AI just broke the supply chain. Apple posted $111.2 billion in Q2 2026 revenue Thursday after the close, its best March quarter ever, with EPS of $2.01 against a $1.95 estimate and iPhone revenue up 22% year-over-year. Mac revenue hit $8.4 billion against an $8.02 billion estimate, but Apple disclosed Mac mini and Mac Studio supply was constrained because AI and agentic-tool demand outran the company’s own forecast. Tim Cook—in his first earnings call since announcing the September 1, 2026 handoff to incoming CEO John Ternus—warned of significantly higher memory costs in the June quarter as AI demand has roughly quadrupled iPhone RAM costs, in what the trade press is now calling a “RAMageddon.” Cook framed Apple’s strategy as weaving AI into hardware rather than building a standalone AI app, which makes the memory squeeze the gating variable on every Apple Silicon device shipped this year. Record demand met record component inflation in the same press release.
While Apple’s bottleneck sits at the chip and memory layer, the same week reset the model layer. Mistral shipped Medium 3.5 on April 29, a 128B dense flagship with a 256k context window, open weights under a modified MIT license, and self-hostable on 4 GPUs. API pricing lands at $1.50 per million input tokens and $7.50 per million output, with the model scoring 77.6% on SWE-Bench Verified and 91.4 on the τ³-Telecom agentic benchmark. It replaces both Mistral Medium 3.1 and Magistral in Le Chat as the company’s first “merged” flagship—instruction-following, reasoning, and coding consolidated into a single set of weights—and powers the new Work mode and Vibe remote agents tier. Distribution runs through the Mistral API, NVIDIA NIM endpoints, and Hugging Face. The same week Apple confirmed the Mac is now an AI workstation, the open-weights frontier just put a 128B reasoning-and-coding flagship inside a 4-GPU footprint.
The cloud layer rearranged itself in the same window. Amazon used its What’s Next with AWS event on April 28-29 to launch Bedrock Managed Agents powered by OpenAI, alongside GPT-5.5 and GPT-5.4 on Bedrock and Codex for in-AWS coding assistance, all in limited preview. The managed-agent service is built on the OpenAI harness, engineered for OpenAI frontier models with what AWS describes as faster execution and reliable steering of long-running tasks, and routed through existing Bedrock APIs for security, governance, and cost controls. The launch materializes the OpenAI–AWS partnership that Microsoft cleared with its exclusivity rewrite on April 27, turning a contractual change into a shipped product the next day. Sitting one layer below the model and far below the device, we worked through xAI’s new Grok TTS and STT endpoints in our Grok Speech API tutorial—$4.20 per million characters versus $100 on ElevenLabs Multilingual v2—because cheap inference is the other half of how the device-to-cloud stack gets reshaped. Three layers, one week, three different repricings.
The Pulse
Legora hits $5.6B as Nvidia and Atlassian back $50M Series D extension to $600M
Legal AI is officially Nvidia’s newest portfolio vertical — Harvey doesn’t get the field anymore.
Accenture leads Netomi’s $110M Series C and signs a global agentic-AI alliance
When the systems integrator becomes the channel, customer-service vendors stop being software and start being labor.
Huawei expects AI chip revenue to surge 60% to $12B as Ascend 950PR ramps
China’s domestic-AI thesis isn’t a hedge anymore — it’s the order book.
Meta’s business AI now handles 10 million conversations a week — up from 1 million
Free today, monetized tomorrow — WhatsApp just became Meta’s next ad platform with extra steps.
Featherless AI raises $20M Series A from AMD and Airbus to host 30,000+ open models
AMD funding the open-model gateway is the quietest shot at the closed-AI stack this week.
Latest from PulseMark
 |
Grok Speech API Tutorial: TTS and STT at $4.20/1M Chars
Five fixed voices, no waitlist, and a 24x price gap to ElevenLabs Multilingual v2 — with working Python, the pricing math, and a candid framework for when the switch makes sense (and when the missing voice cloning kills the deal). |
That’s the signal.
— The PulseMark Team
Get the Daily Pulse
Sharp analysis on what's actually moving in AI. No hype, no filler, no weekly digest.