I’m one of the founders of San Francisco Compute. We sell GPUs. I’ve been thinking about commodities a lot, since they pop up in AI a lot. AI is unique: it’s software with expenses.

Most software people aren’t used to dealing with significant capex, COGS, supply, depreciation, inventory risk, commoditization, and so on. Folks come into the game with decades of advice on how to make high-margin, highly differentiated, product focused software companies, but little information on how to win in games with normal business.

The following piece argues that there’s an objectively correct move to be made at every step, and lays out how to make them.

On Commoditization

By default, commoditization king-makes the player with the cheapest cost of capital. Eventually, all the competitors in the market conform to the cheapest strategy to create widgets and buy from the cheapest vendors. At that point, the only way to drive prices down further is to compete to get a cheaper loan than the next guy. And the folks that get the cheap loan are the biggest and oldest, since they’ll have the best credit history.

If capital is in low supply (no one wants to lend or the commodity requires large amounts of capital), then you end up with only a few, big market participants. All the new & small players didn’t have the credit to get a cheap loan, so died when the big player undercut their prices.

Because there’s only a few large participants, the vendors who sell to the commodity market participants suddenly start to lose. When you can only sell to one of four companies, each of them has strong negotiating power, and form a monopsony.

The alternative strategy is to escape the commodity market all-together by skating up the value chain by building a product that uses the commodity as an input. It’s the “law of conservation of attractive profits”: when one market commoditizes, profit accrues in the adjacent points in the value chain. In other words, if you’re buying a product, you love when the market commoditizes, because the inputs of your business get cheaper and your costs go down.

On Pricing

Be wary of the axis of competition. If customers only care about price, your only option is to reduce your prices. However, once the prices of you and your competitors are too close to matter, the axis of competition shifts away from price. Then your customers start caring about all the other parts of your product.

Many founders incorrectly focus on making their product really good too early. In traditional software, you’re rarely really competing on price — it just doesn’t cost much to run a web server. Thus, you’re often competing to have a better user experience or some particular feature. But AI has actual expenses, and thus prices can be in the thousands, if not millions. And when it’s that expensive, you must compete on price first, not on product quality. If the cost to produce an iPhone was $100k, then Apple should manically focus on how to bring down COGS, not on making an incredible luxury product. They only get to focus on design when the costs are in the range of a customer’s wallet.

On Assets

Every time your competitor makes an asset purchase, you should celebrate. The bigger the purchase, the more you should celebrate.

Each large purchase by your competitor locks them into the strategy implied by the investment. Software people are used to competitors that never buy any assets, where a strong-willed founder can simply shift the company in a completely different direction at will, and fire anyone who disagrees. But if you buy assets, if you take loans, there’s no amount of strong will that can cause the company to shift. Software companies really only die when people give up. But companies with assets die because they go bankrupt — so they’re actually locked into a path.

Anytime your competitor is locked into a path, then there’s a bunch of things they can’t do, and if you can make money doing those things, then your competitor will flee away from you.