Equinix: New CFO Says Agentic AI Could Drive Up to 40% of Data Center Demand Within Three Years, Defends Discipline Over Growth-at-All-Costs
RBC Capital Markets 2026 Global Communications and Infrastructure Conference, September 29, 2026
Equinix's newly installed CFO Olivier Leonetti used his first major investor appearance to lay out a bullish, if carefully hedged, view of the company's positioning in the AI buildout, arguing that the shift toward agentic AI workloads represents a structural demand driver that plays directly to Equinix's strengths in interconnection and neutrality. Just days into the role, Leonetti told RBC's Jonathan Atkin that the team had previously "undersold" the value proposition of the business, and that demand is now strong enough that Equinix is actively turning away orders to preserve ecosystem diversity rather than chase near-term bookings.
Agentic AI Framed as the Next Demand Wave
The most quantified new data point from the session was management's forecast on agentic AI penetration. Leonetti said Equinix believes agentic AI will represent 50% of total AI demand within three years, and that it will account for 30% to 40% of overall data center activity over the same horizon. The logic, as he explained it, is that agentic workloads require constant movement between models and clouds, which plays to Equinix's core interconnection business. "You don't join us to have access to compute. You join us to have access to an ecosystem of participants. You want to be free to move from one AI model to another, free to move from one cloud provider to another," he said. He added that the open-source and open-weight model trend reinforces this dynamic, since "large language model is better one week, better the second week," pushing customers toward open infrastructure rather than locked-in relationships. Orchestration for inference, he said, will increasingly happen inside Equinix's own facilities, pointing to a 60% increase in power deployed per cabinet as evidence that customers are already densifying their footprints for exactly this purpose.
Capacity Delivery Running Well Ahead of Plan
On the supply side, Leonetti disclosed that Equinix has deployed 40% more power than it had originally planned for this year, and that 2026 power delivery will be roughly double the 2025 level. He attributed the outperformance to long-standing, localized relationships with utilities, general contractors and power management partners built up over two decades under the company's longtime head of operations, which he said gives Equinix construction certainty that larger, gigawatt-scale projects competing for the same grid capacity often lack. Looking further out, the company intends to deploy 1 gigawatt of power between now and the end of 2029, with a further 2 gigawatts of land and power already secured and available beyond that. Leonetti also said Equinix's smaller, community-embedded data center footprint has largely insulated it from the political backlash facing larger hyperscale projects in the U.S. and Europe.
Demand Discipline Over Order Book Maximization
Leonetti was explicit that Equinix is deliberately constraining near-term bookings. He said the company recorded total sales velocity of about 30% in the second quarter and could post an even higher order book today, but is choosing not to in order to "nurture the best ecosystem for long-term value" rather than concentrate capacity among a narrow set of large customers. The company's newly launched Fabric One product, introduced roughly a month before the conference, is central to this strategy, cutting the time needed for participants to connect to other ecosystem actors from about a month to seconds. On pricing, Leonetti said Equinix is still evaluating usage-based models for cross-connects but emphasized that interconnection is sold as part of a bundled offering with space, power and ecosystem access, making it difficult to price in isolation. He indicated confidence that future pricing iterations will carry higher value regardless of structure.
Guidance Raise Not Tied to Incremental Capex
Leonetti clarified a point of investor confusion from earlier conversations that day: the company's recent guidance increase of three percentage points on both top-line and bottom-line growth through 2029 is not attributable to the additional capital expenditure Equinix has committed to. In fact, he said the incremental capex is a near-term drag on earnings, since new sites take roughly two years to build and another three years to fill before reaching full economic contribution. "Despite the drag, we have been able to generate better economics," he said, attributing the improved outlook instead to underlying demand strength. On financing, Equinix intends to lean on debt rather than equity, citing low net leverage of 3.5% and room to increase it while preserving investment-grade status. Leonetti said the company is also exploring "innovative vehicles" to finance growth within the retail colocation balance sheet but ruled out equity issuance as an inefficient source of capital.
xScale Deprioritized as Hyperscale Economics Compress
On the off-balance-sheet xScale joint venture platform, Leonetti signaled a clear capital allocation preference away from large-footprint hyperscale deployment and toward Equinix's core retail colocation business. He described the xScale segment as "now starting to be a crowded portfolio" with many competitors and lower-margin, kilowatt-based economics, contrasting it with retail colocation's small-footprint, high-return model. "xScale is part of the portfolio, retail is the focus," he said, suggesting the segment will remain a secondary rather than primary growth vector going forward.
No Signs of Hyperscaler Capex Slowdown, But Training Seen as the Vulnerable Segment
Asked about resilience to a potential pullback in hyperscaler spending, Leonetti pointed to hyperscaler and neocloud capital expenditure running 80% higher this year than the prior year, with no evidence yet of deceleration. If a slowdown does materialize, he said it is more likely to hit large language model training capacity than inference, which is where Equinix's business is concentrated. He noted the company's flexible cost structure, ability to delay capex, and low debt burden as mitigants, saying "we should be able to react better than other players" in an adverse scenario, while maintaining that such a scenario is not currently visible in the data.
M&A Focused on Technology Tuck-Ins, Not Scale
On capital deployment beyond organic buildout, Leonetti said M&A will remain targeted at technology tuck-ins that complement the interconnection business rather than large-scale acquisitions, citing the recent atNorth deal in the Nordics as an example of opportunistic land-and-power acquisition rather than a template for broader consolidation. Regionally, he noted the U.S. is growing modestly faster than EMEA and APAC, driven less by superior demand and more by better P&L economics and greater access to available space and power domestically, with roughly 60% of platform participants maintaining a global, multi-region presence.