OpenAI, Meta, SpaceXAI Compete for More Cost-Efficient AI Models
THE SO WHAT
When three major labs emphasize lower cost per token over raw capability in the same week, AI economics are shifting from ‘can we do it’ to ‘what’s the unit cost’. If you’re budgeting AI into core workflows, this is the moment to re-run your cost models—assume rapid repricing and negotiate for variable usage, not fixed lock-ins.
READ THE SOURCE
MORE FROM THE WIRE
Applied AISources: Apple is already taping out M7 with major NPU upgrades, and plans M7 Ultra with 1.5TB RAM, and M8, for 2028; new Pencils are coming with a new iPad Pro
Apple planning M7/M7 Ultra with major NPU upgrades and up to 1.5 TB RAM is a clear tell—on-device AI is being treated as a long-horizon, system-level bet, not a feature. If you build Mac software, assume a world where local models and massive context windows are standard and architect for offline, privacy-sensitive workloads now.
Applied AIThe AI race has quietly stopped being about who has the biggest model
Procurement is shifting from “who has the biggest benchmark” to a portfolio of models tuned for task, cost, and control. If you’re still designing around a single frontier model, you’re behind the buyer—start planning for multi-model routing, cost-aware selection, and domain-specific evals.
Applied AIMost Americans now say the public should own half of the big AI companies
When 69% of surveyed Americans back forcing AI firms to hand half their equity to a sovereign wealth fund, you’re no longer just managing “regulatory risk”—you’re in the realm of ownership politics. Boards and founders building around AI need to scenario-plan for capital structures and control rights that look very different from standard tech playbooks.
Applied AIChinese developers are climbing the firewall to pay more for GPT-5.6, and doing the maths on why
Mainland developers are paying VPN and proxy premiums for GPT-5.6 because token efficiency makes the total cost pencil out—performance per dollar is beating access friction. If you’re building models or tools for this market, you’re competing not just on price but on effective throughput and latency under gray-market conditions.