News, notes & investors

What moved the market, and why it matters to the book.

Not a link roundup. Each note states what happened, cites where it was reported, and says what it changes for the price of compute — including when the answer is “nothing yet”. If a claim here is wrong, the source is one click away and we will correct it in place with the correction dated.

5notes this quarter

10sources cited

Quarterlyspread report

Market notes

This quarter

ModelsJuly 2026

The open-weight frontier keeps escalating

Moonshot released Kimi K3, reported as a 2.8-trillion-parameter mixture-of-experts model with open weights following release1. It follows K2.6 in April, a 1T-parameter model with 32B active parameters and a 256K context window2.

So what

Open weights move demand from training to serving, and serving is far more price-sensitive than training. Every capable open model that ships enlarges the population of buyers who care about the difference between $1.50 and $3.00 an hour — which is the population a reference rate serves.

PolicyMay–June 2026

Export controls close the overseas-subsidiary route

New US guidance requires export licences for transfers of NVIDIA’s most advanced processors to entities headquartered in China or Macau, explicitly covering their subsidiaries outside China3. It follows the January framework that moved chips such as the H200 from presumption of denial to case-by-case review and applied a 25% tariff to covered imports not destined for the US technology supply chain4.

So what

Policy is now a first-order input to where capacity is worth building, and therefore to the regional basis between instruments. An H100-hour in Virginia and one in Singapore have different regulatory reachability for different buyers — that is a spread, and spreads are what a book prices.

Build-outFebruary 2026

Hyperscaler capex guidance approaches three quarters of a trillion

The four largest US cloud providers have guided to combined 2026 capital expenditure approaching $700 billion6, with the sum of individual guidance ranges put at roughly $725 billion — about 77% above the prior year5.

So what

This is the supply curve every buyer on our book trades against. It is also the strongest argument that the current contracting model breaks: capacity is being committed years ahead against demand nobody can forecast a quarter ahead.

PowerStanding

Power, not silicon, is the binding constraint

The IEA projects data-centre electricity demand more than doubling by 2030 to around 945 TWh, against roughly 415 TWh in 20247. Consumption is heavily concentrated — the US at 45%, China 25%, Europe 15%8.

So what

When the constraint is power, capacity stops being uniform and starts carrying a location-dependent price. That is precisely the condition under which a regional forward curve becomes useful rather than decorative.

MarketJuly 2026

The spread on an H100-hour is still roughly 4.7×

Published on-demand list prices for the same GPU span about $1.49 to $6.98 across more than fifteen providers9, with neoclouds clustered at $1.50–$2.50 and spot trading near $1.0310.

So what

We publish this every quarter because it is the single cleanest measure of whether the market is working. If the spread narrows durably, the thesis is being solved — by us or by someone else. Either way it should be measured in public.

Events

Where we will be

Q3 2026

Compute Markets Roundtable

Closed session for desks and operators on margining an asset with no historical curve.

New York

Q3 2026

Supply origination briefing

For regional operators and energy-adjacent capacity owners.

Abu Dhabi (ADGM)

Q4 2026

Reference rate consultation

Public comment on the methodology before it is fixed.

Virtual

Investors

Materials and reporting

What we report

Notional traded, unique counterparties on each side, spread on the reference instrument, and capacity listed versus capacity sold. Quarterly, to holders of record.

What we will not report

Counterparty names, individual positions, or any metric that lets one member infer another’s book. Neutrality is not a slogan here — it is the thing being sold.

The public spread report

The H100 spread figure above is published every quarter whether it flatters us or not. It is the measure of whether the market is working.

Contact

investors@hyperlink.org. Materials are shared under NDA; we do not circulate a deck without a conversation first.

NoteNothing on this page is an offer to sell or a solicitation to buy any security, and nothing here is investment advice. Forward-looking statements about the market reflect our view at the date shown and are not predictions.

Open the terminal Where demand comes from About the company

Sources

Every figure above is numbered to an entry here. Links last read 27 July 2026.

1
China's Moonshot AI releases Kimi K3, the largest open-source model ever
VentureBeat · July 2026 · Reporting
ParaphraseKimi K3 is a 2.8-trillion-parameter mixture-of-experts model, with open weights following the model's release.
2
Kimi K2.6 — open-weight frontier model
Moonshot AI (via Interconnects) · April 2026 · Reporting
ParaphraseKimi K2.6 is a 1T-parameter mixture-of-experts model with 32B active parameters and a 256K context window, released under a permissive open-weight licence.
3
U.S. takes step to halt Nvidia AI chip shipments to Chinese firms outside China
CNBC · 31 May 2026 · Reporting
ParaphraseNew guidance requires export licences for transfers of Nvidia's most advanced processors to entities headquartered in China or Macau, closing the overseas-subsidiary route.
4
Administration Policies on Advanced AI Chips Codified
Mayer Brown · January 2026 · Reporting
ParaphraseThe framework effective 15 January 2026 moves licensing for chips such as the H200 from a presumption of denial to case-by-case review, and applies a 25% tariff to covered advanced AI chip imports not destined for the US technology supply chain.
5
Google, Microsoft, Meta, and Amazon capex spending to hit $725 billion in 2026, up 77% from last year
Tom's Hardware · February 2026 · Reporting
Google, Amazon, Microsoft, and Meta collectively plan to allocate $725 billion to capital expenditures in 2026 — up 77% from last year's $410 billion.
A sum of separate company guidance ranges, not a reported figure. Individual guidance: Amazon ~$200B, Google $175–185B, Meta $115–135B, Microsoft $110–120B.
6
Tech AI spending approaches $700 billion in 2026, cash taking big hit
CNBC · 6 February 2026 · Reporting
ParaphraseThe four largest US cloud providers have guided to combined capital expenditure approaching $700 billion for 2026, the bulk of it directed at AI data centres.
7
Energy and AI — Executive summary
International Energy Agency · April 2025 · Primary
Electricity demand from data centres worldwide is set to more than double by 2030 to around 945 terawatt-hours (TWh) … slightly more than the entire electricity consumption of Japan today.
8
Energy and AI — Energy demand from AI
International Energy Agency · April 2025 · Primary
The United States accounted for the largest share of global data centre electricity consumption in 2024 (45%), followed by China (25%) and Europe (15%).
9
H100 Rental Prices Compared: $1.49–$6.98/hr Across 15+ Cloud Providers
IntuitionLabs · 2026 · Third-party estimate
ParaphraseOn-demand H100 list prices span $1.49 to $6.98 per GPU-hour across more than fifteen providers, a spread of roughly 4.7× for the same silicon.
Survey of published list prices. Committed and reserved rates are negotiated and are not represented here.
10
GPU Cloud Pricing Comparison 2026
Spheron · 2026 · Third-party estimate
ParaphraseSpecialised neoclouds cluster at $1.50–$2.50 per H100-hour while the large general-purpose clouds remain in the mid-single digits; spot capacity has traded near $1.03.

Where a claim rests on a third-party estimate rather than the party that owns the number, the entry says so. Figures that are Exascale’s own — our rate card, our fee schedule — carry no citation, because they are ours to set rather than facts about the world.