A Trillion Dollars and the Week AI Got Serious About Everything
There are weeks in technology that feel like accumulation, and weeks that feel like acceleration. This was one of the latter. By the time NVIDIA's GTC conference opened on Monday March 16, several of the week's most significant deals had already closed. Legora, a two-year-old Swedish legal AI startup, had tripled its valuation to $5.55 billion in a single round. KAST, a stablecoin neobank barely 18 months old, had pulled in $80 million from QED Investors, Left Lane Capital, and DST Global. Mira Murati's Thinking Machines Lab had signed a partnership with NVIDIA that appears to be worth tens of billions of dollars in chips.
Then came GTC, and Jensen Huang took the stage at San Jose's SAP Center to tell 30,000 attendees — and anyone streaming — that he now sees at least $1 trillion in orders for NVIDIA's Blackwell and Vera Rubin chips through 2027. Last year, the figure was $500 billion. The doubling happened in roughly 12 months. By any measure, that is a fast-moving number.
What made this week notable was not any single announcement, but the pattern across all of them. AI is no longer building infrastructure for its own sake. It is landing in law firms, payment rails, frontier research labs, and agentic software platforms. The construction phase is not over. But the deployment phase has clearly begun.
AI Infrastructure: GTC and the $1 Trillion Signal
NVIDIA's GTC 2026 conference opened March 16 in San Jose, drawing more than 30,000 attendees across 450 sponsors and 1,000 sessions. The keynote from CEO Jensen Huang ran close to two hours and covered a sweep of announcements across chips, AI agents, robotics, and space. But the number that cut through everything else was straightforward: Huang said he sees at least $1 trillion in purchase orders for Blackwell and Vera Rubin systems through 2027.
orders through 2027
12 months ago
Vera Rubin platform
at GTC open
The Vera Rubin platform was the centrepiece. Huang walked the audience through a system comprising seven chips, five rack-scale systems, and what he called a new supercomputer for agentic AI. The platform delivers 10 times more performance per watt than Blackwell, which matters significantly when energy cost is one of the primary constraints on AI infrastructure expansion. Azure was named as the first hyperscaler to bring up Vera Rubin NVL72 systems, with global rollout continuing through the year.
Huang also unveiled the Groq 3 Language Processing Unit — the first product from the AI chip startup NVIDIA acquired in a $20 billion deal last December. The Groq 3 LPX rack, designed to sit alongside Vera Rubin rack-scale systems, improves tokens-per-watt performance of Rubin GPUs by 35 times, Huang said. It is expected to ship in Q3 2026.
Two other GTC announcements are worth flagging. First, agentic AI was the conference's defining theme. Huang referenced it a dozen times, and NVIDIA announced full support for OpenClaw, the open-source agent platform that Huang described as the fastest-growing open-source project in history. A single command deploys an AI agent. Multiple agents can work as teams. The implication for compute demand is not subtle: agents generate tokens continuously, not just when a user types a prompt.
Second, NVIDIA announced Vera Rubin Space-1 — a module designed to support AI data centers in orbit. It is an early-stage initiative, and the technical challenges around radiation hardening are real. But it signals where NVIDIA thinks the physical constraints on data center real estate eventually lead.
Thinking Machines Lab secures gigawatt-scale NVIDIA deal
The partnership between NVIDIA and Thinking Machines Lab, announced March 10, was the week's most consequential deal before GTC even opened. Mira Murati's AI research startup, which raised $2 billion in seed funding and carries a $12 billion valuation, agreed to deploy at least one gigawatt of NVIDIA's Vera Rubin systems starting in early 2027. NVIDIA also made what both companies described as a "significant investment," with the Financial Times reporting the chip supply component alone is worth tens of billions of dollars.
One gigawatt of AI compute capacity costs roughly $50 billion to build, according to Huang's own estimates. That is not a startup investment. It is a frontier lab commitment. The deal comes after a turbulent stretch for Thinking Machines — three co-founders left for OpenAI and Meta between October 2025 and January 2026. The NVIDIA partnership changes the narrative. Whatever the talent departures cost, the lab is clearly still operating at frontier scale.
Startups: AI Lands in Law, and the Vertical Bet Pays Off
If GTC confirmed that AI infrastructure spending is accelerating, Legora's Series D confirmed something else: AI is now generating the kind of traction in professional services that justifies frontier valuations. The Swedish legal AI company raised $550 million on March 10, led by Accel, with participation from Benchmark, Bessemer Venture Partners, General Catalyst, ICONIQ, Redpoint Ventures, Y Combinator, Bain Capital, Menlo Ventures, and Salesforce Ventures, among others. The round valued Legora at $5.55 billion — triple its October 2025 valuation of $1.8 billion.
The numbers behind the raise are instructive. Legora now serves more than 800 law firms and legal teams across 50 markets. The platform is used by tens of thousands of lawyers daily, with the company having grown from 40 to 400 employees in a year. Built primarily on Anthropic's Claude, it is designed for complex legal work — structured document review, bulk edits, agentic multi-step tasks, and citation-backed research — rather than simple question-and-answer queries. Deposition reviews that previously took 20 hours are reportedly running in under two. In-house teams are replacing external counsel reviews that cost $1,200 per hour with work done internally in minutes.
Legora's nearest competitor, Harvey, is reportedly seeking to raise at an $11 billion valuation. Both are on nearly identical revenue trajectories, according to Dealroom data. That two European-origin legal AI companies are racing toward $10 billion-plus valuations, backed by every major US venture firm, says something clear about where AI monetisation is actually happening in professional services.
The US expansion is the immediate priority. Legora is opening new offices in Houston and Chicago alongside its existing New York and Denver presence, and expects more than 300 US employees by year-end.
Fintech & Crypto: The Stablecoin Neobank Arrives
The week's most telling fintech story was KAST, and the numbers behind it are worth sitting with for a moment. The company was founded in July 2024. By March 9, 2026 — 18 months later — it had over a million users, was processing nearly $5 billion in annualised transaction volume, and had just closed an $80 million Series A led by QED Investors and Left Lane Capital, valuing it at $600 million. Revenue has been growing 15 to 20 per cent month-on-month.
KAST is not a crypto exchange. It is built on stablecoin rails rather than traditional correspondent banking infrastructure, offering USD-denominated accounts and payments across 190 countries. The pitch is practical: faster cross-border transfers, dollar-denominated accounts for people in markets with currency instability, payroll and payout tools for businesses operating internationally. The $80 million will fund expansion across Latin America, North America, and the Middle East.
The market context gives the deal additional weight. Global stablecoin transaction volume grew 72 per cent last year to more than $33 trillion, according to Artemis Analytics, exceeding the combined on-chain settlement volumes of major card networks. Stripe acquired stablecoin infrastructure provider Bridge for $1.1 billion in February 2025. Visa is expanding stablecoin-linked cards through Bridge to over 100 countries. PayPal operates its own stablecoin. The category is not experimental. The question is which consumer-facing platforms will hold the relationship with end users as the infrastructure matures. KAST is one of the clearest bets that the answer might not be one of the existing incumbents.
What will matter for KAST's trajectory is whether it can differentiate against Revolut, which already has stablecoin features, and against the infrastructure players now moving into consumer products. Eighteen months in, the growth rate is hard to argue with. Whether it continues as competition intensifies is the more open question.
Markets: Agents, Monetisation, and the Next Question
The macro read from this week is relatively clear. AI infrastructure spending is not slowing. If anything, GTC confirmed that the capex cycle is still accelerating. The $1 trillion order figure from Huang is a demand signal, not just a supply projection. Hyperscalers are committing capital, frontier labs are signing gigawatt-scale compute deals, and enterprises are beginning to embed AI into core operational workflows rather than running pilots alongside them.
The question markets are now asking — and this is audible in how technology stocks have moved around earnings and conference announcements — is about the second layer. Infrastructure spending benefits NVIDIA and the data center supply chain immediately. But the returns from that infrastructure have to materialise somewhere. The companies that can demonstrate measurable productivity gains, sustained revenue growth from AI features, and defensible positions in vertical markets are separating from those still relying on general AI exposure narratives.
| Sector | This week's signal | What it suggests |
|---|---|---|
| AI infrastructure | NVIDIA GTC: $1T orders, Vera Rubin full production, Groq 3 LPU announced | Demand shows no near-term ceiling; energy efficiency now the key metric |
| Frontier AI labs | Thinking Machines locks in 1GW NVIDIA deal; Murati lab still at frontier scale | Compute access is the new credibility signal for serious AI competitors |
| Vertical AI (legal) | Legora raises $550M at $5.55B; 800 customers, 400 staff, real measurable ROI | Professional services are where AI monetisation is actually happening |
| Stablecoin fintech | KAST closes $80M at $600M valuation; 1M users, $5B annualised volume | Consumer-facing stablecoin platforms are attracting serious institutional capital |
| Agentic AI | OpenClaw declared fastest-growing open-source project; NVIDIA announces full support | Agent deployment is moving from demos to developer infrastructure |
There is one structural note worth adding. Agentic AI — the shift from AI that answers questions to AI that completes tasks autonomously over time — is the thread running through almost everything this week. NVIDIA is repositioning its hardware roadmap around agent workloads. OpenClaw is becoming infrastructure. Legora is building agentic workflow tools for legal teams. KAST is running agents across compliance and operations. The common direction is unmistakable: 2026 appears to be the year agents move from research demos into production systems.
Five Themes That Defined the Week
Huang's projection at GTC doubled the figure from 12 months ago. The demand driving those numbers is not speculative. It is committed capex from the world's largest cloud providers.
Thinking Machines' gigawatt deal with NVIDIA signals something specific: serious AI competitors in 2026 are defined by their compute commitments, not just their benchmarks or talent.
Legora's $5.55B valuation is not hype. It reflects real productivity gains at 800 customers and a vertical market worth $1 trillion globally where AI adoption is still below 20%.
KAST's $80M raise at 18 months old is one signal. Stripe, Visa, and PayPal moving aggressively into the same space is the structural context behind it.
From OpenClaw support at GTC to Legora's multi-step legal workflows, agents are crossing the line from interesting demos to the thing enterprises actually run on.
Law, finance, cybersecurity, logistics — the sectors attracting the largest AI funding rounds are those where domain-specific models create clear, measurable ROI over general-purpose tools.
Sources
| Topic | Source | Key data point | URL |
|---|---|---|---|
| NVIDIA GTC 2026 keynote — $1T orders | CNBC | Huang projects $1T in Blackwell + Rubin orders through 2027 | cnbc.com |
| GTC 2026 live coverage | NVIDIA Blog | Vera Rubin platform: 7 chips, 5 rack-scale systems; Feynman next; Space-1 announced | blogs.nvidia.com |
| GTC $1T analysis | TechCrunch | Huang: "I see through 2027, at least $1 trillion" — doubles 2025 projection | techcrunch.com |
| GTC 2026 — AI agents & OpenClaw | CNN Business | NVIDIA announces full OpenClaw support; agents as primary GTC theme | cnn.com |
| Thinking Machines Lab + NVIDIA deal | Axios | Multiyear partnership; 1GW Vera Rubin deployment early 2027; NVIDIA makes significant investment | axios.com |
| Thinking Machines deal detail | TechCrunch | FT reports deal worth tens of billions; company valued at $12B; raised $2B seed | techcrunch.com |
| Legora $550M Series D | Bloomberg | $5.55B valuation; led by Accel; triples October 2025 valuation of $1.8B | bloomberg.com |
| Legora detail + ROI data | Menlo Ventures | 800 customers; 80% of legal tasks within AI reach; deposition review cut from 20hrs to under 2 | menlovc.com |
| Legora Series D — Reuters / Crunchbase | Crunchbase | Full investor list; Harvey comparison; $816M total raised since inception | crunchbase.com |
| KAST $80M Series A | PRNewswire (official) | QED + Left Lane led; $600M valuation; 1M users; $5B annualised volume; 15–20% MoM growth | prnewswire.com |
| KAST stablecoin market context | CoinDesk | $33T stablecoin volume; 72% YoY growth; Stripe/Visa/PayPal competitive context | coindesk.com |
| KAST funding detail | FinTech Global | Peak XV, HSG, DST Global participating; expansion plans for LatAm, MENA, North America | fintech.global |
This wrap goes out every week — verified, no noise, just what actually mattered.
