💰 OpenAI cap table leak shows early investors up to 140x

A cap table circulating in venture circles outlines potential returns for OpenAI investors. The data was compiled from public filings and secondary market estimates, and attributed to investor Sheel Mohnot.

🖱 Early angels — about 140x. Includes names like Reid Hoffman and Peter Thiel, who entered at the very beginning.

🖱 Sound Ventures — about 43x. Ashton Kutcher’s fund invested $20–30M early.

🖱 Khosla Ventures — about 30x. Invested $50M in 2019 for a 5% stake, now around 0.18%.

🖱 Microsoft — about 17.6x. Entered in 2019 and remains the key strategic backer.

Most other investors are shown below 10x in the same table.

The numbers come from mixed sources and secondary estimates, not official disclosures.

✔️@venture

⚔️ OpenAI claims Anthropic overstates revenue by about $8B

OpenAI says Anthropic inflates its annual run rate, putting the gap at about $8B. The claim surfaced through a leaked internal memo after reports that Anthropic crossed $30B while OpenAI sits around $24–25B.

According to the memo, Anthropic reports gross revenue from partnerships with Google and Amazon instead of net figures. That means cloud provider shares are not deducted, which lifts the headline number.

If adjusted, Anthropic’s run rate would drop below OpenAI’s by a few billion.

✔️@venture

🚀 Amazon buys Globalstar for $11.6B to build direct-to-device satellite network

Amazon will acquire Globalstar for about $11.6B, offering $90 per share in cash or stock. The deal implies a 117% premium and is expected to close in 2027.

The acquisition adds satellites, spectrum licenses, and infrastructure to Amazon’s Leo network. The goal is to launch direct-to-device connectivity starting in 2028. Globalstar currently operates about 24 satellites and already powers Apple’s Emergency SOS, with Apple holding a 20% stake.

Amazon has launched over 240 satellites so far, still far behind Starlink with more than 10,000 satellites and over 9M users.

The deal gives Amazon both capacity and spectrum to compete in direct-to-cell services.

✔️@venture

📈 Allbirds pivots from shoes to GPU rental after raising $50M

Allbirds is shifting its business to GPU infrastructure after struggling in footwear. The company faced rising losses, falling demand, and store closures as the brand lost traction.

It is now selling assets and raising $50M to buy GPUs for rental. The plan is to enter the GPU-as-a-service market.

After the announcement, Allbirds stock jumped more than 400%.

✔️@venture

🚖 Uber breaks its asset-light model with $10B robotaxi push

Uber built its business on not owning assets. Cars were never on its balance sheet, and drivers handled everything themselves. Now that model is changing.

The company is committing $10B to robotaxi development and fleet purchases, with another $7.5B reserved for expansion. The plan is to deploy thousands of autonomous vehicles as a starting point.

Autonomy forces a different structure. To run robotaxis, Uber needs to own and operate the fleet.

✔️@venture

💻️ Jensen Huang says China can build Claude-level AI and calls for cooperation

In a new interview with Dwarkesh Patel, NVIDIA CEO Jensen Huang said China already has enough compute and energy to train models at the level of Claude Mythos. The AI race is shifting, and the US needs to stay open while holding its lead.

Export restrictions became a key factor. Limits on chip supply pushed China to accelerate its own ecosystem, including Huawei and SMIC, increasing independence instead of slowing progress.

Demand for TPU and Trainium is largely tied to Anthropic. Early deals with Amazon and Google locked the company into their infrastructure, creating artificial demand for these chips. Nvidia did not invest at that stage, which Huang called a missed opportunity.

The balance between policy and technology is starting to shape the market.

✔️@venture

🚀🔋 Alphabet could make $100B on SpaceX IPO stake

Alphabet’s early investment in SpaceX may turn into a $100B windfall. A new filing shows Google held 6.11% at the end of 2025, likely diluted to around 5% after the xAI merger.

At a $2T IPO valuation, that stake would be worth about $100B. Even a 0.05% holding could make an investor a billionaire at that level.

SpaceX is targeting a listing that could raise up to $75B, potentially the largest IPO ever.

✔️@venture

📈 Madison Air raises $2.23B IPO

Madison Air Solutions raised $2.23B in its US IPO at a $13.2B valuation.

It is the largest industrial IPO of the century.

The company builds ventilation, air quality, and cooling systems for data centers, commercial, and residential use. Its portfolio includes Nortek Air Solutions and Big Ass Fans.

Demand is driven by data center growth, where cooling has become a core constraint.

✔️@venture

📱 Snap cuts 1,000 jobs as AI replaces routine work

Snap will lay off about 1,000 employees, around 16% of staff, and close more than 300 open roles. The move follows pressure from activist investor Irenic Capital, which holds about 2.5%.

The company says AI now generates over 65% of new code, allowing smaller teams to handle core work. Snap expects more than $500M in annual cost savings by the second half of 2026.

The cuts come as Snap continues investing in products like its AR glasses, which have already cost over $3.5B and remain under scrutiny.

✔️@venture

⚡️ Investor turns $225M into $5.5B betting on AI power bottleneck

Leopold Aschenbrenner explained the logic behind his largest bet. His Bloom Energy position alone is now close to $2B after Oracle’s 2.8 GW fuel cell deal.

🖱 2022: GPT-4 cluster used ~10 MW and cost ~$500M
🖱 2024: ~100 MW, around 100,000 GPUs, cost in the billions
🖱 2026: ~1 GW, comparable to a nuclear reactor
🖱 2028: projected ~10 GW
🖱 2030: projected ~100 GW, over 20% of US electricity output

AI compute has been scaling roughly every 12–18 months. Training is only part of the demand, inference adds multiples on top.

US electricity production grew only ~5% over the last decade. Grid constraints, transformer shortages, and delayed data centers are already visible.

The bet: generate power directly at data centers with fuel cells, bypassing the grid.

The constraint is shifting from compute to electricity.

✔️@venture

In 2000, Netflix wanted to sell itself to Blockbuster for $50 million.

Blockbuster laughed at them.

Today, Netflix generates $50 million in profit in less than a day.

✔️@venture

👾 OpenAI loses Kevin Weil and two other leaders amid internal reshuffle

Kevin Weil is leaving OpenAI along with two other senior leaders. He served as VP of research and previously as chief product officer, and recently led the OpenAI for Science initiative.

The company is scaling back side projects, and that team has been integrated into other units. The shift appears to be tied to his departure.

Bill Peebles, head of Sora, and Srinivas Narayan, CTO for enterprise applications, are also leaving.

✔️@venture

🧩 Microsoft plans to build frontier AI models by 2027

Microsoft said it aims to develop its own frontier models by 2027 to compete with OpenAI and Anthropic.

The company no longer wants to rely only on external AI providers. A previous agreement with OpenAI restricted Microsoft from building large general-purpose models in exchange for access to ChatGPT.

That restriction was removed last year after the contract was revised.

✔️@venture

🟢 OpenAI buys Hiro to build personal finance layer inside ChatGPT

OpenAI acquired personal finance startup Hiro, a tool that tracks income, expenses, and debt and generates dashboards and recommendations. The deal closed in October 2025, but was disclosed only now. The ~10-person team joins OpenAI, and Hiro shuts down on April 20 to become a native feature.

This follows another October deal for finance assistant Roi. OpenAI is building vertical layers inside ChatGPT, adding finance alongside earlier efforts like healthcare.

🖱 Finance is a core agent use case. Tasks like paying bills or checking subscriptions require action, not just answers
🖱 Domain expertise matters. Financial logic and regulation are easier to acquire through teams than build from scratch
🖱 User data becomes richer. Spending and income data provide strong signals for personalized agents

There are constraints. Financial advice is tightly regulated, and existing fintech players are building their own AI tools.

✔️@venture