Newsroom
Artificial Intelligence
Models, compute, business models, and the economics of AI.
EU opens 11 Android features to AI rivals. Reliability remains unsolved
On 16 July 2026, the European Commission ordered Google to give rival AI services equally effective access to 11 Android features, with most changes due in Android 18 by 1 August 2027. The order removes a system-access disadvantage, but a controlled mobile benchmark in which three frontier models completed 47.8% to 55.6% of GUI-only tasks shows why equal access cannot guarantee equal performance. https://digital-markets-act.ec.europa.eu/developer-portal/interoperability/alphabet-specification-proceedings-interoperability-ai-services_en; https://ec.europa.eu/competition/digital_markets_act/cases/202629/DMA_100220_2683.pdf; https://aclanthology.org/2026.acl-long.278.pdf
Did Meta's layoff process penalize protected leave? Public records cannot calculate the selection rates
Twenty-six Meta employees allege that AI-assisted performance measures converted protected absences or disability-related reductions in output into lower scores before Meta's May layoff selections. The mechanism is plausible, but every company-wide comparison group is incomplete, so the public record cannot show whether leave increased a worker's probability of selection.
Microsoft Cut an Xbox That Was Still Growing
On 6 July 2026 Microsoft cut 4,800 jobs and began shedding game studios it had built or bought years before the Activision deal, with Xbox chief Asha Sharma calling the unit's margins '3 to 10 times lower' than comparable platforms. But the profitable part of Xbox was still growing; only the console hardware was shrinking. That points to a capital-reallocation decision, not a gaming collapse, with the rising internal price of capital, set by the AI build-out, as the deciding variable.
One Word, 'Excess', Sank the AI Landlords. Meta's Own Contracts Point the Other Way
When Bloomberg reported that Meta was building a cloud to resell 'excess' AI computing power, CoreWeave fell about 14% and Nebius about 17% in a single session while Meta rose about 10%. But Meta's contracting behavior, locking in more outside capacity, not less, contradicts the glut story, which reframes the drop as a bet on thinner future margins rather than vanishing present demand. Which of those it is will not be settled until the next quarterly filings.
The OpenAI IPO Delay Did Not Trigger the AI Sell-Off. It Confirmed One Already Days Old.
Headlines blamed a June 25 report that OpenAI would push its listing to 2027 for a global rout in AI stocks. But lining the price moves up against the clock shows most of the damage landed on June 22-23, before the report crossed, when analysts blamed SpaceX's slump and a more hawkish Federal Reserve. The delay was the thermometer reading the fever, not the match that lit it.
onsemi's 23.7% drop turned a 27.6% premium for Synaptics into a 2.6% discount, in one day
On 25 June 2026 onsemi agreed to buy Synaptics for about $7bn, paying entirely in its own shares at a fixed rate of 1.350 onsemi shares per Synaptics share. The companies called it roughly a 19% premium, measured against a 10-day average of both stock prices. The next trading day onsemi's shares fell 23.7%. Because the share count was locked, the same deal was now worth only $122.38 per Synaptics share at the close, 2.6% below Synaptics' own price the day before the deal. It is a clean lesson in why a fixed-rate all-stock premium is a promise paid in the buyer's wobbling currency, not cash in hand.
Apple's memory-shortage price hikes, tested against the arithmetic: mostly real cost, with the one markup it cannot explain landing on the cheapest iPad
On 25 June 2026 Apple raised prices across Macs, iPads, HomePod and Vision Pro, blaming a shortage of memory chips. We checked Apple's own price list device by device. The dollar increases grow with how much memory each device carries, not with how expensive the device is, which is what you would expect if Apple were simply passing on a genuine rise in component costs. But our test is a forgiving one, the evidence leans heavily on a single pair of devices, and the one increase that clearly costs more than even the most expensive plausible memory bill lands on the $349 entry iPad, the device with the least memory to blame.
