Musk capped drone links, officials say; Anthropic draft puts computing near $9B

Marcus here.

Today's picks turn on who controls the services people pay for.

Elon Musk said on Oct. 14 that the U.S. military and the president decide Starshield's limits, as Ukrainian drone operators lost connections in flight and units canceled missions without any warning.

Anthropic racked up nearly $9 billion in computing costs in the first half of 2026, a draft IPO prospectus shows, and the bigger bills are still coming.

Yelp's Check for Me calls restaurants about wait times for diners. Its own voice product for businesses, sold since 2025, could end up answering the call.

Stay curious,

Marcus Schuler

Elon Musk ordered the Starshield slowdown that disrupted Ukrainian drone connections from Oct. 14, U.S. officials and people familiar with the matter said.

Starshield runs on Starlink, and the U.S. government pays for it and controls access. On Oct. 14, Ukraine's 7th Rapid Reaction Corps reported 10 to 20 second outages at 110 to 120 km/h.

One person familiar said Musk discussed the cap with Trump beforehand; some officials said he acted alone. Musk had written that limits are "decided by the US military" and the president. Trump said: "it wasn't me, and I don't think it was Elon."

Why This Matters:

• Ukrainian interceptors need speed to catch Russia's jet-powered Shahed variants; satellite interruptions hamper the air defenses that rely on them.
• Nine senators asked the Pentagon's inspector general to investigate restrictions on a service the U.S. government pays for.

Anthropic incurred nearly $9 billion in computing capacity and infrastructure costs in the first half of 2026, a draft IPO prospectus shows.

First-half revenue reached $16.3 billion. Non-cash charges from changes in the value of financing instruments accounted for $72 billion of its $80.7 billion net loss. Future infrastructure commitments exceed $600 billion, and most of its $450 billion in cloud deals is payable regardless of usage.

Cloudflare’s Oct. 19 post explains how it cut average AI code-review costs to $0.77 per review.

The baseline was $2.43 across its two costliest weeks before the changes. “Much of the drop in critical findings is the reviewer no longer calling minor things critical,” Cloudflare writes.

• eMudhra estimates about 30,000 certificates need migration after eight weakly generated signing keys were compromised and reported Oct. 17.
• A fired OpenAI engineer was ordered held without bail by Judge Alex Tse in San Francisco on Oct. 19, citing a risk of non-appearance.
• SpaceX generated $7.5 billion of its $14.6 billion third-quarter 2026 revenue from its AI segment.
• Google released TorchTPU on Oct. 20, letting PyTorch applications run on its TPU chips through a configuration change.
• Otter.ai received $320 million in financing without equity dilution on Oct. 20 from General Catalyst’s Customer Value Fund.
• Meta put 39% of its total October ad spending behind Muse, up from 23% in September.

Run a blind read on a board recommendation.

A polished board memo can inherit the sponsor's authority. Strip identity cues before judging whether the recommendation earns approval.

Your raw input:

The prompt:

Why this works:

Neutral labels reduce status and brand bias before the model scores the case. The evidence test separates persuasive writing from support, while the rejection test makes the model look for disconfirming facts.

What to use:

Use a current reasoning model in ChatGPT or Claude. A smaller model is adequate when the memo is short and the criteria are explicit.

Profound sells software that shows marketing teams how brands appear in AI-generated answers, then gives them agents to act on the findings. Its bet is that visibility inside ChatGPT, Gemini and similar systems becomes a budget line before larger marketing suites absorb the category.

Founders:

James Cadwallader and Dylan Babbs founded Profound in New York in August 2024 after meeting at South Park Commons. Cadwallader previously ran a media company; Babbs was a design engineer on Uber's maps team.

Product:

The platform samples answers from ChatGPT, Perplexity, Google AI products, Gemini, Copilot, Claude and Grok, tracks brand mentions and citations, and links those findings to marketing workflows. Enterprise marketing teams and agencies pay for it. Profound said more than 1,000 enterprise brands used the platform as of September 15, 2026, including Walmart, Royal Bank of Canada, Zoom, ServiceNow and Figma.

Financing:

Profound had raised more than $300 million across five rounds as of September 2026. Its $180 million Series D on September 15, 2026 was co-led by Sequoia Capital and Kleiner Perkins at a $1.8 billion valuation.

The risk:

Semrush, Adobe and other marketing-software incumbents can add similar monitoring to broader suites. Profound also depends on answer engines leaving enough output and access available for outsiders to measure reliably.

Credit: Midjourney

Prompt: девушка в горах (a girl in the mountains)

Yelp's Check for Me feature calls restaurants for wait times and texts diners the answer.

Our take: Yelp has offered restaurants and service providers a voice product since 2025. It now supplies the caller too. A diner enters a party size and phone number so software can telephone a restaurant that may have software answering. This is an elaborate way to ask how long dinner will take. The conversation might break down before the answer comes back.

Yelp also launched Smart Lead Reply at $30 a month for business owners. Booking through OpenTable already happens inside the app. Yet checking the wait gets a telephone delegation. Apparently the missing ingredient in dinner was a conference call for software.

Marcus Schuler

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai