The AI Boom Is Now Sitting in Your Bill of Materials

For a while, the cost of the AI boom lived in someone else's budget. It was a line in an infrastructure spreadsheet, a data center problem, a footnote in a sustainability report. That's over. The bill is showing up in the price of phones, in the parts you buy, and in the questions your customers now ask about how you power your product.

Let me catch you up on where the physical costs landed and what they mean for the choices you make this quarter.

Memory got expensive, and it's not slowing down

Samsung is pushing for another memory price hike of up to 20% next quarter, and that comes after three straight quarters of major increases: roughly 90% in Q1 2026 and 50 to 60% in Q2. Data centers are buying up nearly all the supply. For the first time in years, Samsung, SK Hynix, and Micron hold real pricing power, and they're using it.

The reason is simple. Chip makers are shifting production toward high-bandwidth memory for AI accelerators because it pays more per wafer than the standard RAM in your phones and laptops. That leaves less capacity for everyday electronics, so prices climb. If your product ships with a chip in it, you are now competing with hyperscalers for the same wafers, and you do not have their leverage.

The phone market already broke

India is the clearest proof this is real. Smartphone shipments there fell 10% year over year in the April to June quarter, the steepest June-quarter drop in six years. China fell just 2% over the same stretch. India got hit harder because about 60% of its market sits under $210, exactly where higher memory costs bite hardest. The sub-$150 segment collapsed 45%.

Prices there rose between 4% and 68% depending on the model. Buyers are stretching replacement cycles from about 3.5 years to four, or moving to secondhand. Kiranjeet Kaur at IDC expects the shortage and elevated prices to last until at least the end of 2027. This is a preview of what happens to any price-sensitive product when a core component gets scarce.

When margins tighten, brands retreat

Watch what OnePlus did. It said it would stop launching new products in Europe and North America while keeping its India business, after China grew to 74% of its shipments. When each unit earns less, you cannot afford to serve markets where you barely turn a profit. You pull back to where the math still works.

Samsung, meanwhile, was the only major brand to post growth in India last quarter, up 2%. Premium buyers, cushioned by financing, barely flinched. The lesson for your roadmap is uncomfortable but clear: rising component costs push the whole industry upmarket, and low-end products become the first thing you cut.

The energy cost has a face now

The other half of the bill is power, and it's getting loud. Microsoft's carbon footprint grew 25% last year, moving it further from its 2030 carbon-negative goal. Its electricity use jumped 24%. Amazon rose 16%, Google 18%. Data centers are the driver. Microsoft even signed a deal with Chevron for a 2.67 gigawatt gas plant in Texas.

This is no longer an internal report nobody reads. When Microsoft's sustainability chief spoke at a Seattle climate event, protesters drowned her out with chants: "Microsoft, you can't hide. We can see your dirty side." How you power your AI is becoming a brand question, asked in public, by people who no longer accept a footnote as an answer.

The deep cut

The companies pulling ahead are the ones treating supply as a design decision, not a purchasing one. India just committed about $6.5 billion to move from "assemble more" toward local component sourcing, because Tarun Pathak at Counterpoint says brands are now trying to "save every cent" on parts. Google locked in a 1.8 gigawatt solar and battery project that can deploy in three years, while xAI runs nearly 60 gas turbines without federal clean air permits near predominantly Black neighborhoods.

Both are the same move: decide your inputs early, on purpose. If your product depends on memory you buy at spot prices, or compute you run on power you never chose, you are exposed on cost and on reputation at the same time. Bring your bill of materials and your energy sourcing into the same review. They are now the same conversation.

Three questions for your team

1. If memory prices stay elevated through 2027, which products in our lineup stop being profitable, and are we willing to cut or reprice them the way OnePlus cut whole regions?

2. Are we buying components and compute at spot prices, or have we locked in supply and power the way Google locked in solar years ahead? Where are we most exposed?

3. If a customer or reporter asked how our AI features are powered and what they cost the grid, could we give a straight answer, or would we get drowned out?