Most Discussed
Geothermal's moment has arrived
The AI build-out has spent the past year solving for compute. The next bottleneck is showing up one layer deeper: steady, around-the-clock, decarbonised electricity, generated and stored on site at data centres. In a single week, announcements landed on nuclear-powered data centre projects, autonomous wave-powered floating compute funded by a prominent venture backer, and a 20-year, 150MW power purchase agreement routing geothermal output through the regulated grid to a major hyperscaler. Different angles, same pattern. Hyperscalers need clean baseload power, and they need it on schedule.
Two capabilities solve that. The first is baseload that does not depend on the weather: a renewable that runs above 90% of the time regardless of cloud, season or wind, with negligible fuel cost and a low carbon footprint. Geothermal is the only source that fits that description, replacing fossil generation one-for-one rather than supplementing it. The historical cost gap has closed; federal tax credits for next-generation geothermal in the US, alongside permitting reform, have unlocked commercial-scale deployment for the first time.
The second is the ability to move energy through time. As zero-marginal-cost renewables flood the grid and power prices grow more volatile, anything that shifts energy from production time to consumption time is disproportionately valuable. Storage, broadly defined, is seeing margins expand as the arbitrage between cheap-hour absorption and peak-hour discharge widens.
The infrastructure constraints are not sequential but multiplicative. Cooling advances push rack density from 50kW to 120kW+, which worsens the power bottleneck. Copper demand is 3-5x higher per data centre, with EVs, renewables and robotics all competing for the same supply. Off-grid power generation bypasses the 5-7 year grid interconnection queue, but gas turbines are now sold out with 3-5 year lead times, shifting rather than eliminating the constraint. Transformers and switchgear carry 2-3 year lead times and depend on copper. Skilled electricians are the shared bottleneck across data centres, utilities, renewables and power plants. Washington has now formally classified the grid supply chain as a national defence bottleneck, reinforcing the structural bid for any operator who can deliver dispatchable clean power outside the queue.
The investable insight across the energy stack: follow the bottleneck rotation rather than picking end-market winners, with copper, transformers, cooling and photonics as the durable choke points.
Who survives the software reset?
"This is a reset. It is not an evolution. Companies that become native AI or reinvent themselves quickly will be the companies that win. It will be winner takes all."
That was the verdict from a veteran software CEO who has run three major enterprise platforms and spent a decade inside one of the world's largest database companies. His filter for survivors is simple: watch the language CEOs use. Those trying to defend moats and wall off their user base are going to struggle, because disruption will happen in their backyard.
Q1 2026 earnings delivered the first hard rebuttal to February's "Software-mageddon" sell-off, which wiped c.$1tn from public software on the thesis that agents would bypass the application layer entirely. Three reports told the story: a cloud communications platform grew revenue 20% with voice accelerating for a sixth consecutive quarter; an observability platform grew 32% with AI-integrated customers representing 20% of the base but c.80% of ARR; and a collaboration software firm grew 32%, beating consensus by c.$100m, with AI users growing ARR at twice the rate of non-users. All three raised guidance. The pattern: AI-native customers are the heaviest spenders, non-AI revenue is also reaccelerating, and pricing is shifting from per-seat to consumption-based.
The risk underneath the sell-off is no longer theoretical. A car-rental SaaS company gave an AI agent root access to its production database. The agent deleted the entire database and all the backups, even citing its own "do not delete" policy before ignoring it. No enterprise is going to deploy autonomous agents at scale without a managed, supervised, auditable layer underneath. The names that have spent years building that layer are precisely the ones the market just cut. The right filter is leadership quality and AI-readiness, not the headline multiple.
Hyperscaler capex commitments for 2026 total c.$575bn across the three largest cloud providers alone, with one openly compute-constrained. The companies embedding AI inside their own products, treating internal deployment as both a margin lever and a live customer demo, are the ones whose forward earnings multiples look most mispriced relative to growth.
What Drove The Collective Ideas This Week
WoW: -2%
The Conviction basket slipped -2% WoW, but the thematic underpinnings strengthened beneath the headline. Geothermal went mainstream: a pure-play IPO raised nearly $2bn into a 15x oversubscribed book and closed at a $10bn+ market cap, a second hyperscaler signed its first geothermal deal, and established operators saw storage margins nearly triple. The US-China thaw triggered broker upgrades across Chinese cloud and AI infrastructure. Precious metals reshaped industrial metal cost curves, with by-product credits collapsing copper production costs 35% and delivering decade-best quarterly earnings. Frontier digital platforms posted 58% digital revenue growth and drew endorsement from a top-tier global asset manager. The IPO cycle reopened decisively, signalling institutional willingness to price duration and optionality again. On the other side, geopolitical de-escalation compressed energy risk premiums, and security infrastructure was sold despite beating and raising as a margin transition spooked holders. The -2% was positioning rotation, not thematic deterioration.
Best Content Shared This Week
🎧 Anthropic and the Economics of Intelligence
Why listen? Anthropic's CFO breaks down why compute is no longer just a cost line, but the core asset behind frontier AI. The discussion explains how multi-chip flexibility, enterprise trust and explosive token demand are turning intelligence into a high-return infrastructure business, with clear read-throughs for Amazon, Alphabet, Broadcom and Nvidia. Listen here.
📺 Agentic AI Is Already Replacing Elite Knowledge Work
Why Watch? Ken Griffin describes watching AI agents complete work in days that previously required teams of highly paid finance PhDs working for months. The important point is not automation at the margins, but that agentic AI is now moving directly into high-value decision-making and analytical workflows inside elite institutions. The shift from copilots to autonomous agents is no longer theoretical; it is already showing up in the real economy. Watch here.
Stock Of The Week
The Digital Operator Re-rating
VEON is increasingly looking less like a traditional telecom operator and more like a digital platform business embedded across emerging markets. Q1 revenues grew 17% year-on-year, while digital revenues surged 57.7% and now account for more than a quarter of group sales, driven by fintech, entertainment, ride-hailing and AI-enabled services.
The key shift is the platform model. Products like JazzCash, which issued more than 200,000 digital loans per day in Q1, are deepening customer engagement beyond connectivity, while AI1440 is layering local-language AI tools across the ecosystem. With leverage at just 1.07x, a $100m annual buyback and raised 2026 guidance, the question is how long the market continues valuing VEON as a legacy telecom rather than a scaling digital operator.
.png)
