Liquidity flowed where compute lives. SoftBank’s $5.8 billion Nvidia sale financed its $22.5 billion commitment to OpenAI, which disclosed $20 billion ARR and a staggering $1.4 trillion in infrastructure deals. Anthropic answered with its $50 billion Fluidstack partnership. CoreWeave’s results and Quantinuum’s breakthrough reinforced the same trend: investors are rotating into hard assets that power intelligence. Even Einride’s SPAC, Cursor’s $29 B round, and Oura’s growth forecast fit the pattern — compute and capital converging.
From world-saving mission to private-sector beast
Last week, OpenAI found itself caught between two contradictory narratives. On one side, it is developing life‐saving AI, contributing to national defence, and undertaking projects so consequential that a government stepping in to ensure success seems plausible. On the other side, it behaves very much like any top‐tier private company: bristling when outsiders ask for more than a headline.
This week Sam Altman disclosed that OpenAI expects $20 billion in annualised revenue run‐rate (ARR) by year-end and has locked in roughly $1.4 trillion of data-centre commitments over the next eight years. He named forthcoming revenue streams: enterprise offerings (OpenAI claims already serving one million business customers), new consumer devices, robotics, and even a cloud compute business.
In parallel, OpenAI has asked the U.S. government to expand the 35 % tax credit created under the CHIPS Act (via the Advanced Manufacturing Investment Credit) so that it applies not just to semiconductor fabrication, but also to AI servers, data centres and grid components. In a letter to the White House’s science policy director, its chief global affairs officer argued that broadening the credit would lower capital cost, de-risk early investment, and accelerate the AI build-out in the U.S.
Would you trade Nvidia for OpenAI?
That’s exactly what SoftBank just did — offloading $5.8 billion worth of Nvidia stock to help fund a $22.5 billion commitment to OpenAI.
SoftBank’s latest earnings reveal a company liquidating whatever’s easy to sell and borrowing heavily to keep its AI bet alive.
The Nvidia stake wasn’t a crown jewel. It was a short-term trade, not a legacy position. According to MST Financial’s David Gibson, SoftBank likely started buying in early 2024, when Nvidia traded at roughly half today’s level. The position amounted to 32.1 million shares — just 0.13 % of the company.
Anthropic’s $50 B Data-Center Bet
Anthropic announced a $50 billion partnership with U.K. neocloud provider Fluidstack to build new U.S. data centers tailored to its expanding AI workloads. The facilities — in Texas and New York — will come online through 2026 and are described as “custom-built for Anthropic,” optimized for model training efficiency.
It’s a massive infrastructure commitment, putting Anthropic in the same strategic lane as OpenAI’s Stargate and CoreWeave’s hyperscale expansion. The deal effectively locks in long-term compute sovereignty — ensuring Anthropic isn’t hostage to external cloud shortages as demand for GPUs and specialized racks continues to outstrip supply.
The IPO Window Cracks Open
After 42 days of paralysis, the U.S. government shutdown is finally ending. The Senate has passed the funding bill to reopen federal agencies, and President Trump has indicated he’ll sign it once the House gives final approval — a step expected within days.
That means the SEC can resume operations, unfreezing the IPO pipeline that’s been stuck for six weeks.
Follow our IPO Barometer for real-time updates as the window reopens and filings flood back in.
CoreWeave’s Signal: AI Cloud Economics Are Getting Stickier
Neocloud player CoreWeave posted $1.4 billion in Q3 revenue (+134% YoY) and a $25 billion backlog, operating in what it called a “highly supply-constrained environment” where demand for its AI infrastructure far exceeds capacity. Revenue diversification improved — the number of customers generating $100 million+ tripled — but debt costs rose, and Q4 guidance disappointed because of delayed data-center builds. The stock fell 9%.
The standout metric wasn’t top-line growth. It was customer behavior: one client’s 10,000-plus H100 GPU contract, set to expire soon, was re-upped two quarters early at nearly the same price.
H100s are now last-gen hardware — yet customers are willing to pay essentially flat pricing for continued access. It suggests the useful revenue life of GPUs is extending, even as newer chips roll out.
The Next Big Quantum Computer Has Arrived
Quantinuum, the $10 billion quantum leader backed by Honeywell and Cambridge Quantum, has unveiled Helios — its most advanced machine yet.
Helios marks a real engineering milestone: it packs 98 physical qubits and can deliver 48 logical, error-corrected qubits — an unprecedented 2:1 ratio. That efficiency, said UCLA professor and DCVC partner Prineha Narang, sets a new standard. Most competitors need dozens or even hundreds of physical qubits to produce a single stable one.
Quantinuum’s clients — including JPMorgan Chase — will use Helios to test real-world applications, from portfolio optimization to risk modeling, and stress-test where quantum could start delivering business value.
The company plans to scale further with Apollo, a 2029-targeted system boasting thousands of physical qubits and hundreds of logical ones, aiming for the holy grail: a fault-tolerant, commercially viable quantum computer.
Einride Drives Toward a $1.8 B SPAC Listing
Swedish self-driving freight startup Einride plans to go public via a SPAC merger that values the company at $1.8 billion, with a NYSE debut targeted for 2026. The move comes just weeks after its $100 million capital raise, signaling investors’ willingness to back autonomous-logistics plays despite a tougher mobility market.
Einride’s model fuses “freight-capacity-as-a-service” — electric, autonomous truck fleets — with a SaaS platform that orchestrates routing, charging, and compliance. The company has seen recurring revenue growth and regulatory progress in both the U.S. and Europe, even after a year of steep losses and layoffs.
The SPAC route gives Einride a faster path to scale in the U.S., where regulatory and infrastructure momentum could make 2026–27 the breakout phase for autonomous freight.
Mira Murati’s Thinking Machines Targets $50 B Valuation
Thinking Machines Lab, led by former OpenAI CTO Mira Murati, is in talks to raise new funding at a $50 billion valuation — just five months after closing its previous round at $10 billion.
Launched in February, the startup has quickly positioned itself as a bridge between open-source models and enterprise-grade performance. Its first product, Tinker, is an API that lets developers finetune open-source models for specific use cases — effectively turning public models into private, performance-optimized ones.
Murati has told investors the next step is “KPI-aligned modeling” — producing AI systems customized to the metrics that matter most to each client, from revenue acceleration to margin expansion.
Prediction Markets Go Mainstream
Polymarket and Kalshi just earned a spot on Google Finance. A select group of users will soon see prediction-market data integrated directly into Google’s finance dashboard — a milestone that blurs the line between financial markets and information markets. It’s also validation for prediction as a new form of financial signal — one that could soon compete with analyst forecasts and polling models alike.
Dealflow Snapshot — AI, Cloud, Legal, and Fintech
Developer-focused AI tool Cursor raised $2.3 billion at a $29 billion valuation, with participation from Nvidia, Google, and Accel. The funding will expand its in-house Composer model, built to compete directly with OpenAI and Anthropic in the dev-tools segment.
Vast Data signed a $1.17 billion deal with CoreWeave to power AI data management at scale — deepening the vertical integration between AI cloud and software infrastructure.
Metropolis, the computer-vision parking startup, raised $500 million at a $5 billion valuation (Series D) to expand across North America.
Clio, a legal-tech platform for law firms, raised $500 million at a $5 billion valuation and completed its $1 billion acquisition of vLex, consolidating AI-driven legal research.
Gamma, the AI-native presentation and site builder, raised $68 million at a $2.1 billion valuation (Series B, led by Andreessen Horowitz) and crossed $100 million ARR.
d-Matrix, focused on inference chips, raised $275 million at a $2 billion valuation — part of the new “post-GPU” wave of silicon innovation.
Zilch, the U.K. buy-now-pay-later platform, raised $175 million in a mix of debt and equity led by KKCG, and is exploring strategic acquisitions to consolidate the fragmented BNPL space.
Oura — maker of the smart ring — projects $1 billion in 2025 sales and is forecasting nearly $2 billion for 2026 as it doubles down on AI health analytics and international expansion.