Pick your subsector. Each case below is a real episode from the feed: who said it, from what seat, and the second it was said. Nothing here came from a filing.
Sanjay Mehrotra · Chairman, President and CEO, Micron · August 21
The memory cycle is the semis debate of the year, and the man who runs the only US memory maker put numbers on it in a half-hour interview. Take-or-pay prepayments, the size of the shortfall, and when new wafers actually arrive. If you model Micron, Hynix, Samsung or anything that buys HBM, this was the quarter's most useful thirty minutes.
$250 billion of US memory manufacturing, backed by $22 billion of customer prepayments under five-year take-or-pay agreements.▶ 1:10
Supply is constrained, 2027 is expected to be tighter than 2026, and customers want 50% more memory than Micron can commit.▶ 4:02
Boise produces its first wafers in mid-2027 and ramps in 2028; the New York fab ramps in 2029 to 2030.▶ 3:07
Prepaid, take-or-pay, and short by half for two years. That is a pricing model, not a soundbite.
Pete Johnson · Field CTO of AI, MongoDB · September 1
The software debate is whether AI is a cost line or a revenue line. MongoDB's field CTO spends his days inside customers' AI bills, and on a ninety-minute technical podcast he gave the numbers: who ran out of budget, why context windows are the wrong lever, and what the embedding layer is worth. That is the unit economics of every "agentic" product you cover.
Uber burned through its 2026 token budget in 13 weeks, the cost of naively maxing out context windows.▶ 16:20
Voyage AI models, acquired by MongoDB, retrieve up to 14% better than other embedding models; embeddings are not commoditised.▶ 49:11
There is no standard stack for agents yet; most Fortune 500 companies start with employee-facing use cases and a human in the loop.▶ 1:10:49
The AI bill is being renegotiated in public, on technical podcasts. Your margin model should hear it.
Matthew Prince · CEO, Cloudflare · September 2
Cloudflare sits in front of roughly a fifth of the web. In one hour on Big Technology its CEO gave four numbers that reprice every ad-supported publisher, every marketplace and every business that lives on search, and announced what Cloudflare would do to Google's crawlers two weeks later.
Automated traffic passed human traffic online in May 2026, years earlier than expected.▶ 1:44
Crawl-to-click doubled in a year: Google 18 to 36, OpenAI 1,500 to 3,000, Anthropic 60,000 to 120,000.▶ 23:18
Cloudflare will default-block Google's AI crawlers from millions of ad- and subscription-supported sites from mid-September.▶ 30:04
If you own a publisher, a marketplace or anything that lives on search, that hour was your quarter.
Juan Hernandez · Head of Credit and Underwriting, Block · September 1
Nobody puts an underwriting model's performance in a filing. On a niche credit podcast the person who runs Block's underwriting did: how the Cash App score performs against bureau data, who its borrowers are, and the plan to distribute it outside Block. That is a new revenue line for one company and a threat to the bureau moat for three others.
The Cash App score, built only on first-party data, performs as well as or better than models using bureau data, with higher approvals and lower losses.▶ 7:45
Most users of Cash App Borrow have FICO scores below 580; a large part of that population is mispriced by traditional scoring.▶ 28:18
Traditional scores are slow to update and miss real-time deterioration, as the lag in card delinquencies after the pandemic showed.▶ 16:24
The bureau moat was questioned by a lender with the data to test it. That is a thesis, and it was free to hear.
Robert Kyncl · CEO, Warner Music Group · April 4
The value of a catalog in the AI era depends on the licensing structure the majors accept. At a business-school fireside, months before the deals were public, Warner's CEO said exactly what the company would and would not sign. If you model music, publishing or any rights owner, that is the term sheet.
Warner insists on revenue sharing for AI model training and outputs, rejecting fixed fees for training content.▶ 39:25
Financial participation in both generic and derivative AI outputs, with artist and songwriter opt-ins when name, likeness or voice is used.▶ 38:43
Warner has gained market share over the last twelve months.▶ 24:37
The structure of every AI music deal was stated in April by the man signing them. Read it in ten minutes.
Niraj Paliwal · SVP, Silicon IP Product Management, Synopsys · August 21
Synopsys sells the IP inside nearly every AI chip, so it sees the roadmaps before the systems exist. On an engineering trade podcast its SVP described what the hyperscalers now buy, how it is budgeted, and where the power goes. That is the demand curve for power, cooling, networking and the whole server chain.
Rack power went from 8 kW to 130 kW with the NVIDIA GB200 NVL72 and is projected to reach one megawatt within five years.▶ 16:56
Hyperscalers now buy AI capacity, not servers; the design and budget unit is the rack.▶ 4:14
The next bottleneck is interfaces and memory, the memory wall and the IO bandwidth wall, not compute.▶ 11:23
An eight-fold power step per rack is a content-per-rack model for a dozen names. It was said on a trade podcast nobody on your desk has time for.
Yago Tenorio · CTO, Verizon · August 10
Carrier opex is the telecom investment case, and the only person who can say what AI does to it is the person running the network. On a CIO interview series Verizon's CTO described the target, the results so far, and how fast the workforce was given the tools. That is the operating-leverage argument, from the source.
AI agents resolve network incidents in 90 seconds that previously took five to eight hours.▶ 10:47
Verizon targets a Level 4 autonomous network, the equivalent of a Waymo with human supervision.▶ 7:12
33,000 employees were given access to agentic coding tools in six weeks, for rapid in-house development.▶ 21:24
Ninety seconds versus eight hours is an opex line. It was said on a podcast with a few thousand listeners.
Aiman Ezzat · CEO, Capgemini · September 5
The services debate is whether AI deflates billable hours faster than it creates new work. Capgemini's CEO gave the delivery mix, the client engagement and the positioning in a short interview with Indian business television. Read it against every peer's offshore ratio and you have the sector's cost curve.
Capgemini's India operations, more than 230,000 people, are now 60% of the total workforce.▶ 0:19
The India team hosted over 1.8 million client visits last year.▶ 2:16
The strategy is a "shift to value": tangible economic benefit for clients, with AI agents embedded in operations and business impact before technology.▶ 2:25
A peer's delivery mix and positioning, in the CEO's words, five days before your next model update. Ten minutes.
14 days free. No card to start. Cancel anytime.
Work email, a one-time code, your names. Two minutes to your first brief.