> daily_signal(2026_07_11)
Apple just signed a $30 billion-plus chip deal with Broadcom to build the AI server silicon behind Apple Intelligence — in the US.
PickBits Daily Signal · Saturday, July 11, 2026
1. Apple just signed a $30 billion-plus chip deal with Broadcom to build the AI server silicon behind Apple Intelligence — in the US.
On July 9, 2026, techinasia reported that Apple has signed a supply agreement worth more than $30 billion with Broadcom for custom accelerator chips — the silicon that runs Apple Intelligence on Apple's own Private Cloud Compute servers. The deal deepens a Broadcom partnership the two companies had already extended through 2031, and it moves Apple toward designing the processors behind its AI features in-house rather than buying general-purpose parts off the shelf. The strategic tell is where the work lands: the chips are slated for US fabrication, aligning the program with Apple's standing US-investment commitments. For everyone downstream, this is another of the largest technology buyers writing a multibillion-dollar check that says purpose-built AI accelerators — designed for one company's workloads — are now the piece of the stack worth owning outright.
Key fact: If you buy or budget cloud/AI compute, open your provider's roadmap and audit where your workloads are pinned: as Apple, Meta, Google, and Amazon each bring custom accelerators online, fleets are going heterogeneous (custom chips + Nvidia + AMD) — map which of your pipelines are locked to a single vendor's toolchain now, and prioritize portable runtimes/abstractions where a cheaper accelerator could pick up the load later.
techinasia.com · broadcom.com · apple.com · primary source
2. Meta just cut its Muse Spark 1.1 coding-model API below OpenAI and Anthropic — starting a price war over the AI tools developers actually build with.
On July 9, 2026, the-decoder reported that Meta released Muse Spark 1.1 with API pricing that undercuts OpenAI and Anthropic on coding workloads — the tokens developers spend the most on — and techinasia confirmed the 1.1 rollout on 07-10. The move is Meta's hard push into the AI-coding-assistant market, the segment Anthropic's Claude and OpenAI's models have led, and it turns model choice into a pricing fight rather than a capability standoff. For developers and the platform teams that pay their bills, cheaper frontier-grade coding tokens are an immediate win; the risk is the one every price war carries downstream — a scramble to switch providers for cost, then lock-in, then a correction once someone has to make the unit economics work on inference that is genuinely expensive to serve. The bigger signal is that the AI-coding tier has matured fast enough that the biggest labs now compete on price per token, not just benchmark scores.
Key fact: If you run engineering or platform spend, open your current AI-coding provider's pricing page and Meta's Muse Spark 1.1 API rates side by side and compare cost-per-task on your real workloads — but before you migrate, verify rate-limit terms, data-retention/training policies, and whether the introductory price is durable (DeepSeek's peak-hour surcharge is the cautionary tale); pilot behind an abstraction layer so a price reversal doesn't strand you.
the-decoder.com · theverge.com · techinasia.com · primary source
3. Patreon just switched on Cloudflare's crawler blocking to stop AI companies from scraping creators' paid work to train models — without consent or pay.
On July 9, 2026, 404 Media reported that Patreon has switched on Cloudflare's crawler-blocking to stop AI companies' bots from scraping the members-only posts creators sell — art, writing, music, video — for use as model training data. It runs on Cloudflare's July 6 upgrade from an all-or-nothing AI-bot toggle to granular controls that sort crawlers by what they're doing — indexing for search, collecting training data, or acting as an autonomous agent — so a site can welcome search bots while turning away training scrapers. The practical stakes for the creator economy are direct: a Patreon post sits behind a paywall creators charge for, and an AI crawler that copies it feeds a model the creator was never paid for and never agreed to. Patreon setting the block platform-wide means every creator page behind it is covered without anyone touching a setting, and it lands as a growing list of platforms flip their default for AI bots from open to blocked — narrowing the free training data that model builders had treated as there for the taking.
Key fact: If you're a creator or run a site with original work, log into your Cloudflare dashboard (or ask your host) and review the AI crawler controls now: enable the training-scraper block while keeping search crawlers allowed, check your robots.txt and any AI-specific opt-out signals, and confirm the setting actually took effect — Patreon just did this at the platform level, but on your own domain it's a checkbox you have to flip.
404media.co · the-decoder.com · cloudflare.com · primary source
4. Anthropic just launched its own AI drug-discovery programs — aimed at the diseases Big Pharma has decided are too unprofitable to chase.
In a rollout reported across late June and early July 2026, Anthropic said it will develop its own drugs — standing up in-house AI drug-discovery programs and a bioscience product, Claude Science, to run them. The through-line the company put front and center: target the diseases the pharmaceutical industry has judged too unprofitable to pursue — the neglected and rare conditions where the economics never closed, not the blockbuster markets. It is a notable turn for a frontier lab: instead of only selling models to biotechs, Anthropic is pointing its own compute and models at drug discovery directly, using AI to compress the early, expensive stages — target identification, molecule design, and candidate triage — where cost and time have historically kept unprofitable diseases un-researched. The promise is the genuinely constructive case for the whole buildout: if AI can lower the cost of discovery enough, the calculus that leaves rare-disease and neglected-tropical-disease patients without options can change. The caution is equally real — announced programs are not approved drugs, and the distance from an AI-designed candidate to a validated, regulator-cleared therapy is measured in years and clinical trials, not model runs.
Key fact: If you work in biotech, drug discovery, or health policy, open the STAT and Endpoints coverage and read Anthropic's own program description, then compare its stated targets against your own pipeline or the neglected-disease lists (WHO NTDs, rare-disease registries): identify one indication where AI-accelerated target ID or candidate generation could change the cost math, and treat announced programs as a lead to track through preclinical milestones, not a cleared therapy.
the-decoder.com · statnews.com · theverge.com · endpoints.news · primary source