Tech news picked on 2026-07-09
Memory Has Shifted from Cost to Allocation, Shaking the Foundation of AI Startups
- AI data centers are consuming about 70% of global memory production, causing consumer DRAM prices to soar, and Apple has raised the prices of MacBooks and iPads by hundreds of dollars.
- Memory is no longer a cost that decreases every year but a resource allocated first to large customers.
- Smaller teams lower in the supply order must adopt designs that use less memory and pricing structures that withstand cost shocks to enter the market.
EU Revives Private Message Scanning, Challenging the Premise of End-to-End Encryption
- On July 7, the European Parliament passed an emergency procedure vote 331 to 304, setting a vote on July 9 to revive the temporary private message scanning regulation that had expired after being rejected in March.
- A separate permanent regulation is pushing for mandatory device-level scanning.
- If inspection can occur on devices before encryption, the promise of end-to-end encryption, that we cannot see the content, is only half-kept.
- For messaging products, it is more honest to clearly disclose under what conditions that promise is broken.
The Era of Cheap, Free Compute Is Over; Power Has Become the Ceiling for AI Businesses
- The US FERC ordered in June that AI data center grid connections must be processed within 90 days, with the condition that applicants either bring their own power or reduce demand.
- There is currently 2.2TW stuck in the connection queue, and more than 30 states have introduced regulatory bills.
- The real bottleneck for compute has shifted from chips to power.
- Let's calculate now the price and usage limits that can be sustained even if inference costs double.
Orange Planet and B-Startup Challenge Are Accepting Teams
- Orange Planet's regular recruitment and the B-Startup Challenge are accepting AI and digital healthcare teams.
- They offer residency space and connections to investments of 500 million KRW, but the main remaining question is what can be proven within 12 months.
- News about support programs comes back every year, but only a few teams have reached the next round.
- If your plan is to introduce AI in the public sector, keep in mind that the procurement phase tends to slow down the schedule.
The Speed at Which AI Startups Gain Revenue Has Changed
- This article says that AI startups are gaining revenue more quickly.
- However, gaining revenue quickly and the market growing are different things.
- The growth reported is a curve for a small number of startups that have already secured large funding and partnerships with big corporations. For teams that don't meet those conditions, industry growth rates are just someone else's numbers.
- What founders need to verify is one thing: whether their customers are moving payments forward.
Insights for AI Startups' Global Leap
- Sixsense and J&P Medi are accelerating their overseas expansion by being selected for Y Combinator's Summer 2026 batch and the Unicorn Bridge program.
- While the schedules for batches and support programs move quickly, the pace at which manufacturing sites and clinical trials are adopted do not move at the same rhythm.
- The variable that can advance the timetable is the speed of collecting data and proving efficiency on site rather than just prior information.
- Participating in the programs should be viewed not as the completion of overseas expansion but as the start of validation.