AI & Tech News Roundup — April 29, 2026
NVIDIA crosses $5 trillion. Big tech commits $660B to infrastructure. Liquid cooling becomes the new baseline. And the grid is the bottleneck nobody budgeted for. Here is what mattered this week in AI and compute.
AI & Tech News Roundup — April 29, 2026
Ultra Labs tracks the signals that matter for compute infrastructure, Bitcoin mining, and the AI buildout. This roundup publishes every 2-3 days when the news warrants it.
1. NVIDIA Crosses $5 Trillion. The Infrastructure Market Is Just Getting Started.
NVIDIA hit a $5 trillion market cap on April 24, driven by a 4.3% single-day surge. Q1 2026 Data Center revenue came in above $30 billion, with gross margins above 75%. CEO Jensen Huang continues to project a $1 trillion global AI infrastructure upgrade market, and the quarterly numbers are not disagreeing with him.
For context: that is a single product segment generating more quarterly revenue than most tech companies generate annually. Intel posted its best single-day performance since 1987 the same week, up 24% on Q1 results of $13.58B. AMD jumped 14%. The semiconductor sector as a whole is on track for its third consecutive year of double-digit growth, approaching $1 trillion in global scale.
The Ultra Labs angle: NVIDIA's rack density problem is the same problem Bitcoin miners solved years ago. The liquid cooling systems keeping Antminer S21 XP Hydro units running at scale are what hyperscalers are now engineering into their AI facilities. Miners who understand heat management at density have a real operational edge as this market develops.
Sources: Nvidia Hits $5 Trillion Market Cap · Semiconductors & AI Chips Weekly Briefing Apr 24
2. Big Tech Is Committing $660-690 Billion to Infrastructure in 2026
The five largest US cloud and AI infrastructure companies have committed $660-690 billion in capital expenditure for 2026, nearly double 2025 levels. The World Economic Forum puts the broader AI hardware buildout figure at $7 trillion over the coming years.
That is not software spending. That is physical infrastructure: land, power, cooling, fiber, and compute hardware.
TSMC reported a 35% year-over-year revenue surge in Q1 2026, driven by 80% growth in its advanced CoWoS packaging technology — the process that makes NVIDIA Blackwell chips viable at scale. TSMC is now expanding into Arizona specifically to de-risk US supply chains. They also unveiled A13 and N2U process nodes targeting 2029 production.
The constraint is no longer chip design. It is manufacturing, power delivery, and thermal management — problems deeply familiar to anyone who has run a large-scale mining operation.
Sources: AI Investments: $7 Trillion Buildout · TSMC and Nvidia Packaging Expansion
3. Liquid Cooling Just Became the Default
Liquid-cooled capacity equaled air-cooled capacity in data centers globally in 2025. By end of 2026, it is projected to double it. NVIDIA Blackwell racks are already pushing past 100 kW per rack — a density level where air cooling is physically impossible, not just inefficient.
TechRepublic reports that hyperscalers are now treating entire campuses as integrated thermal systems rather than optimizing individual buildings.
This is the moment Bitcoin hydro miners have been building toward. Immersion and direct liquid cooling at high rack densities is exactly what next-generation AI compute requires. The operational playbook already exists — it just lives inside mining facilities right now.
Ultra Labs runs 58 hydro-cooled miners in the Pacific Northwest. That is not legacy infrastructure. That is a template.
Source: AI Demand Forces Rethink of Data Center Power and Cooling
4. 7 GW of US AI Capacity Is Stuck in the Queue
Up to 7 GW of announced US data center capacity remains unbuilt as of April 2026, stalled by grid equipment shortages — specifically transformers, switchgear, and battery systems with 18-36 month lead times. Tech Insider puts the capacity at risk figure at 7 GW of delayed or cancelled projects.
The bottleneck is not compute. It is not cooling. It is not even land. It is the utility grid.
This is where energy-proximate miners have a structural advantage. Facilities with existing grid interconnects, substations, and power contracts are worth significantly more than the hardware inside them. Access to power is the moat — and most hyperscalers are discovering this for the first time. Mining operators have been living it for years.
Source: U.S. AI Data Center Delays: 7 GW Capacity Crisis · AI 2026: Data Centers Restart Growth of U.S. Electrical Grid
5. The Mining-to-AI Pivot Is Repricing the Sector
HIVE Digital Technologies raised $115 million in fresh capital specifically for GPU fleet expansion and AI data center development. Keel Infrastructure — formerly Bitfarms — completed the sale of its Paraguay site and declared itself an infrastructure-first HPC/AI operator. Both stocks gained around 11% on the news.
CoinShares projects that publicly listed Bitcoin miners could derive up to 70% of revenues from AI by December 2026, up from roughly 30% today.
Core Scientific is already there: its CoreWeave partnership covers 200 MW of AI capacity under a 12-year deal, with a projected 71% net return on the contract.
The market is not pricing these companies as Bitcoin miners anymore. It is pricing them as compute infrastructure operators — which is exactly what the good ones are becoming.
Sources: Hive and Bitfarms Lead Bitcoin Miner AI Rally with 11% Gains · AI Pivot Accelerates as HIVE Raises Fresh Capital and Keel Reshapes Portfolio
Ultra Labs is a US Bitcoin mining and technology company, operating hydro-cooled miners in the Pacific Northwest and building on Cardano and Midnight. Read more: Bitcoin Miners Are Becoming AI Infrastructure Companies · Bitcoin Mining in 2026: Air, Hydro, Immersion, and the Bitaxe Revolution