The fastest SaaS company ever just doubled again

Cursor Doubles

‍

There's a number circulating that should make every private market investor recalibrate their mental models: Cursor reportedly hit $2 billion in annualized revenue in February, doubling its run rate in three months.

To put that in context: Cursor launched its product roughly two years ago. It hit $1B ARR in November 2025. It doubled to $2B by February 2026. That's a trajectory that makes Wiz, Deel, and Ramp look leisurely by comparison. The company, valued at $29.3B after its $2.3B Series D last fall, is now generating revenue at a pace that would make it one of the most valuable enterprise software companies in the world if it were public.

But the headline number isn't the story. The composition of that revenue is. According to Bloomberg, roughly 60% of Cursor's revenue now comes from corporate customers — a sharp pivot from its origins as a tool beloved by individual developers paying $20–40 per month. That shift from consumer-ish to enterprise fundamentally changes the company's durability profile. Enterprise revenue is stickier, higher-margin on a per-seat basis, and far more predictable. It also means Cursor isn't just riding a wave of developer enthusiasm. It's becoming embedded in corporate engineering workflows at a pace that would make traditional SaaS companies envious.

For private market investors, Cursor represents something genuinely new: a company whose growth rate is so extreme that traditional valuation frameworks start to break down. At $29.3B on $2B ARR, it's trading at roughly 15x revenue — a number that looks almost modest given the growth rate. If Cursor sustains even half this pace, it will be the most valuable private software company in the world within a year.

The broader signal is that AI-native developer tools are a platform shift. And the companies that own that shift are accumulating value at a pace the private markets haven't seen before.

The Musk Debt Reset

Buried beneath the Anthropic headlines this weekend was a significant financial development: X and xAI plan to repay approximately $17.5 billion in combined debt in full.

The mechanics matter. X inherited about $12.5 billion in debt from Musk's acquisition. xAI added another $5 billion through bonds and loans last June. The combined entity, xAI Holdings, has been servicing tens of millions in monthly interest payments — a drag on cash flow for a company that Bloomberg reports burns roughly $1 billion per month on data center infrastructure, chips, and talent.

The repayment is being facilitated by Morgan Stanley, which handled both companies' original debt raises. Notably, xAI's $3 billion in high-yield bonds will be redeemed at approximately 117 cents on the dollar — a premium that compensates bondholders for losing two years of expected interest income.

Where is the money coming from? xAI raised $20 billion in a Series E in January. SpaceX acquired xAI in February at a $250 billion valuation. And SpaceX itself is reportedly preparing a confidential IPO filing as early as this month, targeting a June listing.

The strategic logic is clean: consolidate the Musk empire's balance sheet, eliminate the drag of legacy acquisition debt, and present a cleaner capital structure ahead of what could be the largest tech IPO in history. For private market investors holding xAI or SpaceX positions, this is a de-risking event. The debt overhang that made xAI's financial picture murky is about to disappear.

Deal Flow

Reflection AI — reportedly $2B+ at $20B+ valuation

Per the Financial Times, the Nvidia-backed open-source AI lab founded by former DeepMind researchers is courting investors for another massive raise — just five months after closing $2B at an $8B valuation. If the reporting holds, that's a 2.5x valuation jump in under six months for a company that hasn't yet released its frontier language model. The bet here is on the team: CEO Misha Laskin led reward modeling for Gemini, and co-founder Ioannis Antonoglou co-created AlphaGo. But the pace of valuation escalation in frontier AI continues to test the boundaries of conventional venture risk assessment.

Revel — $150M Series B

Led by Index Ventures with participation from Redpoint Ventures and Thrive Capital. Revel is building a software platform for hardware test and control across aerospace, defense, robotics, and industrial markets. The raise reflects a broader theme: as AI moves from software into physical systems — autonomous vehicles, robots, drones, satellites — the testing infrastructure for those systems becomes a critical bottleneck. Revel is the picks-and-shovels play for the physical AI buildout.

M&A to Watch

Canva acquires Cavalry and MangoAI — The design platform picked up a 2D motion animation startup and a stealth video ad optimization company. Canva continues its strategy of acqui-hiring specialized creative AI teams to build an end-to-end visual content platform.

Gilead acquires Arcellx for $7.8B — The largest deal of the week by far, as the biopharma giant buys a cell therapy biotech. Not a tech deal, but instructive for private market watchers: this is what a strategic acquisition at full price looks like in a sector where clinical validation commands premium multiples.

What We're Reading on Anthropic

Dean W. Ball on the skirmish (Hyperdimensional)

Ball frames the Anthropic-Pentagon dispute as the first real test of who controls frontier AI in America. His core argument: every institution involved — Anthropic, the Pentagon, the White House, OpenAI — behaved erratically and without clear strategic doctrine. The piece is worth reading for its refusal to lionize any party. The Pentagon's "supply chain risk" designation was disproportionate. Anthropic's negotiating posture, while principled, was arguably naive about how government procurement actually works. And OpenAI's opportunistic contract signing, while commercially rational, undermines its own stated commitments to safety. The through-line: we're making up the rules for AI governance in real time, and nobody has a playbook.

Ben Thompson on alignment (Stratechery)

Thompson takes a different angle than Ball. He acknowledges Anthropic's concerns as legitimate but argues the company's position is fundamentally untenable in a world where US adversaries are actively developing autonomous military AI capabilities. Thompson's framing: Anthropic is trying to apply peacetime ethics to wartime procurement, and the result is a principled stance that may be strategically self-defeating. The tension between these two analyses — Ball's "everyone was wrong" vs. Thompson's "Anthropic was right on values but wrong on strategy" — captures the actual complexity that got flattened in the weekend's social media discourse.

Quick Takes

MongoDB drops 24% after hours despite Q4 beat — Revenue up 27% YoY to $695M, EPS of $1.65 vs. $1.45 estimate. But Q1 FY2027 guidance came in light: $659–664M revenue vs. $662.5M consensus. The "beat and lower" pattern strikes again. For private market investors watching eventual IPO pricing: even 27% growth at scale gets punished when the forward outlook softens. Valuation is about trajectory, not snapshots.

Nvidia plans $2B investments in photonics makers Lumentum and Coherent — Strategic investments to secure its optical networking supply chain as data center interconnect bandwidth becomes the next bottleneck. Follow the capital: when Nvidia starts investing in its suppliers, it's telling you where infrastructure constraints will emerge next.

Software engineering jobs up 11% YoY — Citadel Securities data shows developer job postings are rising, not falling, despite the vibe coding wave. The counterintuitive logic: lowering the cost of building software increases demand for the people who can build it well. If you lower the price of something that was supply-constrained, demand goes up.

Qualcomm launches Snapdragon Wear Elite for AI wearables — A 3nm SoC that supports up to 2B-parameter models on-device. The AI wearables race is about to intensify: Google, Samsung, Meta, Apple, and Motorola are all developing smart glasses and wearable AI devices. Qualcomm is positioning itself as the picks-and-shovels play for all of them.

📊 Data Point of the Day

90 days

That's how long it took Cursor to double its annualized revenue from $1 billion to $2 billion. For comparison, it took Slack roughly 18 months to make the same jump. Zoom did it in about 5 months, but that was during a pandemic. Cursor is doing it in a normal market — if you can call the current AI investment environment "normal."

🎓 Manual

Make-Whole Premium

When a company repays bonds before maturity, bondholders lose the future interest payments they expected to earn. A make-whole premium compensates them — typically calculated as the present value of remaining interest payments plus the principal, discounted at a Treasury rate plus a small spread.

xAI's $3B bond redemption at 117 cents on the dollar is a practical example: bondholders are receiving 17% above face value because the debt was expected to remain outstanding for at least two more years. The make-whole provision is standard in high-yield corporate bonds and effectively represents the "exit fee" for early repayment.

What We're Watching

1. CrowdStrike earnings (today after close) — CRWD reports Q4 FY2026 results tonight. After MongoDB's guidance-driven selloff, the market has zero patience for "beat and lower" in enterprise software. Watch for commentary on AI-driven security workloads and whether the Falcon platform expansion is translating into durable net-new ARR growth.
2. MWC fallout — Beyond the headline product launches, watch for signals on the 6G timeline (Qualcomm is targeting 2029), the competitive dynamics in AI wearables silicon, and whether any handset OEMs announce GrapheneOS or privacy-focused Android partnerships following Motorola's move with the GrapheneOS Foundation.
3. SCOTUS punts on AI copyright  — The Supreme Court declined Monday to hear the Thaler case, leaving intact lower court rulings that AI-generated works without human authorship can't be copyrighted. The narrow ruling matters less than what it leaves unresolved: the court explicitly didn't address works created with AI assistance, only works created by AI autonomously. That distinction is where the money is. A separate case in Colorado (Allen v. Perlmutter) is testing whether human-directed AI outputs qualify for protection — and the outcome will directly affect the IP defensibility of every AI-native content company in venture portfolios. If AI-assisted works can't be copyrighted, the moat around AI creative tools shrinks dramatically.

‍