The digital infrastructure carve-out is worth reading closely. Compute rented so that a buyer can deploy and run its own software sits outside the tax base, which is where the AWS and GCP bill lands [9]; the right to access someone else's prewritten software sits inside it [2]; and the exclusion for human effort delivered electronically is drafted so that it does not reach a SaaS access right [10]. What California has built is a wedge equal to the full district rate, 7.25% to about 10.75% [6], between renting a machine and renting a subscription [6]. That only bites where the untaxed alternative was already close on cost, which is a narrow set of buyers, and the wedge is likely to change invoices long before it changes architecture.
Vendors control the part that matters most: how a bundle gets billed. Under the true object test SaaStr describes, a bundle billed as one line item is taxable in full, while the same deal with implementation, support and managed services separately stated may be taxable only on the subscription [20]; the custom software exclusion likewise reaches modifications only to the extent they are separately stated [8]. The cheapest compliance response available to a vendor is therefore not a product change but a new invoice template, and a buyer asking how its top ten vendors plan to bill in January is asking the only question that sets its own number.
The collection mechanics invert where you would expect the burden to sit. Once one vendor's digital product sales to one purchaser pass $5M, the vendor is relieved of collecting and the purchaser self-assesses and remits [15], which is ten times the $500,000 remote seller threshold that now pulls software companies into CDTFA registration in the first place [14][5]. So the collection stack a vendor builds serves its long tail while its largest accounts get handed a use tax function, and the same account describes the $5M test both as a calendar year figure and, per CDTFA's July workshop, as measured over 12 months with the shift landing on the transaction that crosses the line [15][16].
The size of the tax base is the other half of the story. A projection of roughly $2B a year in combined state and local revenue [4] implies a taxable California software base between about $18.6bn at the top district rate and $27.6bn at the state rate alone [1], somewhere in the twenties of billions of annual subscription spend that sat outside the base across the three decades California taxed the disc and not the browser [22]. None of it is recoverable, since there is no input credit and no refund mechanism [17], so it stacks multiplicatively with vendor uplift: a 7% renewal on top of 9% tax is 16.6% [19][3]. A buyer holding its software budget flat buys roughly 8% less software [4].
All of this rests on one publisher's reading of the statute plus one CDTFA workshop, and that published text works through nexus, sourcing, bundling and the $5M flip without addressing resale or exemption certificates [7], which is where channel and reseller treatment will actually be settled. SaaStr's claim that most B2B founders have not yet done the vendor-side work is the one assertion here carrying no number [21]. What would make the 8-10% baseline [7] too high is CDTFA guidance that reads a typical managed subscription as predominantly service under the true object test. Signed June 29, 2026, effective January 1, 2027 [1][2]: 186 days [2] to reprice a quote, and $160-200K of new expense on a $2M tool budget [18].
