AI Industry News Brief — August 27, 2026

Coverage window: stories published between 21:07 UTC on August 25 and 21:07 UTC on August 26, 2026.

Nvidia beats estimates with $96.2B quarter as data center alone hits $89B

Nvidia reported fiscal Q2 2027 results after the close on August 26, posting revenue of $96.22 billion against an expected $92.17 billion. The data center segment accounted for the overwhelming majority at $89.02 billion, up 117% year over year, and adjusted earnings per share came in at $2.22 versus $2.10 expected. Guidance for the current quarter is $108.0 billion, plus or minus 2%, and that figure explicitly excludes any data center compute revenue from China. Hopper shipments to China represented less than 1% of data center revenue in the quarter.

Reference
CNBC — Nvidia earnings live updates
NVIDIA Newsroom — Q2 FY2027 financial results

OpenAI’s in-house Jalapeño chip tops Blackwell in its first published benchmarks

OpenAI released benchmark results for Jalapeño, its first custom-designed inference accelerator. By OpenAI’s numbers the chip delivers 1.5x to 1.9x more AI work per watt at peak throughput than the best commercially available systems, with 1.7x to 3.6x lower end-to-end latency. The runs used SemiAnalysis’s public InferenceX benchmark across three models — GPT-OSS 120B, DeepSeek R1 670B and Kimi K2.5 1T. Two caveats travel with the result: Jalapeño handles inference only and does not train models, and analysts note the fairer comparison is Nvidia’s newer Vera Rubin platform, which shares HBM4 memory. OpenAI says it will keep buying Nvidia silicon regardless.

Reference
CNBC — OpenAI’s Jalapeño AI chip and Nvidia margins
The Decoder — Jalapeño inference benchmarks

Meta settles with 29 states for $16.68B and rewrites the rules for teen accounts

Meta agreed to pay up to $16.68 billion to resolve claims from 29 states that it designed Facebook and Instagram to addict children, misled consumers about safety, and improperly collected data on users under 13. The deal was struck mid-trial in federal court in Oakland, California, before Judge Yvonne Gonzalez Rogers. Beyond the payment, Meta committed to substantial product changes for users under 18:

A mandatory night mode blocking access between midnight and 6 a.m.
A default two-hour daily time limit
Notifications blocked during school hours
Prompts after every 15 minutes of consecutive use
Like counts hidden on teen users’ posts
Cosmetic surgery and makeup filters disabled

Meta denied wrongdoing, and the payment is scheduled in annual installments over ten years.

Reference
CNBC — Meta settles social media addiction case
Al Jazeera — Settlement and platform changes
Forbes — $16.68 billion settlement

Amazon closes Mechanical Turk on September 30 after 21 years

Amazon confirmed it will shut down Mechanical Turk, its crowdsourced labor marketplace, on September 30, 2026. Launched in 2005, the platform matched businesses with workers for Human Intelligence Tasks — data labeling, audio and video transcription, survey completion — the kind of work Jeff Bezos famously described as “artificial artificial intelligence” because computers could not yet do it reliably. The service once drew more than 500,000 workers but had been in decline as dedicated data-labeling vendors pulled that workforce away, and Amazon stopped accepting new customers last month. Insurance and travel companies still using the platform are now looking for replacements. It is a fitting bookend: the human labor market that trained the machine-learning era is being wound down by what it helped build.

Reference
Quartz — Amazon shutting down Mechanical Turk
Tech Startups — 21 years of Mechanical Turk

Apple aims the M6 Mac mini and M5 Ultra Mac Studio squarely at local AI

Apple introduced a new Mac Studio with M5 Max and the all-new M5 Ultra, alongside an M6 Mac mini, both positioned for running large models and agents locally rather than in the cloud. The M5 Ultra can be configured with up to a 36-core CPU, an 80-core GPU and 512GB of unified memory, with memory bandwidth reaching 1.2TB per second. Apple’s own benchmarks put the M5 Ultra at up to 4.3x the peak AI compute of the M3 Ultra, and up to 9.8x faster LLM prompt processing and 8.2x faster text-to-image generation than an M1 Ultra system. The M6 Mac mini is Apple’s first 2nm chip, with a dual Neural Engine and neural accelerators in every GPU core. Most configurations ship September 22; Mac Studio starts at $2,499 with M5 Max and $5,499 with M5 Ultra.

Reference
Apple Newsroom — New Mac Studio with M5 Max and M5 Ultra
gHacks — M6 Mac mini and M5 Ultra Mac Studio

Emerald AI raises $150M to turn AI data centers into flexible grid assets

Emerald AI closed a $150 million Series A at a $1.05 billion valuation, co-led by Energize Capital and DCVC. The company builds software that lets power-hungry AI data centers modulate their draw in response to grid conditions, making them participants in the electricity system rather than just a load on it. Autonomous trucking company Gatik announced a $200 million round in the same news cycle. The pattern is worth noting: capital is moving toward the layer that connects AI to the physical economy — power, freight, infrastructure — rather than another generic software tier.

Reference
Tech Startups — Startup funding news, August 26, 2026

Today’s takeaway

On the same day Nvidia confirmed that appetite for AI infrastructure is still enormous — data center revenue up 117% — OpenAI announced that its own inference silicon outperforms Blackwell per watt, exposing a crack in the accelerator market. Add Apple pushing a 512GB M5 Ultra Mac Studio for running frontier models on a desk, and the center of gravity for inference is now contested by cloud, custom silicon and the desktop simultaneously. Meanwhile Meta closed its youth-safety exposure at $16.68 billion and Amazon retired the 21-year-old Mechanical Turk — the regulatory bill AI generated and the human labor market AI replaced, settled on the same day.