HYBSIS Outlook

What We Think Happens Next

Predictions are HYBSIS interpretations, not facts. Each one shows the sourced facts it rests on, the observed market signals, our reasoning, and a review date. Predictions are never deleted or rewritten: status changes are recorded in new dated snapshots.

Open: 11
Tracking: 0
Supported: 0
Partially Supported: 0
Not Supported: 0
Superseded: 0

Current snapshot: 25 Sept 2026 · 1 snapshot(s) on record · Prediction history

Fact

Established and sourced.

Observed signal

A measured pattern in collected evidence.

Interpretation

HYBSIS reasoning connecting facts and signals.

Prediction

A forward-looking, reviewable claim.

Next 12 months

Rack power density & cooling Open Confidence: medium

Prediction

Newly announced large-scale AI deployments tracked by HYBSIS will predominantly specify rack-scale, liquid-cooled systems rather than air-cooled servers.

Fact: NVIDIA's GB200 NVL72 is a liquid-cooled, rack-scale system connecting 72 GPUs in one NVLink domain. Source: NVIDIA — NVIDIA GB200 NVL72
Observed signal: Liquid Cooling Adoption (rising, 31 obs.); Rack Power Density (rising, 22 obs.)
Interpretation: Leading accelerator platforms are now designed as racks rather than servers, and their thermal envelope assumes liquid cooling. Facilities not designed for this become retrofit projects.

Power availability Open Confidence: medium

Prediction

Power availability, rather than GPU supply, will be the constraint most frequently cited in AI data-centre announcements HYBSIS tracks.

Fact: The IEA projects data-centre electricity consumption to more than double by 2030 in its base case, with AI as the main driver. Source: International Energy Agency — Energy and AI
Observed signal: Power Availability (rising, 58 obs.); Grid Capacity Pressure (rising, 72 obs.)
Interpretation: Accelerator supply has expanded with each generation while grid connections and generation follow multi-year utility and permitting cycles.

NeoCloud consolidation Open Confidence: medium

Prediction

At least one further acquisition or merger involving a tracked NeoCloud provider will be announced.

Fact: CoreWeave announced an agreement to acquire Weights & Biases in March 2025. Source: CoreWeave — CoreWeave Investor Relations
Observed signal: NeoCloud Consolidation (rising, 27 obs.); AI Infrastructure Funding (rising, 139 obs.)
Interpretation: Public-market access gives the largest providers acquisition currency, while smaller providers face rising capital and power requirements.

GPU generation transition Open Confidence: medium

Prediction

Blackwell-generation systems will be the default in new NeoCloud capacity announcements, and the first Rubin-generation deployments will be announced by tracked providers.

Fact: NVIDIA's published roadmap runs from Blackwell and Blackwell Ultra to the Vera Rubin platform on an annual cadence. Source: NVIDIA — NVIDIA Newsroom
Observed signal: GPU Generation Transition (rising, 40 obs.)
Interpretation: NeoClouds compete on time-to-availability of new generations; being early is part of their commercial positioning.

Next 3 years

NeoCloud growth Open Confidence: medium

Prediction

NeoClouds will differentiate primarily on secured power, delivery capability and time-to-capacity rather than on access to GPUs alone.

Fact: The IEA projects data-centre electricity consumption to more than double by 2030 in its base case. Source: International Energy Agency — Energy and AI
Observed signal: NeoCloud Capacity (rising, 152 obs.); Power Availability (rising, 58 obs.)
Interpretation: As accelerator supply normalises, the scarce inputs become energised sites and the ability to deliver them on schedule.

AI networking Open Confidence: medium

Prediction

Ethernet-based AI fabrics will take a growing share of new large-cluster deployments, alongside InfiniBand rather than wholly replacing it.

Fact: The Ultra Ethernet Consortium published its 1.0 specification for Ethernet-based AI and HPC networking. Source: Ultra Ethernet Consortium
Observed signal: AI Network Bandwidth Demand (rising, 21 obs.)
Interpretation: Standardisation lowers adoption risk for operators who prefer multi-vendor Ethernet operations at scale.

Sovereign AI Open Confidence: medium

Prediction

Sovereign and regional AI programmes will become a material source of demand for NeoCloud and AI-factory capacity, most visibly in Europe and the Middle East.

Fact: The European Commission announced the InvestAI initiative including funding intended for AI gigafactories. Source: European Commission — AI Factories
Observed signal: Sovereign AI Infrastructure (rising, 60 obs.)
Interpretation: Governments increasingly treat AI compute as strategic infrastructure, favouring locally operated capacity.

Optical networking Open Confidence: low

Prediction

Co-packaged optics will move from demonstrations into production AI network deployments.

Observed signal: AI Network Bandwidth Demand (rising, 21 obs.)
Interpretation: Rising port speeds make optical power and reliability a first-order design concern, favouring integration closer to the switch silicon.

Next 5+ years

Energy generation Open Confidence: medium

Prediction

Large AI campus developers will routinely participate directly in new energy generation — on-site or contracted, including firm low-carbon sources — making energy development a core AI infrastructure competence.

Fact: The IEA projects data-centre electricity consumption to more than double by 2030 in its base case. Source: International Energy Agency — Energy and AI
Observed signal: Power Availability (rising, 58 obs.); Grid Capacity Pressure (rising, 72 obs.)
Interpretation: When grid supply cannot keep pace, developers that bring their own generation reduce schedule risk.

Grid capacity Open Confidence: medium

Prediction

Grid interconnection and transmission capacity, not chips, will be the binding constraint on AI capacity growth in major markets.

Fact: The IEA projects data-centre electricity consumption to more than double by 2030 in its base case. Source: International Energy Agency — Energy and AI
Observed signal: Grid Capacity Pressure (rising, 72 obs.); Power Availability (rising, 58 obs.)
Interpretation: Transmission build-out timelines are long relative to data-centre and accelerator cycles.

Distributed inference Open Confidence: low

Prediction

Inference will drive a growing share of AI infrastructure build-out and favour more distributed, regionally deployed capacity alongside concentrated training campuses.

Observed signal: GPU Compute Demand (rising, 55 obs.); AI Data Centre Capacity (rising, 610 obs.)
Interpretation: Inference is latency-, data-residency- and cost-sensitive in ways training is not, which favours geographic distribution.

Signals used as evidence

Computed from the shared dataset — including NeoCloud Intelligence data — by the same engine that powers HYBSIS Market Signals.

GPU Compute Demand

Rising

Announcements of GPU capacity purchases, reservations and cluster build-outs.

Strength: strong
Confidence: high
Observations: 55
Latest evidence: 29 Oct 2026

How this was derived

39 matching item(s) in the last 90 days vs 0 in the previous 90 days, from 8 distinct source(s). Rising/falling requires a ≥1.5× change and a difference of at least 2.

AI Data Centre Capacity

Rising

New AI data-centre campuses, expansions and capacity announcements.

Strength: strong
Confidence: high
Observations: 610
Latest evidence: 1 Nov 2026

How this was derived

588 matching item(s) in the last 90 days vs 3 in the previous 90 days, from 10 distinct source(s). Rising/falling requires a ≥1.5× change and a difference of at least 2.

NeoCloud Capacity

Rising

Capacity expansion announced by tracked NeoCloud providers.

Strength: strong
Confidence: high
Observations: 152
Latest evidence: 30 Oct 2026

How this was derived

135 matching item(s) in the last 90 days vs 1 in the previous 90 days, from 7 distinct source(s). Rising/falling requires a ≥1.5× change and a difference of at least 2.

GPU Generation Transition

Rising

Adoption of newer accelerator generations across providers.

Strength: strong
Confidence: high
Observations: 40
Latest evidence: 1 Nov 2026

How this was derived

33 matching item(s) in the last 90 days vs 0 in the previous 90 days, from 5 distinct source(s). Rising/falling requires a ≥1.5× change and a difference of at least 2.

AI Infrastructure Funding

Rising

Equity, debt and public-market financing for AI infrastructure.

Strength: strong
Confidence: high
Observations: 139
Latest evidence: 30 Oct 2026

How this was derived

100 matching item(s) in the last 90 days vs 6 in the previous 90 days, from 22 distinct source(s). Rising/falling requires a ≥1.5× change and a difference of at least 2.

Rack Power Density

Rising

Evidence of rising per-rack power in AI deployments.

Strength: strong
Confidence: high
Observations: 22
Latest evidence: 30 Oct 2026

How this was derived

20 matching item(s) in the last 90 days vs 0 in the previous 90 days, from 5 distinct source(s). Rising/falling requires a ≥1.5× change and a difference of at least 2.

Liquid Cooling Adoption

Rising

Deployments and products using direct-to-chip or immersion liquid cooling.

Strength: strong
Confidence: high
Observations: 31
Latest evidence: 30 Oct 2026

How this was derived

30 matching item(s) in the last 90 days vs 0 in the previous 90 days, from 5 distinct source(s). Rising/falling requires a ≥1.5× change and a difference of at least 2.

Power Availability

Rising

Power procurement, PPAs, on-site generation and energy partnerships for data centres.

Strength: strong
Confidence: high
Observations: 58
Latest evidence: 31 Oct 2026

How this was derived

52 matching item(s) in the last 90 days vs 1 in the previous 90 days, from 7 distinct source(s). Rising/falling requires a ≥1.5× change and a difference of at least 2.

Grid Capacity Pressure

Rising

Grid interconnection queues, moratoria, and utility constraints affecting data centres.

Strength: strong
Confidence: high
Observations: 72
Latest evidence: 1 Nov 2026

How this was derived

71 matching item(s) in the last 90 days vs 0 in the previous 90 days, from 6 distinct source(s). Rising/falling requires a ≥1.5× change and a difference of at least 2.

AI Network Bandwidth Demand

Rising

Advances and deployments in AI fabrics, optics and interconnect.

Strength: strong
Confidence: high
Observations: 21
Latest evidence: 29 Oct 2026

How this was derived

20 matching item(s) in the last 90 days vs 0 in the previous 90 days, from 6 distinct source(s). Rising/falling requires a ≥1.5× change and a difference of at least 2.

Sovereign AI Infrastructure

Rising

National or regional AI infrastructure programmes and sovereign compute.

Strength: strong
Confidence: high
Observations: 60
Latest evidence: 30 Oct 2026

How this was derived

56 matching item(s) in the last 90 days vs 0 in the previous 90 days, from 7 distinct source(s). Rising/falling requires a ≥1.5× change and a difference of at least 2.

NeoCloud Consolidation

Rising

Mergers, acquisitions and strategic combinations among AI infrastructure providers.

Strength: strong
Confidence: high
Observations: 27
Latest evidence: 30 Oct 2026

How this was derived

20 matching item(s) in the last 90 days vs 1 in the previous 90 days, from 5 distinct source(s). Rising/falling requires a ≥1.5× change and a difference of at least 2.

Prediction history

Every prediction ever published, with its status at each snapshot. Snapshot files are immutable and hash-verified.

Prediction	Horizon	First published	Status history
Newly announced large-scale AI deployments tracked by HYBSIS will predominantly specify rack-scale, liquid-cooled systems rather than air-cooled servers.	Next 12 months	25 Sept 2026	2026-09-25: Open
Power availability, rather than GPU supply, will be the constraint most frequently cited in AI data-centre announcements HYBSIS tracks.	Next 12 months	25 Sept 2026	2026-09-25: Open
At least one further acquisition or merger involving a tracked NeoCloud provider will be announced.	Next 12 months	25 Sept 2026	2026-09-25: Open
Blackwell-generation systems will be the default in new NeoCloud capacity announcements, and the first Rubin-generation deployments will be announced by tracked providers.	Next 12 months	25 Sept 2026	2026-09-25: Open
NeoClouds will differentiate primarily on secured power, delivery capability and time-to-capacity rather than on access to GPUs alone.	Next 3 years	25 Sept 2026	2026-09-25: Open
Ethernet-based AI fabrics will take a growing share of new large-cluster deployments, alongside InfiniBand rather than wholly replacing it.	Next 3 years	25 Sept 2026	2026-09-25: Open
Sovereign and regional AI programmes will become a material source of demand for NeoCloud and AI-factory capacity, most visibly in Europe and the Middle East.	Next 3 years	25 Sept 2026	2026-09-25: Open
Co-packaged optics will move from demonstrations into production AI network deployments.	Next 3 years	25 Sept 2026	2026-09-25: Open
Large AI campus developers will routinely participate directly in new energy generation — on-site or contracted, including firm low-carbon sources — making energy development a core AI infrastructure competence.	Next 5+ years	25 Sept 2026	2026-09-25: Open
Grid interconnection and transmission capacity, not chips, will be the binding constraint on AI capacity growth in major markets.	Next 5+ years	25 Sept 2026	2026-09-25: Open
Inference will drive a growing share of AI infrastructure build-out and favour more distributed, regionally deployed capacity alongside concentrated training campuses.	Next 5+ years	25 Sept 2026	2026-09-25: Open