The Next AI Bottleneck: Everything Between the Grid and the GPU
By Ieva Bružaitė | Founder, Orion Capital | 26 October 2026
The next AI infrastructure bottleneck is no longer just about chips. It is about power, grid connections and cooling.
For much of the artificial intelligence investment cycle, the market has focused on one question: who can secure the most advanced chips?
NVIDIA, Broadcom and the wider semiconductor industry have been central to that discussion. But recent developments suggest that investors need to pay closer attention to something less visible: the infrastructure required to turn computing capacity into an operating business.
A data centre can have customers, contracts and billions of dollars in committed investment. None of that guarantees it will have electricity when it needs it.
The question is no longer simply how much AI computing capacity companies intend to build. It is this:
How much of that capacity can actually be connected, powered and brought online on schedule?
Three developments this week offer a useful way to examine that question.
When the Grid Delays Revenue
On 21 October, Australian data-centre operator NEXTDC revised its FY2027 financial guidance after delays to electricity network works affecting its M3 Melbourne facility.
The next increase in grid supply is now expected during FY2028. Approximately A$140 million of previously anticipated FY2027 revenue associated with the affected contracts was removed from the company’s revised guidance.
The financial consequences were substantial.
| Net revenue | A$615–640m | A$475–500m |
| Underlying EBITDA | A$385–410m | A$245–265m |
NEXTDC shares fell sharply following the announcement.
The underlying problem was not insufficient demand for computing capacity. The delay involved electrical network works, including complications related to community opposition and local approvals.
This distinction matters. Contracted demand is not necessarily operational capacity, and operational capacity is what ultimately generates revenue.
For investors evaluating the AI infrastructure build-out, NEXTDC offers a concrete example of how a grid connection can become a financial variable rather than simply a construction milestone.
The Cable Constraint
A day later, Nexans highlighted a different part of the same infrastructure challenge.
In an interview with Reuters on 22 October, the French cable manufacturer described tight global production capacity. Its chief financial officer indicated that establishing a new cross-linked polyethylene cable production line can take approximately three years.
That is a long adjustment period for an industry facing rapidly growing investment requirements.
Nexans reported third-quarter organic growth of 9.0% at group level, while its electrification activities recorded organic growth of 7.1%.
The company highlighted momentum in its PWR-Grid and PWR-Connect businesses, which serve electrical infrastructure markets.
Not all of that growth comes from AI. Grid modernisation, renewable energy projects and broader electrification are also important demand drivers.
But that is precisely why the supply question is interesting. Data centres are competing for electrical infrastructure within a much larger investment cycle.
This makes cable supply another critical part of the AI infrastructure bottleneck.
A shortage of advanced processors may be solved by additional semiconductor production. A shortage of suitable electrical equipment can involve a different set of factories, investment decisions and construction timelines.
AI Infrastructure Bottleneck: Power and Cooling Reach the Income Statement
The latest results from Bloom Energy and SLB provide another perspective.
On 22 October, Bloom Energy reported third-quarter revenue of approximately $1.236 billion, an increase of 138.1% year over year. It raised its full-year 2026 revenue guidance to $4.4–4.6 billion.
Bloom’s on-site power generation technology is relevant to customers seeking reliable electricity without depending exclusively on conventional grid connections.
However, its overall revenue growth should not be treated as a direct measure of AI data-centre demand. The company serves a broader customer base.
SLB’s results, released on 23 October, showed a more specifically identified data-centre business.
Its Data Center Solutions revenue reached $230 million for the quarter, increasing 76% year over year and 23% sequentially.
The business includes modular infrastructure and cooling-related solutions. SLB also discussed its planned acquisition of Kelvion, a specialist in thermal management.
Yet there was a qualification in the results: costs associated with ramping up new projects affected profitability.
That is an important reminder for investors. A rapidly expanding market can produce impressive revenue growth without immediately producing equally impressive margins.
The AI infrastructure bottleneck creates demand, but profitability remains the real test.
What Is Changing in the AI Investment Chain?
The conventional AI investment narrative begins with semiconductors and moves toward computing capacity.
The physical infrastructure introduces additional dependencies:
CHIPS → COMPUTE → POWER → GRID → COOLING
In practice, these systems must develop together. A facility cannot operate at full capacity simply because its processors have arrived.
This creates different investment exposures across the supply chain.
The AI infrastructure bottleneck is reshaping investment opportunities across the entire supply chain.
| AI processors and custom silicon | NVIDIA, Broadcom | Will computing demand support expected growth? |
| Data-centre networking | Arista Networks | How quickly will new capacity be deployed? |
| Electrical distribution | Eaton, Schneider Electric | Can orders convert into profitable deliveries? |
| Critical power and cooling | Vertiv | Can growth be sustained without margin pressure? |
| Power generation and grid equipment | GE Vernova | How much demand becomes firm orders? |
| On-site power | Bloom Energy | Can expansion generate durable cash flow? |
| Cables and electrical connections | Nexans | Will lead times support pricing power? |
| Modular infrastructure and cooling | SLB | Can growth translate into profitability? |
These companies should not be valued as though they offer identical exposure to the AI cycle.
Their competitive positions, capital requirements, margins and customer relationships differ substantially.
The investment opportunity, if one exists, must be established company by company.
The Difference Between an Announcement and an Operating Asset
There is another distinction that deserves attention as AI infrastructure commitments grow larger.
An announced data-centre project is not the same as a completed facility.
A power purchase agreement does not necessarily mean the required electricity can be delivered immediately.
A preliminary supply framework is not equivalent to a firm equipment order.
And an order backlog is not the same as recognised revenue.
These differences become increasingly important when investors attempt to estimate future earnings from large infrastructure announcements.
The NEXTDC case demonstrates the consequences of treating project schedules as more certain than they actually are.
For equipment suppliers, the corresponding question is whether growing demand produces firm orders, deliveries and cash generation on commercially attractive terms.
Three Possible Paths for the Infrastructure Trade
1. AI Infrastructure Bottlenecks persist
Grid connections, electrical equipment and cooling systems remain difficult to deliver quickly. Established suppliers may benefit from stronger order visibility and sustained demand, although valuations and execution costs still matter.
2. Supply catches up
Manufacturing expansion, faster approvals and new infrastructure investment reduce delivery constraints. Data-centre operators could benefit from improved schedules, while suppliers might face less favourable pricing conditions.
3. AI capital expenditure slows
A moderation in hyperscaler investment could ease infrastructure constraints without any dramatic increase in supply. In that scenario, companies that expanded production capacity aggressively might face underutilisation, margin pressure or weaker returns on capital.
The same industry can therefore produce very different investment outcomes depending on how demand, supply and valuations evolve.
Why Strong Demand Is Not Enough
It would be tempting to conclude that electrical infrastructure companies must be the next major beneficiaries of the AI boom.
That conclusion would be premature.
An AI infrastructure bottleneck does not automatically create an attractive investment.
Investors must distinguish between a structurally attractive market and an attractively priced stock.
Strong revenue growth may already be reflected in share prices. New manufacturing capacity may require substantial capital expenditure. Competition may increase as more suppliers enter the market.
For companies such as Eaton, Vertiv and GE Vernova, the next important evidence will come from order quality, delivery schedules, margins and management guidance.
For NVIDIA, Broadcom and Arista, infrastructure delays could affect the timing of hardware deployment, even if long-term demand remains strong.
These are potential transmission mechanisms, not evidence that a specific company has already suffered a measurable earnings impact.
What Investors Should Watch Next
- Changes in transformer, cable and switchgear delivery times
- Firm orders versus preliminary capacity announcements
- Electrical equipment and cooling suppliers’ backlog conversion
- Grid interconnection delays affecting data-centre operators
- Power availability and project commissioning schedules
- Gross margins, free cash flow and capital expenditure requirements
- Hyperscaler spending plans and changes in expected deployment timelines
The most useful signal will not necessarily be another multibillion-dollar project announcement. It may be evidence that existing infrastructure suppliers are converting constrained capacity into sustainable earnings.
The Orion View
The recent developments do not suggest that semiconductors have become less important to artificial intelligence.
They suggest that semiconductor availability alone cannot determine the pace of the AI build-out.
NEXTDC demonstrates how electricity network delays can alter financial guidance. Nexans illustrates the difficulty of rapidly expanding critical electrical manufacturing capacity. Bloom Energy and SLB show that demand for supporting infrastructure is already visible in reported revenue.
These are independent signals. They do not establish a universal global shortage, and they do not prove that every power or cooling supplier is undervalued.
What they do establish is a stronger reason to examine the industrial infrastructure surrounding AI computing.
The AI infrastructure bottleneck is becoming an investment question as much as an engineering challenge.
For Orion Capital, the next question is not simply which companies will sell more equipment.
It is which businesses can convert the infrastructure requirements of AI into profitable, sustainable growth — and whether investors are paying a reasonable price for that opportunity.
The next constraint on AI growth may not be the chip itself, but everything required to make that chip productive.
Investments with a Soul.
Disclaimer: This material is provided for informational and research purposes only and does not constitute investment advice or a recommendation to buy or sell any security.
Sources
NEXTDC – M3 Melbourne Power Delay and Revised FY27 Guidance, 21 October 2026
Reuters – NEXTDC shares fall following revenue delay, 21 October 2026
Nexans – 9M 2026 Financial Information, 22 October 2026
Reuters – Cable capacity emerges as bottleneck for AI data centres, 22 October 2026
Bloom Energy – Q3 2026 Financial Results, SEC filing
SLB – Third-Quarter 2026 Results, 23 October 2026
About the Author
Ieva Bružaitė is the Founder of Orion Capital, an independent research platform covering AI, semiconductors and long-term investing.