New York Enacts Data Center Moratorium, Australia Mandates Energy Offsets, and Ollama Raises $65M Series B

Mid-July 2026 brings a structural realignment: AI infrastructure growth is colliding with physical constraints that no amount of capital can immediately overcome. New York Governor Hochul signs a one-year moratorium on new data centres consuming ≥50MW — the first US state to halt hyperscale expansion — triggering a Generic Environmental Impact Statement over grid stability, water use, and household utility cost socialisation. Australia follows 24 hours later with a policy that inverts the problem: mandatory energy underwriting (data centres must generate new renewables equal to their draw) and an explicit rejection of copyright exemptions for AI training ("scraping is theft"). And against this backdrop, Ollama closes a $65M Series B (Theory Ventures lead, Benchmark + 8VC + YC), reaching 8.9M monthly active developers and 85% Fortune 500 adoption — proof that "local-first inference" is now an enterprise-grade alternative to cloud APIs.

---

⚡ New York's 50MW Moratorium — Grid Capacity as the New Scaling Limit

The AI Data Centre Energy Crisis

How much power frontier AI actually consumes:

A single 2GW AI mega-campus would consume 5.7% of New York State's entire grid capacity. Five such facilities would consume 28% — pushing the state's grid to capacity just for AI cooling.

Key facts about New York's power grid constraints:

• Con Edison (New York City utility) is already running at 95–98% capacity on summer peak days
• New York's offshore wind buildout (25 GW by 2035) is behind schedule — current installed: ~1.1 GW
• Transmission upgrades needed to move power from upstate hydro to NYC data centres: 10+ years of permitting
• Average household electricity rate in NY: $0.24/kWh (5th highest in US) — already strained

The Moratorium — Legal Structure and Industry Impact

Executive Order details:

What the GEIS must assess:

1. Grid stability: cumulative impact of all approved + pending facilities on NYISO reliability margins
2. Water consumption: aggregate cooling water use vs watershed capacity (Hudson, Delaware basins)
3. Environmental justice: whether low-income communities bear disproportionate impact (heat, noise, truck traffic)
4. Rate impact: modelling household utility bill increases from data centre load growth through 2030

Industry consequences — where hyperscalers will now go:

---

🇦🇺 Australia's Energy Underwriting + Copyright Position — A Coherent Alternative

The Energy Underwriting Requirement

Australia's policy (announced July 15, 2026) — key provisions:

Why energy underwriting is smarter than a moratorium: New York's moratorium stops building. Australia's underwriting requirement accelerates renewable energy buildout — every megawatt of new AI compute must be paired with a megawatt of new clean generation. This turns hyperscale data centres from grid burdens into renewable energy catalysts.

Australia is uniquely positioned:

• Solar irradiance among the world's highest (most of the continent)
• Offshore wind potential: 2,200 GW (vs current ~5 GW installed)
• Natural gas export revenue funds the energy transition
• Low population density allows large land-use solar + wind projects without community conflicts

The Copyright Stance — Creator-Protective Outlier

What AI labs wanted from Australia (lobbied for): A "text and data mining" (TDM) exception to Australian copyright law — similar to the one in the UK and EU's CDSM Directive — that would allow AI training on any publicly accessible content without licence or payment.

PM Albanese's July 15 statement:

"The unauthorised scraping of artists' and creators' work to train AI models is, in plain terms, theft. Australia will not be creating TDM exceptions that benefit corporations at the expense of our artists."

The legal framework Australia will implement instead:

• Mandatory licensing: AI companies must negotiate licence agreements with content creators for training data
• Opt-in by default: No Australian-hosted content may be used for AI training without explicit rights clearance
• Provenance tracking: AI companies must maintain verifiable records of all training data sources

Comparison to other jurisdictions:

---

💻 Ollama $65M Series B — Local Inference Enters the Enterprise

Why Local Inference Is Growing

The three enterprise drivers of local AI inference:

Ollama's growth metrics:

What models run locally on consumer hardware (2026):

The hybrid inference architecture (Ollama's Series B focus): Rather than pure local vs pure cloud, Ollama is building a dynamic router that:

1. Classifies each inference request by: data sensitivity, latency requirement, model size needed
2. Routes sensitive/low-latency requests → local model
3. Routes complex/large-model requests → cloud API
4. Manages the handoff transparently (developer writes one API call; routing is automatic)

This hybrid approach means enterprises don't need to choose — they get local privacy + cloud capability simultaneously, with cost optimisation as a bonus.

---

📌 The Bottom Line

• new-york-50mw-data-center-moratorium: Gov. Hochul, July 14; ≥50MW threshold; 1-year halt; GEIS (grid stability, water, environmental justice, rate impact); a single 2GW mega-campus = 5.7% of NY grid; Con Edison 95-98% summer peak; wind buildout behind schedule (1.1 GW of 25 GW target); alternatives: Texas ERCOT (isolated grid, wind+solar), Quebec hydro (surplus, near-zero carbon), Iceland geothermal.
• australia-mandatory-ai-energy-copyright: Energy underwriting: new ≥10MW builds must generate new renewables equal to draw (turns data centres into renewable catalysts); water + location + community agreement requirements; copyright: "scraping is theft" (PM Albanese quote); mandatory licensing (opt-in default) + provenance tracking; highest creator protection of any major jurisdiction (US=low, EU=medium, Japan=very low, Australia=highest).
• ollama-65m-series-b-local-inference: $65M Series B (Theory Ventures + Benchmark + 8VC + YC); 8.9M MAU, 85% Fortune 500, 150+ models; privacy/compliance/cost ($12/M tokens cloud vs $0.02/M effective local)/latency (300-1500ms → 30-80ms) drivers; hardware frontier: DeepSeek V4 MoE (600B, 37B active) runs on M4 Max 128GB at 12 tok/s; Series B focus: hybrid dynamic router — classify by data sensitivity+latency → route local or cloud transparently.

---

📬 Stay Updated

Get the best of AI & technology delivered to your inbox every week. Subscribe to our free newsletter →

---

Disclosure: This post contains affiliate links. If you purchase through our links, we earn a small commission at no extra cost to you. We only recommend products we believe in.