Industry News

A Data Center Knowledge commentary by Patrycja Sobera of Unisys argues that AI projects stall when the facilities behind them can’t keep up. She lists five areas to evaluate before committing budget: whether servers, storage and networks work as one system; power and cooling capacity; security across the AI environment; whether spending should be predictable or flexible; and room to grow.

The power example is concrete: a rack built for 5 kW can’t carry a 20 kW AI node, and power and cooling have to scale together. The article also notes that a fast GPU cluster can still be held back by storage throughput or network congestion.

For enterprises, the network is often the part of that list that gets overlooked. Moving training data, replicating between sites and reaching cloud GPU capacity all depend on bandwidth and latency between facilities, not just inside the rack. Readiness planning should cover the routes between data centers, colocation and cloud providers, and how those links will scale as AI workloads grow.

Sports led the activity, with NFL and college football markets among the busiest, while crypto markets added another $2.16 billion on Kalshi. Cross-category, multi-leg “combo” markets made up the largest share of Kalshi’s volume. Separately, a PitchBook analysis estimates Kalshi could be worth as much as $42 billion ahead of a possible IPO, up from the $22 billion valuation of its May funding round. That estimate depends on the outcome of a pending Supreme Court case that could let individual states regulate the platform.

For trading firms, rising volume makes prediction markets a venue category worth evaluating for reliable order access, market data and resilient connectivity. The right design depends on the venue’s current endpoints and approved access methods. Firms can compare cloud and colocation options with routes back to their existing equities and futures infrastructure after confirming the actual trading path.

Annual data center capital expenditure is projected to climb from roughly $800 billion a year in 2026 to $1.8 trillion a year by 2050. Notably, PwC’s baseline case points to recurring chip upgrades, rather than new construction, driving most of that long-term capital spending — a sign that ongoing hardware refresh cycles, not just new-build data centers, are becoming the bigger long-term cost driver. The report also flags supply-chain risk: disrupted trade flows for chips could reduce global investment by nearly 20%, while a shift toward national tech sovereignty would redirect investment rather than shrink it overall. Global investment in AIinfrastructure to hit US$31.6 trillion through 2050 | PwC +2

For firms evaluating where to locate AI infrastructure or how to plan connectivity around it, the throughline is the same one we’ve been tracking: power and interconnection access are becoming the real constraints, not just chip supply or floor space.

Kalshi and the Polymarket US app are both federally regulated by the CFTC and are technically legal in all 50 states and Washington, D.C., though a handful of states have moved to ban prediction markets outright, with sports-specific contracts often singled out — Arizona, for instance, is pursuing legal action against Kalshi over state betting law. The regulatory picture remains a genuine patchwork even as volume grows. CBS SportsCBS Sports

Separately, mainstream finance is starting to build products around this growth: an ETF provider recently filed to launch a fund carrying exposure to both Kalshi and Polymarket, treating the platforms themselves as an investable business rather than just a venue for placing predictions. For firms in the connectivity and market-data space, the combination of rising volume and expanding regulatory legitimacy is worth tracking — it points to growing infrastructure and data demands from an asset class that barely existed at scale two years ago.