~10 days after mid-Aug 2026 (Aug 26, 2026) ▶ 00:35:48.
Nebius and CoreWeave posted strong Q2 2026 earnings ~Aug 11–12 with large positive moves on AI demand ▶ 00:34:09.
Leopold Aschenbrenner’s Situational Awareness fund and other overleveraged players (incl. large South Korean entities) faced liquidation/margin-driven unwinds in the late-July 2026 AI correction ▶ 00:35:03.
Bears including Michael Burry, Jim Chanos, and Ed Zitron have called an AI top/end since late 2024 ▶ 00:36:52.
Clarity Act widely viewed as a major potential regulatory catalyst for Bitcoin/crypto ▶ 00:24:23.
California top state income tax rate 13.3% on high incomes ▶ 00:32:56.
NVIDIA announced collaboration with IREN and Australian partners (incl. Firmus) for up to 2 GW AI infrastructure buildout by 2027; IREN applying blueprint to Bundey campus ▶ 00:01:21.
Cipher developing natural-gas laterals/pipelines supporting up to 2.5 GW behind-the-meter/on-site generation for HPC/data centers ▶ 00:04:50.
ERCOT Batch Zero applies to qualifying large-load projects ≥75 MW ▶ 00:10:19.
IREN Sweetwater 1 (~1,400 MW) and Sweetwater 2 (~600 MW) received conditional baseload designation in ERCOT Batch Zero; additional large-scale projects in IREN’s broader pipeline also included ▶ 00:30:40.
IREN targeting ~480 MW AI Cloud capacity by end-2026 and ~1.2 GW (1,210 MW) by end-2027 ▶ 00:18:40.
IREN Bundey (Bundy) South Australia campus planned for up to 800 MW; energization expected to commence from 2028 (not 2027) ▶ 00:13:54.
Directional Goldman fireside math: ~500 MW at ~$25M/MW → ~$12.5B incremental ARR; contract length drives actual guidance (unverifiable exact quote) ▶ 00:17:06.
U.S. 30-year Treasury yield recently hit ~5.2–5.3% ▶ 00:39:29.
IREN shares touched ~$49 (high ~$49.29 on 2026-09-08) and held near $45 support ▶ 00:39:56.
Firmus facing community opposition/backlash in Tasmania on AI data-center projects ▶ 00:16:20.
From-scratch 500 MW is a ~2032 problem; batch-zero power is a multi-year moat. Peak IT build cadence estimated ~850 MW/year around 2028–29. Childress: Horizon 3 CDUs installing, H3 mid-Q4 revenue expected, H4 likely late Dec with grace into Q1.
Bitcoin ranging since last ~58k sweep (June/July 2026 low), mid-to-high $70ks mid-Sep 2026, major resistance ~82k coinciding with 50-week MA ▶ 00:08:59.
Bank of Japan widely expected to hike at Sep 17-18 2026 meeting (to 1.25% from 1%), priced in ▶ 00:08:12.
2017-era altcoins (e.g. NEO, Lisk ATH near $35 early 2018) failed to reclaim prior ATHs in later cycles despite QE/ZIRP: accurate ▶ 00:01:11.
Claim AVAX/LINK reached 75-80% of prior ATHs last cycle: inaccurate (AVAX Nov 2021 ATH ~$146, 2024 high ~$65 ~45%; LINK May 2021 ATH $52.88, later highs ~$30-31 ~57%) ▶ 00:00:50.
Circle Arc mainnet launched Sep 16 2026 messy: bridge opened early overnight ahead of schedule, RPCs/platforms failing, OTC stables ~2x, copycat rugs: accurate ▶ 00:02:56.
Dansec missed at least one ~58,000% extreme early runner after premature bridge open ▶ 00:03:08.
Sui meme/DJN attention short-lived around Sep-Oct 2024 then faded; chain characterized as ghost town: misleading (maintained DeFi TVL, transactions, development) ▶ 00:04:16.
Historical BoJ hikes “always” negative for Bitcoin (only one short-lived positive exception): misleading (2024-25 hikes typically 18-32% BTC drawdowns; June 2026 hike little reaction as priced in, slight uptick) ▶ 00:08:26.

Thesis & bull case

AI accelerates diffusion on existing digital infrastructure unlike dot-com era, producing higher utilization via Jevons paradox and enabling longer buildout ▶ 00:04:45. Profits concentrate among few layers while open source captures limited revenue; closed models and orchestration drive value. Vertical integration on physical layer provides significant advantage as buildout scales ▶ 00:08:44. "AI is a more extreme revolution compared to industrialization, financialization and digitalization" — Jim Liu ▶ 00:01:56. "This concentration results in less inflation" — Jim Liu ▶ 00:02:24. AI supercycle intact after overleveraged washout; summer seasonality was mistaken for a cycle end, with bounce continuing toward Nvidia earnings; Micron, related memory/storage, and power names favored. Bitcoin pullback viewed as mid-cycle (closer to 2020 than 2022). Financialization and institutional builds on Bitcoin (e.g., Saylor-style) are the path across the Geoffrey Moore adoption chasm; pure orange-pilling is tapped out ▶ 00:27:19. Prefer concentrated megatrends (AI, Bitcoin, power/data centers) and add hardest when short-term sentiment is negative; price leads narrative/headlines. Full upside capture preferred over covered calls on core high-upside names early in the cycle. Long duration, equanimity, and mostly doing nothing compound better than short-term instruments, margin, or overtrading. Settled baseload (IREN ~2 GW already in Batch Zero, with undisclosed additional batch-zero baseload beyond Sweetwater 1–2 at bare-minimum ~500 MW and possibly more) is a multi-year moat versus peers still securing power; scarcity favors holding contract duration open for higher 2027 ARR/GPU pricing rather than early TCV-heavy deals. Financing mix prefers asset-backed loans over ATM/equity or convertibles. Sweetwater building fast; Childress H3/H4 on track for late-2026/early-2027 revenue. Nvidia Australia PR is generic lift, not near-term Bundy 2027 energization. Recycled last-cycle alts will not enter price discovery even under easy policy: "even if Trump turns on the money printer and rates go back to zero percent, it doesn't necessarily mean that your ticker from last cycle is gonna go into price discovery" ▶ 00:01:25. Prefer multi-market wealth-creation over single-asset bull-market heroes. Bitcoin range-bound until clean 82k break; add toward 50-55k on breakdown or ladder after breakout; ETH/SOL/BNB/LINK have cleaner structure as possible front-runners. Fed rate hikes/cuts a nothing-burger; BoJ more relevant.

Risks & bear case

Overcapacity, accelerated depreciation, commoditization, and regulatory capture. Bust in prior cycles limited by human reorientation speed in physical world ▶ 00:03:48. Overleveraged liquidations (e.g., Aschenbrenner fund, Asian agents) amplify drawdowns in AI names. Short-term instruments, options income strategies, and margin make participants easiest to exploit. Crowding visible in 13Fs. Tax avoidance contortions can impair compounding even though high taxes signal large gains. Fundamentalist fracture and noisy discourse can distract from product/financialization progress in Bitcoin. Long-end yields (~5.2–5.3% 30y) and oil damp near-term momentum. Cipher-style high-GW PRs that omit clear baseload leave peers in precarious competitive spots versus settled-power operators; community opposition (e.g., Firmus Tasmania) can delay sites. Publicizing undisclosed sites risks backlash. Convertible financing disliked; over-reliance on equity/ATM dilutive. Bundy energization 2028 (not 2027) removes near-term Australia catalyst. Last-cycle alt nostalgia and meme-coin charting played out (2017 names failed ATH reclaim despite QE/ZIRP). Messy new-chain launches (Arc/Argus shitshow: early bridge, RPC fails, ~2x OTC). BoJ hike (even priced) historically associated with BTC weakness (caveat June 2026). BTC 82k/50-week MA resistance; crypto Spaces somewhat played out. Single-asset hero risk versus multi-market accounts. Sui-like degen fade without unique usage.

Timeline of developments

2026-08-04: Jim Liu delivers macro thesis tying AI CapEx to liquidity, production, and inflation dynamics; analyzes buildout layers from inference to physical power ↗
2026-08-16: Mike Alfred et al. discuss Bitcoin as mid-cycle correction amid equity ATHs, AI supercycle bounce post-washout (Nebius/CoreWeave strength, Aschenbrenner liquidation), Clarity Act as BTC catalyst, financialization crossing Bitcoin’s adoption chasm, and megatrend allocation (AI/Bitcoin/power-data centers) with selective trims into miners/neoclouds ↗
2026-09-10: Frans Bakker / Bitcoin Butcher / Mark Cleary OnlyFrans space covers IREN Batch Zero baseload edge (Sweetwater 1–2 + additional undisclosed), Cipher 2.5 GW pipeline PR skepticism (baseload omission), Nvidia Australia 2 GW collaboration PR (generic, Bundy 2028), directional Goldman ~$12.5B incremental ARR math, financing preferences, Sweetwater/Childress construction cadence, yields/oil headwinds, and $45 IREN support ↗
2026-09-16: Wabi/Dansec Market Talk: last-cycle alts (2017 NEO/Lisk) failed ATH reclaim despite QE/ZIRP so easy policy ≠ recycled tickers in discovery; Circle Arc/Argus launch chaotic (early bridge, RPC fails, ~2x OTC, missed 58k% runner); Sui degen fade; Robinhood potential Base killer; Fed nothing-burger vs BoJ hike (priced, Sep 17-18); BTC range since 58k with 82k/50w MA resistance, prefer 50-55k add or post-breakout ladder; ETH/SOL/BNB/LINK cleaner structure ↗

Open questions

Whether Anthropic $80B ARR and OpenAI $42.6B ARR figures hold ▶ 00:05:18.
Extent to which 1999 dot-com promises materialized exactly 15 years later via Google, AWS, Facebook, Microsoft, and Apple ▶ 00:03:35.
Timing and magnitude of next Bitcoin catalysts beyond Clarity Act.
Scale and timing of any physical-AI or robotics rotation relative to current digital/infrastructure AI trade.
How much near-term opportunity cost justifies trimming core Bitcoin into miners/data-center equities (e.g., Cipher, IREN).
Exact size and mix of IREN’s undisclosed additional Batch Zero baseload beyond Sweetwater 1–2.
Whether holding contract duration open fully monetizes 2027 scarcity pricing versus peers locking early TCV.
Impact of sustained 5%+ long-end yields and oil on AI/power equity momentum and financing costs.
Whether Bitcoin breaks 82k/50-week MA resistance or revisits 50-55k.
Realized impact of expected BoJ Sep 17-18 2026 hike (to ~1.25%) on Bitcoin given mixed 2024-26 history.
Whether Robinhood L2 becomes a Base killer or Circle Arc/Argus gains durable unique usage versus Sui-like degen fade.

Notable predictions to track

AI will operate 24/7/365 at PhD+ IQ level, resulting in deflation ▶ 00:02:59.
Higher diffusion rate allows AI build-out to last significantly longer than dot-com boom ▶ 00:04:45.
AI bounce continues through Nvidia earnings; multi-year AI top calls since late 2024 remain wrong.
Bitcoin mid-cycle correction resolves higher; financialization/products get it across the adoption chasm ▶ 00:27:19.
Batch-zero settled power remains multi-year moat; from-scratch 500 MW is ~2032 problem; IREN captures elevated 2027 ARR via open contract duration and ~$25M/MW pricing.
IREN delivers ~480 MW AI Cloud by end-2026 and ~1.2 GW by end-2027; Childress H3 mid-Q4 2026 revenue, H4 late-Dec/Q1 grace; Bundy energization from 2028.
Peak IT build cadence ~850 MW/year around 2028–29.
Recycled last-cycle alt tickers will not go into price discovery even if Trump money printer/rates return to zero ▶ 00:01:25.
Bitcoin remains range-bound until clean 82k break; prefer adds toward 50-55k on breakdown or ladder after breakout; ETH/SOL/BNB/LINK possible front-runners.
Fed hikes/cuts a nothing-burger; BoJ hikes more consistently negative for Bitcoin (historical caveat) ▶ 00:07:33.
Robinhood as potential Base killer ▶ 00:05:29.