DeepSeek Eyes In-House AI Chip as Data Centers Hit Power Costs

Two fresh bottlenecks are showing up in AI: custom chips in China and power costs in the US industrial belt.

🗄️ DeepSeek Starts Building Its Own AI Chip

Decoded: Reuters reported July 7 that DeepSeek is developing its own AI chip, aiming to reduce reliance on outside suppliers after its rise as one of China's most closely watched AI startups. The project is focused on inference, the stage that turns trained models into real products, and the immediate investor read is straightforward: every major AI lab now wants tighter control over silicon, not just model weights. The move also points to a shifting supplier map inside China, where access to top-end foreign chips remains politically fragile and expensive. (Reuters, July 7)

Why it matters: If DeepSeek moves even part of its inference stack in-house, it reinforces a trend investors cannot ignore: frontier AI economics are pushing labs to internalize more of the hardware layer. That pressures merchant chip vendors over time, especially in markets where export rules already distort supply. It also raises the odds that AI competition increasingly looks like a full-stack race, not a model-only race.

🏛️ AI Data Centers Push PJM Power Prices Into Factory Margins

Decoded: Reuters reported July 7 that power demand from AI data centers is driving sharp electricity-cost increases across the PJM region, which spans a large manufacturing corridor from the Mid-Atlantic into the Midwest. Belden Brick in Ohio said its monthly capacity charge jumped from $1,600 to $12,000, while PJM capacity prices climbed from $28.92 per megawatt-day in 2024 to $329.17 now, a 1,038% increase. Reuters calculations using Energy Department data showed industrial power prices were up 31% in Pennsylvania and 26% in Ohio as of December 2025 from a year earlier. (Reuters via U.S. News, July 7)

Why it matters: This is the cleanest sign yet that AI infrastructure costs are spilling past hyperscalers and into the real economy. When grid scarcity starts hitting factory margins, the political response usually follows, whether through higher charges on Big Tech, slower data center approvals, or forced investment in new generation. For investors, the bottleneck is no longer only GPUs, it is also who gets affordable power and who pays for the next wave of capacity.

Stay decoded. See you tomorrow.

— The Get AI Decoded Team