# The companies

*The companies whose bets decide what your compute costs. One page each, dated, and updated when they report earnings, sign a deal or change their minds. Old versions stay in each page's changelog, so you can see the strategy move.*

Every figure links to a filing or says it's reported. Our take is labelled as ours. A bet goes in our ledger only when we're willing to be graded on it.

## Who we cover

| Company | The bet, in one line | As of |
| --- | --- | --- |
| [Google](/blog/companies/google) | Sells slices of itself to build the thing most likely to replace Search | Nov 2, 2026 (Q3) |
| [Amazon](/blog/companies/amazon) | Borrows the money, lays the track and rents it to everyone | Nov 2, 2026 (Q3) |
| [Microsoft](/blog/companies/microsoft) | Still pays cash, and just opened a credit card anyway | Nov 2, 2026 (FY27 Q1) |
| [Meta](/blog/companies/meta) | Sells stock to build a railroad only its own trains run on | Nov 2, 2026 (Q3) |
| [SpaceX](/blog/companies/spacex) | A rocket company turned landlord, and Google is one of its tenants | Nov 2, 2026 (Q3) |
| [Apple](/blog/companies/apple) | Doesn't lay track. It buys a ticket | Nov 3, 2026 (FY26 10-K) |

*Updated as each company reports. Q4 season starts late January.*

## Coming soon

The supporting cast, roughly in order:

- **Nvidia**, which sells the shovels to everyone above, and has been asked to help pay for some of the digging.
- **Oracle** and **CoreWeave**, which borrow to build data centers other companies have promised to fill.
- **TSMC**, which makes almost everyone's chips, including the ones designed to get away from Nvidia.
- **OpenAI** and **Anthropic**, the biggest customers on these pages, and big investments for some of them.
- **Broadcom** and **AMD**, which help the others build chips that aren't Nvidia's.

## Requests, agents included

Want a company covered that isn't here? Email [feedback@unbiased.ai](mailto:feedback@unbiased.ai): the company, one line on why, and your take if you have one. Agents are welcome to write in themselves; we'd just ask that your human is fine with you sending it (and, ideally, has read it first, which we know is a big ask). We read everything, and we'll say where a request came from when we use it.

*Disclosures: Louis previously worked at Google. Drafted with multiple agents, including models from companies discussed here. All of our relationships are on the [disclosures page](/blog/disclosures).*