Weekly digest
Weekly · 2026-10-17 · 94 fetched · 41 episodes ranked · 42 with transcripts · 176 claims
94 episodes fetched; 41 with transcripts are ranked and 52 had no usable transcript and are not ranked, so audio-only shows are underrepresented. Same-transcript duplicates merged: 1.
Agents: claims.jsonl
Amazon CEO gives checkable operator numbers: AWS AI run-rate over $25B, 2026 capacity sold out and 2027 largely sold, server payback about 3 years; firsthand and material to AMZN and the AI supply chain.
Andy Jassy (Amazon CEO)
25:06AWS capacity for 2026 is sold out, 2027 is largely sold out and a good chunk of 2028 is sold out, apart from a portion held back for on-demand use.
AMZN novelty 2/3 · verifiable 2/3 Andy Jassy (Amazon CEO)
13:09AWS's AI business has passed $25 billion in annual run rate within the first three years of the generative AI wave.
AMZN novelty 2/3 · verifiable 2/3 Andy Jassy (Amazon CEO)
15:05Amazon buys servers and networking gear only a few months ahead of demand and expects them to break even in about three years with 5 to 6 year useful lives, while data centers need about two years of lead time, creating free cash flow headwinds during the build.
AMZN novelty 2/3 · verifiable 2/3 Andy Jassy (Amazon CEO)
32:11Amazon's Trainium AI chip is about 30 to 40 percent more price-performant than comparable processors, per Jassy.
Andy Jassy (Amazon CEO)
24:53Jassy does not expect the shortage of data-center capacity, power and memory to end in 2027, and says supply constraints will show up in different components for several years.
AMZN novelty 1/3 · verifiable 1/3
Intel's CTO says global tool stock allows only ~2x leading-edge wafers, and Intel would invest more; firsthand exec detail on High-NA/backside power vs TSMC and Terafab. Few hard figures.
guest (Intel CTO Pushkar Ranade)
50:40Intel's CTO says that with the world's existing chipmaking tools leading-edge wafer output could roughly double but not rise tenfold.
INTC novelty 3/3 · verifiable 1/3 guest (Intel CTO Pushkar Ranade)
1:08:47Intel's CTO says that, knowing today's demand, Intel would have invested more a year ago and will now invest more.
INTC novelty 2/3 · verifiable 2/3 guest (Chris Auth, Intel process development)
15:04Intel's process lead says backside power lets 14A use High-NA EUV in a single pass, while TSMC without backside power would still need double-patterned EUV.
guest (Intel CTO Pushkar Ranade)
1:01:52Intel says it is working with Elon Musk's Terafab to help accelerate bring-up of Musk's own fab.
INTCTerafab novelty 2/3 · verifiable 2/3 guest (Intel CTO Pushkar Ranade)
56:08Intel's CTO says Intel is already within a few percent of TSMC on performance, power or density and that the remaining gap is wafer scale.
INTCTSM novelty 1/3 · verifiable 2/3
CEO-level firsthand data: Gemini 4 Argon did about 25 trillion tokens in 72 hours (3x Gemini 3.0), Google capacity-constrained, committed-revenue contracts behind capex, Waymo above 500k paid rides a week.
In the first 72 hours after launch Gemini 4 Argon processed almost 25 trillion tokens, about 3x what Gemini 3.0 did, and Google is capacity constrained on the launch.
GOOGL novelty 3/3 · verifiable 2/3 Alphabet went free-cash-flow negative this summer for the first time since its IPO (host's framing; Pichai clarified the term but did not dispute it).
GOOGL novelty 2/3 · verifiable 3/3 Pichai says Google is signing durable long-term contracts with committed revenue, alongside demand signals from its own use, as part of the case for its heavy compute investment.
GOOGL novelty 2/3 · verifiable 2/3 Waymo has just crossed half a million paid rides per week.
GOOGL novelty 2/3 · verifiable 2/3 Pichai expects practical commercial quantum applications within about five years, with Google Cloud negotiating quantum deployments at some scale by then.
GOOGL novelty 1/3 · verifiable 1/3
SemiAnalysis raises its 2027 data-center forecast above 43 GW, maps 400+ moratoriums touching only ~10% of capacity, and details Oracle New Mexico, Childress, and 2028 oversupply risk.
SemiAnalysis is raising its 2027 data-center build forecast from 43 GW (article title says 50 GW in a single year), against critics who say only about 25 GW will be built.
novelty 3/3 · verifiable 2/3
guest analyst (SemiAnalysis)
7:19SemiAnalysis mapped over 400 data-center moratoriums against sites it tracks and found only about 10% of capacity was affected.
novelty 3/3 · verifiable 2/3
guest analyst (SemiAnalysis)
13:47Oracle's New Mexico campus has over a gigawatt of IT capacity under construction, but SemiAnalysis assumes a slow ramp because of power delivery problems, so little counts in 2027.
ORCL novelty 3/3 · verifiable 2/3 guest analyst (SemiAnalysis)
27:01SemiAnalysis says chips are already in oversupply relative to data-center megawatts today, with 2028 a risk of flipping to data-center oversupply.
novelty 3/3 · verifiable 1/3
guest analyst (SemiAnalysis)
19:03Signed data-center leasing exceeds $1T, and crypto-miner backlogs at Applied Digital and Core Scientific are now in the tens of billions of dollars.
CEO firsthand with dated numbers: Boom's first natural gas turbine ships for test this week, 2 GW/yr factory early 2027, targets $5-10B revenue by 2030, vertically integrated blade foundry.
Boom is building a 2 gigawatt-per-year turbine factory operating early 2027, producing about a quarter gigawatt next year, a gigawatt in 2028 and 2 gigawatts in 2029.
Boom Supersonic novelty 3/3 · verifiable 2/3
Boom says its baseline plan makes it a $5 to $10 billion revenue business by 2030 while filling less than one-tenth of the turbine shortfall.
Boom Supersonic novelty 3/3 · verifiable 1/3
Boom has built its first natural gas turbine core for AI data centers, which ships to its test site the next day with a goal of running by Thanksgiving.
Boom Supersonic novelty 2/3 · verifiable 2/3
Conventional gas turbines (GE named) lose about 20% of capacity on hot days when using water and about 30% waterless, which Scholl says his design avoids.
GEV novelty 2/3 · verifiable 1/3 Boom is building an all-digital single-crystal blade foundry aiming for blade design to casting in 24 hours, because only a couple of firms can make such blades and they are supply constrained.
Boom Supersonic novelty 2/3 · verifiable 1/3
Analyst note built on ASML patent filings (unannounced die-to-wafer bonder), HBM hybrid bonding not before 2029, China 2030 roadmap, reticle limits; specific and checkable, but unlabeled speakers.
Per patent filings, ASML is building a not-yet-announced die-to-wafer hybrid bonding tool.
SemiAnalysis analyst (Howie or Sherry Zhang)
11:23SemiAnalysis analysts expect HBM to adopt hybrid bonding no earlier than 2029, because it costs more than thermal compression bonding and customers want fewer layers.
novelty 2/3 · verifiable 1/3
SemiAnalysis analyst (Howie or Sherry Zhang)
24:34Chinese roadmaps target a fully hybrid-bonded 'logic folding' data-center accelerator by 2030, with up to three layers.
Huawei novelty 2/3 · verifiable 1/3
SemiAnalysis analyst (Howie or Sherry Zhang)
25:53High-NA EUV's half-size reticle limits die size to about 400 mm2 versus about 850 mm2 today; a 6x12 mask would restore the full size.
ASML novelty 1/3 · verifiable 3/3 SemiAnalysis analyst (Howie or Sherry Zhang)
27:51ASML management said on its recent earnings call that pricing is value based, so higher tool throughput raises prices.
ASML novelty 1/3 · verifiable 3/3
Ramp's lead economist brings proprietary spend data: ~1% of firms drive ~80% of OpenAI/Anthropic revenue, top-1% spend is slowing, and the restricted Fable model peaked at 5-10% of business spend.
Ara Kharazian (Ramp lead economist)
11:30About 1% of companies account for roughly 80% of OpenAI and Anthropic spending on Ramp's platform, concentrated in software, media, finance and technical services.
OpenAIAnthropic novelty 3/3 · verifiable 2/3
Ara Kharazian (Ramp lead economist)
12:57AI spend by Ramp's top 1% of customers has slowed over the last couple of months even as their usage rises, because they spread across models and prices keep falling.
OpenAIAnthropic novelty 3/3 · verifiable 2/3
Ara Kharazian (Ramp lead economist)
29:30Anthropic's restricted Fable model (as named in the episode) topped out at about 5 to 10% of business spend because businesses would not accept the data-collection conditions and chose a slightly cheaper OpenAI model.
Anthropic novelty 3/3 · verifiable 1/3
Ara Kharazian (Ramp lead economist)
20:24Ramp finds firms using AI at high intensity (at least $50 per employee per month) grew headcount 10% over the following two years while average AI spenders showed no difference from controls.
novelty 3/3 · verifiable 2/3
host (Derek Thompson)
2:45One outlet reported OpenAI's revenue pace at about $70 billion a year and the Financial Times a week later put it nearer $50 billion, a roughly 30% gap that moved markets.
OpenAI novelty 1/3 · verifiable 2/3
Two large-company CEOs give specific numbers: open-weight 40x cheaper, 75-90% of tokens, half of IBM code from agents; the open-weight share forecast is the new angle.
Arvind Krishna forecasts 75% to 90% of all token usage will go to open-weight models, with the Chinese-Western open-weight gap at most 6 to 12 months.
IBM novelty 2/3 · verifiable 1/3 Krishna says a Chinese open-weight model (transcribed 'Kimikay3', ~800B parameters) on own GPUs costs about $1 to $2 per million tokens versus ~$20 for a prior-generation frontier model.
IBM novelty 2/3 · verifiable 2/3 Nikesh Arora says his company inspects endpoint traffic with on-endpoint machine learning at about $20 to $30 per endpoint per year, and trains small language models to 99.9% accuracy rather than using frontier tokens.
PANW novelty 2/3 · verifiable 2/3 Krishna says agents write probably half the code inside his company, and that it is superior to what was written before.
IBM novelty 2/3 · verifiable 1/3 Krishna says DARPA's quantum benchmarking program is down to about 13 companies from a couple of hundred, with ~2030 the midpoint for quantum breaking today's cryptography.
IBM novelty 2/3 · verifiable 2/3
Pichai discloses Gemini launch usage (~25T tokens in 72 hours, 3x Gemini 3.0), capacity constraints, and trade-offs where Cloud forgoes short-term revenue for frontier compute.
Sundar Pichai (Alphabet CEO)
2:43Google's newest Gemini model (referred to as Argon) used almost 25 trillion tokens in its first 72 hours, about 3x Gemini 3.0, and Google is capacity constrained serving it.
GOOGL novelty 3/3 · verifiable 2/3 Sundar Pichai (Alphabet CEO)
22:50Google Cloud is at any given moment forgoing short-term revenue because of compute tradeoffs made to support frontier AI development.
GOOGL novelty 2/3 · verifiable 2/3 Sundar Pichai (Alphabet CEO)
19:17A customer asking for compute for 2027 now is incredibly difficult to serve, and Google leased some short-term compute last year to accommodate customers.
GOOGL novelty 2/3 · verifiable 2/3 Sundar Pichai (Alphabet CEO)
38:12In the four weeks since Argon launched, Google saw a 30% increase in 'sales' (as stated) and about a one-third decrease in rollbacks.
GOOGL novelty 2/3 · verifiable 1/3 Sundar Pichai (Alphabet CEO)
39:29Google says it is bringing incremental power and covering 100% of the cost of its power needs for these data center projects.
GOOGL novelty 1/3 · verifiable 2/3
Specific technical read-through on optical interconnect: Credo's first active light cable uses microVCSELs not microLEDs, with speed/reach numbers and supplier names; analyst hosts, no new company financials.
Credo's first active light cable shown at OCP uses microVCSELs, not the microLED technology it bought through its Hyperloom acquisition.
CRDO novelty 2/3 · verifiable 2/3 MicroLEDs top out near 3 to 5 Gbps per emitter, so about 300 parallel lanes are needed to reach 1 Tbps.
novelty 2/3 · verifiable 2/3
MicroVCSELs can reach 50 meters at 50 Gbps or more per lane, about 10 times the data rate of a microLED.
novelty 2/3 · verifiable 2/3
Lumentum and Coherent are working on 1060 nanometer VCSELs for this application.
Credo acquired the microLED startup Hyperloom (as named in the transcript) for $92 million.
CRDO novelty 1/3 · verifiable 3/3
Analyst conference roundup with a few checkable items (Nvidia Rubin Ultra CPO, TSMC site checks, ASML bonder, Terafab tool orders) but secondhand and unlabeled; mostly known themes.
Nvidia said at OCP that Rubin Ultra, due next year, would use co-packaged optics for scale-up, earlier than the 2028 Feynman timing most expected.
NVDA novelty 2/3 · verifiable 2/3 TSMC said on its earnings call that it sends its own employees to data centers to check how far along they are.
TSM novelty 2/3 · verifiable 3/3 ASML is reportedly developing a hybrid bonding tool; a host says Besi and Applied Materials each took about 10 years to mature theirs.
A host says a person he spoke to had received tool orders from Terafab, Musk's planned fab, the first tangible sign of activity.
TerafabASML novelty 2/3 · verifiable 1/3 Applied Materials is bullish on hybrid-bonding HBM wafers to stack higher, contrary to HBM buyers at Hot Chips who asked not to stack higher; Applied said it depends on workload.
AMAT novelty 2/3 · verifiable 2/3
5#12
Democratizing Security for the Agentic Web with CloudFlare CTO Dane Knecht
Cloudflare CTO on security workload and AI-first restructuring; some new color (weekly Log4j-class events, red team expansion, first-ever layoff, agent payments) but little quantified.
Dane Knecht (Cloudflare CTO)
2:36Cloudflare now sees vulnerabilities as serious as Log4j happening weekly rather than rarely, and expects most of its teams' time over the next 18 months to be spent on security.
NET novelty 2/3 · verifiable 1/3 Dane Knecht (Cloudflare CTO)
35:46Cloudflare has one red team today and expects about four or five by the end of next year.
NET novelty 2/3 · verifiable 2/3 Dane Knecht (Cloudflare CTO)
17:43Cloudflare released a monetization gateway that lets sites take microtransactions, including wallets for agents and a stablecoin, and is working with OpenAI on a project called Signals for agent content discovery.
NET novelty 2/3 · verifiable 1/3 Dane Knecht (Cloudflare CTO)
24:20Cloudflare's AI-first restructuring included letting colleagues go, which was the first time the company had ever done so.
NET novelty 2/3 · verifiable 2/3 Dane Knecht (Cloudflare CTO)
30:26Cloudflare decided this past year to hire a thousand interns, at a company the host described as about 5,000 people.
NET novelty 1/3 · verifiable 2/3
Private-company pitch with specifics: Point2's 1.6T RF-over-plastic module at under 10 W (~5 pJ/bit) and a cited investor list, but no revenue or named customer; volume is months away.
Sean Park (Point2 CEO)
8:53Point2 Technology's investors include Nvidia, Arm, Taiwan foundry UMC and cable maker Molex.
NVDAARMPoint2 Technology novelty 2/3 · verifiable 2/3 Sean Park (Point2 CEO)
38:28Point2's first-generation 1.6T module uses under 10 W per module at about 5 pJ/bit, claimed 3x better than optics, with a target under 2 pJ/bit for near-packaged and co-packaged use.
Point2 Technology novelty 2/3 · verifiable 1/3
Sean Park (Point2 CEO)
23:01Point2 says its interconnect chip is built on a 55 nm process, and that its tape-out costs about 50x less than a GPU tape-out (as stated).
Point2 Technology novelty 2/3 · verifiable 1/3
Sean Park (Point2 CEO)
27:31Point2's first-generation cable cuts cable volume 4x and weight to one-sixth versus copper interconnects.
Point2 Technology novelty 1/3 · verifiable 1/3
Sean Park (Point2 CEO)
44:51Point2 is only now delivering product to customers for lab testing, after which qualification will take many months.
Point2 Technology novelty 1/3 · verifiable 1/3
Kong CEO gives fleet-level figures (~500T tokens, 140x growth as stated; ~1/3 open-weight use) and routing savings claims; vendor self-reported, unaudited.
Aghi Marietti (Kong CEO)
14:35Kong sees about 500 trillion tokens through its AI gateways, growing 140x year over year (as stated; he also said about 100 times bigger than a year ago).
Kong novelty 3/3 · verifiable 1/3
Aghi Marietti (Kong CEO)
10:22About one-third of enterprise model usage Kong sees is now open-weight models, and sophisticated enterprises use dozens of models.
Kong novelty 2/3 · verifiable 1/3
Aghi Marietti (Kong CEO)
24:26Kong says its intelligent model routing, shifting tokens to lower-cost models for the same output, saves large enterprises hundreds of millions of dollars annually (as stated).
Kong novelty 2/3 · verifiable 0/3
Aghi Marietti (Kong CEO)
13:00Marietti says only about 1% of enterprise is agentic, with 95-98% not yet agentified, mostly chatbots and coding tools.
Kong novelty 1/3 · verifiable 0/3
Aghi Marietti (Kong CEO)
15:51Kong's AI token cost management products, previewed at its summit, go GA in a few weeks.
Kong novelty 1/3 · verifiable 2/3
5#15
The New Apex Predator: CrowdStrike CEO George Kurtz on Adversaries Armed With AI Agents
CrowdStrike CEO gives firsthand but promotional detail: $10 vs 3 cents per remediation for its own model, Nvidia Nemotron tie-up, 90 agents per employee, bank CIO shifting to open weights.
George Kurtz (CEO, CrowdStrike)
21:08Kurtz says general remediation with the top frontier cyber model costs about $10 versus about 3 cents with CrowdStrike's own post-trained model.
CRWD novelty 3/3 · verifiable 2/3 George Kurtz (CEO, CrowdStrike)
15:10CrowdStrike built its Safe Mind security models (red Tempest, blue Solano) with Nvidia and uses Nemotron as a main base model on the blue side.
George Kurtz (CEO, CrowdStrike)
18:42Kurtz says an employee can have about 90 AI agents on average, which he thinks is low, a bigger opportunity for CrowdStrike than endpoint detection alone.
CRWD novelty 2/3 · verifiable 1/3 George Kurtz (CEO, CrowdStrike)
24:33A New York bank CIO told Kurtz the bank will move most AI work to self-trained open-weight models after token costs spiked, keeping frontier models for high-value alpha cases.
CRWD novelty 2/3 · verifiable 1/3 George Kurtz (CEO, CrowdStrike)
17:05CrowdStrike now has 33 modules, up from one at its founding.
CRWD novelty 1/3 · verifiable 3/3
Microsoft CEO on security and strategy: OpenAI IP access through 2032 and a deliberate lag behind the frontier are useful context, but few new numbers and much of it is opinion.
Satya Nadella (Microsoft CEO)
22:35Microsoft is deliberately building its own models roughly 3 to 6 months behind the frontier at around 20 percent of the cost, which Nadella calls good for shareholders (6 months is the interviewer's figure; 20 percent is Nadella's illustrative figure).
MSFT novelty 2/3 · verifiable 1/3 Satya Nadella (Microsoft CEO)
30:39Rajesh Jha, who led the Office 365 effort, is retiring from Microsoft.
MSFT novelty 2/3 · verifiable 3/3 Satya Nadella (Microsoft CEO)
20:53Microsoft has access to OpenAI's models and IP through 2032.
MSFTOpenAI novelty 1/3 · verifiable 3/3 Satya Nadella (Microsoft CEO)
13:30Microsoft's consumer Office 365 subscription has more than 100 million subscribers.
MSFT novelty 1/3 · verifiable 3/3 Satya Nadella (Microsoft CEO)
16:37Nadella expects enterprises to hold back model deployment until they have a control and assurance layer that sits outside the model.
MSFT novelty 1/3 · verifiable 1/3
News show with some primary detail: SpaceX Q3 results and Q1 2027 revenue guide, Gerstner's SpaceX compute build vs the 43 GW US forecast, Anthropic revenue forecast, United's fuel pass-through.
SpaceX guided Q1 2027 revenue to $33B-$37B versus a roughly $22B analyst average, and Starlink subscribers grew to 13.6M from 12M in the prior quarter.
novelty 2/3 · verifiable 2/3
Brad Gerstner (Altimeter)
16:51Brad Gerstner said SpaceX is standing up 200,000 GB300s per month, 3 GW by year end and 10 GW next year, about 25% of the 43 GW forecast for total US compute additions next year.
novelty 2/3 · verifiable 1/3
Brad Gerstner (Altimeter)
35:41Altimeter's forecast has Anthropic at about $140-150B of GAAP revenue next year, which at 10x-12x implies about $1.5T-$2T valuation.
Anthropic novelty 2/3 · verifiable 1/3
SpaceX reported its first-ever profit for Q3, with revenue up more than 160% year over year and its AI business up ninefold.
novelty 1/3 · verifiable 2/3
Scott Kirby (United CEO)
44:37United expects to pass on about 72% of its fuel cost increase and wants to get to 100%.
UAL novelty 1/3 · verifiable 2/3
Altman is a top-access source but gives few numbers: claims OpenAI internal math results beat any public model, expects a disease-drug candidate within years, and calls infrastructure scale the moat.
Altman says OpenAI's internal models are ahead of public ones, citing math results that no widely available model can approach.
OpenAI novelty 2/3 · verifiable 1/3
Altman says OpenAI's own chips, robots that build data centers and supply-chain bets could yield far lower-cost, higher-efficiency-per-watt compute than anyone thinks possible.
OpenAI novelty 2/3 · verifiable 1/3
Altman hopes AI yields a candidate treatment for a major disease within the next couple of years, with the FDA process still to follow.
OpenAI novelty 1/3 · verifiable 1/3
Altman says the lasting moats are customer integration, network effects and data, and for OpenAI the scale of its infrastructure, if it has the most cost-effective AI infrastructure.
OpenAI novelty 1/3 · verifiable 1/3
Altman would bet a lot that demand for compute keeps rising, with periods of supply constraint and glut along the way.
novelty 0/3 · verifiable 0/3
Private robot-data company CEO gives useful industry detail (data recipe, 100 Gb lines, hundreds of petabytes) but no public ticker and little checkable; some promotion.
Josh Gao (CEO, Mecka AI)
32:59Robot training is converging on about a million hours of human data, tens of thousands of hours of UMI data and a few hours of in-domain teleoperation.
Mecka AI novelty 2/3 · verifiable 1/3
Josh Gao (CEO, Mecka AI)
14:28Mecka has installed 100 GB (transcribed; description says 100-gigabit) upload lines to handle hundreds of petabytes of data.
Mecka AI novelty 2/3 · verifiable 2/3
Josh Gao (CEO, Mecka AI)
3:31Gao says robotics pre-training data is already a large market, unlike LLM pre-training data, which never became one.
Mecka AI novelty 2/3 · verifiable 1/3
Josh Gao (CEO, Mecka AI)
19:34Collected human-demonstration data for robotics has reached millions of hours, versus about 4,000 hours in Ego4D two years ago.
Mecka AI novelty 1/3 · verifiable 1/3
Josh Gao (CEO, Mecka AI)
23:18Gao expects pick-and-place and packaging manipulation tasks on assembly lines to be solved within the next year or two.
Mecka AI novelty 1/3 · verifiable 1/3
Two supply-chain analysts on Terafab: TSMC CEO's public interest in spare capacity, Intel sharing its process flow and aggressive hiring from Samsung and Intel; mostly sourced chatter, hard to verify.
Intel has shared its process flow and tool list with Terafab, according to people the host spoke to.
INTCTerafab novelty 3/3 · verifiable 1/3 TSMC's CEO said on its earnings call that TSMC is looking at Texas and talking to Terafab about using some of its spare capacity.
TSMTerafab novelty 2/3 · verifiable 3/3 Tesla and SpaceX have recruited heavily from Intel, including its fab manager, and from Samsung's Taylor, Texas fab, offering roughly 1.5 to 2 times prior market pay (as stated).
Musk's company rented its Colossus data center to Anthropic at about 1.5 million or billion dollars a month (unit unclear as stated), and Musk later said he might take the capacity back.
Anthropic novelty 2/3 · verifiable 1/3
Leaked secondhand Anthropic investor-day figures (Q3 revenue $16.8B, NRR 900%) and CoreWeave CDS math; rest is panel opinion on the neocloud debt story.
CoreWeave's credit default swap spread spiked to about 953 basis points that week, which the speaker said implies about a 16% annual and about 55% five-year chance of default.
CRWV novelty 2/3 · verifiable 2/3 Reading secondhand reports of Anthropic's investor day, the host cites Q3 revenue of $16.8B (over $5B a month), 900% net revenue retention, positive free cash flow, 85% adjusted gross margin and 11% adjusted operating margin; ARR estimates of $70B to over $100B are investors' guesses, not company disclosures.
Anthropic novelty 2/3 · verifiable 1/3
CoreWeave says its payback period is 2.5 years on typical 5-year deals versus roughly one year for SpaceX (as stated), with SpaceX monetizing about $50B per gigawatt versus about $20B for CoreWeave on deals cancellable within 90 days.
CRWV novelty 2/3 · verifiable 1/3 Coinbase fine-tuned the Qwen 3.5 9B open-weight model for fraud detection and in testing cut fraudulent transactions by about 30%, while faster and much cheaper than Opus 4.5, per a Brian Armstrong tweet.
COIN novelty 2/3 · verifiable 2/3 CoreWeave gets 75% of its revenue from three customers, and some of those customers are lowering their future revenue expectations.
CRWV novelty 1/3 · verifiable 2/3
5#22
Bloomberg Tech · Tier 2 · 2026-10-22
Markets-recap show but with filing-level Meta facts: spending commitments up to $825B and record 5-year CDS; also Tesla FSD take rate. Mostly known earnings and analyst opinion; capped at 5.
Meta's filing shows spending commitments of $825 billion, up from roughly $700 billion in July (July figure as recalled by the host).
META novelty 2/3 · verifiable 3/3 Meta's 5-year credit default swap spread is at a record level since trading began.
META novelty 2/3 · verifiable 3/3 Meta's new non-voting Class C shares let Zuckerberg sell 20 percent immediately and the rest over eight years, in return for staying with the company for eight years.
META novelty 2/3 · verifiable 3/3 analyst (New Street Research)
9:57About 75 percent of new Tesla vehicles in the US are sold with FSD, including buyers who get the first month free, up about 20 points sequentially (per a New Street analyst citing the CFO).
TSLA novelty 2/3 · verifiable 2/3 Meta's third-quarter ad revenue was above $63 billion versus a street estimate below $62 billion, with price per ad up about 14 percent.
META novelty 1/3 · verifiable 3/3
5#23
Bloomberg Businessweek · Tier 2 · 2026-10-19
Anchor-led market roundup capped at 5; best items are SpaceX cloud at a $25-30B run rate versus a $100B ARR guide, Anthropic ARR growth slowing, and Ives' 13-to-1 memory demand/supply claim.
Mandeep Singh (Bloomberg Intelligence)
2:03SpaceX's cloud segment guided to $100 billion ARR by year-end but is closer to a $25-30 billion run rate, with Anthropic and Google as its only named customers.
SPACEX novelty 3/3 · verifiable 3/3 Nscale and Anthropic are in talks on a roughly $45 billion compute deal (Bloomberg report) while investors question Nscale's $30 billion valuation as perhaps worth half.
NscaleAnthropic novelty 3/3 · verifiable 2/3
Perplexity built its own agent sandbox because renting a third-party one would cost an extra $50 million a year or so at its traffic.
Perplexity novelty 3/3 · verifiable 1/3
Mandeep Singh (Bloomberg Intelligence)
5:09Frontier labs have grown revenue about fivefold since the start of the year, but Anthropic's monthly ARR growth now appears to be slowing.
Anthropic novelty 2/3 · verifiable 2/3
Memory chip demand versus supply is 13 to 1 based on checks in Korea last week.
novelty 2/3 · verifiable 1/3
CNBC morning-show recap: OpenAI round, SpaceX tower and bond reports, Moderna data are second-hand reports already in the news; capped as a news show.
Semafor reports the lowest-rated bonds in SpaceX's $40 billion debt deal are expected to yield 8.3%.
SpaceX novelty 2/3 · verifiable 3/3
OpenAI is working on a roughly $30 billion funding round valuing it at about $1.25 trillion before new capital, led by UAE investors.
OpenAI novelty 1/3 · verifiable 2/3
Bloomberg reports SpaceX seeks access to at least 80,000 cell towers via talks with Crown Castle for a fourth national wireless network.
Moderna and Merck's melanoma vaccine with Keytruda cut the risk of recurrence or death by 42% versus Keytruda alone, yet Moderna stock fell.
MRNAMRK novelty 1/3 · verifiable 3/3
Warp Sciences exits stealth with $215M for a US pre-clinical chemistry CRO; no revenue or customers disclosed; reshoring argument is mostly opinion.
Kevin Weil (Warp Sciences CEO)
13:13Warp Sciences raised $215 million, led by Lightspeed (transcribed 'Lightseed'), with IBP, Mayfield, Lux, Menlo and Scribble Ventures participating.
Warp Sciences novelty 2/3 · verifiable 2/3
Kevin Weil (Warp Sciences CEO)
2:07Warp is building a next-generation pre-clinical CRO starting with small-molecule chemistry in America, and is not developing drugs of its own.
Warp Sciences novelty 1/3 · verifiable 2/3
Kevin Weil (Warp Sciences CEO)
3:57Weil says the 3-5 years of making and testing molecules in drug discovery has been outsourced entirely to China.
novelty 1/3 · verifiable 0/3
CNBC morning-show recap capped low; some checkable items (Tesla Q3 positive FCF and capex cut, Meta non-voting share plan, Nasdaq IPO share) but little new primary information.
host (Andrew Ross Sorkin)
22:29Meta is reportedly planning to issue new non-voting shares to fund its AI buildout.
META novelty 2/3 · verifiable 2/3 Bloomberg reports Musk wants a Tesla-SpaceX deal done before President Trump leaves office and may announce something in the first half of next year.
TSLA novelty 2/3 · verifiable 2/3 Adena Friedman (Nasdaq CEO)
14:53Nasdaq had 8 of the top 10 IPO wins for the quarter, 7 of the top 10 for the year, $49 billion of index-business flows last quarter and 14% revenue growth.
NDAQ novelty 2/3 · verifiable 3/3 host (Andrew Ross Sorkin)
1:53Tesla's Q3 beat expectations with surprise positive free cash flow, a cut to its annual capex forecast, a drop in automotive revenue and services revenue up by a third.
TSLA novelty 1/3 · verifiable 3/3 The Kospi rose 115% in six months this year then fell 40% in four weeks, driven by Hynix and Samsung.
novelty 1/3 · verifiable 3/3
Meta product director explains ad delivery (Andromeda, LTV products in beta) for marketers; mostly known mechanics, few figures, little investor-relevant new data.
Sukhesh Miryala (Director of Product Management, Meta)
21:47Meta is alpha and beta testing lifetime-value optimization products that have not yet shipped.
META novelty 2/3 · verifiable 2/3 Sukhesh Miryala (Director of Product Management, Meta)
28:17Meta's director of product management says Andromeda lets the retrieval stage rank about 10,000 more ads per impression than before (figure as stated).
META novelty 1/3 · verifiable 1/3 Sukhesh Miryala (Director of Product Management, Meta)
38:24Meta says its ad supply is healthy and growing with user engagement, and that CPMs are an outcome of auction supply and demand rather than a set lever.
META novelty 1/3 · verifiable 2/3 Sukhesh Miryala (Director of Product Management, Meta)
52:49Meta says better relevancy ranking yields more clicks and conversions from the same impressions, so advertisers need fewer impressions for the same value.
META novelty 1/3 · verifiable 1/3 Connor Rolain (Head of Growth, HexClad)
11:52HexClad's growth head says about 50% of its Meta ad account is optimized against value rather than lowest cost.
METAHexClad novelty 1/3 · verifiable 1/3
Figma CEO interview is mostly opinion on design and AI; only checkable items are the 48% growth figure (host) and no meaningful Claude Design impact.
Dylan Field (CEO, Figma)
21:46Field says App Store app counts are growing exponentially while app usage and utilization are flat.
novelty 2/3 · verifiable 2/3
Dylan Field (CEO, Figma)
38:39Field cites an OpenAI math release the night before covering 700-plus papers on famous problems.
OpenAI novelty 2/3 · verifiable 2/3
A host says Figma grew 48% year over year in the quarter reported after the Claude Design announcement (figure and metric as stated by host).
FIG novelty 1/3 · verifiable 3/3 Dylan Field (CEO, Figma)
14:09Field says Anthropic's Claude Design launch did not significantly affect Figma's business.
FIGAnthropic novelty 1/3 · verifiable 2/3
Startup guest on AI materials; $500M CHIPS award and Claude distribution are new but company is private and plans are early. Industry revenue figure looks implausible and is flagged.
Shalini Sharma (SandboxAQ)
17:45SandboxAQ received a $500 million CHIPS research award from the US Department of Commerce, to be spent entirely on semiconductor materials research.
SandboxAQ novelty 2/3 · verifiable 3/3
Shalini Sharma (SandboxAQ)
23:10SandboxAQ has released its large quantitative models on Claude with usage-based inference pricing.
SandboxAQAnthropic novelty 2/3 · verifiable 2/3
Shalini Sharma (SandboxAQ)
29:40SandboxAQ names Vertiv and Eaton, with Schneider Electric, as likely users of its battery-chemistry models for fab backup power.
VRTETN novelty 2/3 · verifiable 1/3 Shalini Sharma (SandboxAQ)
20:25More than 90 percent of the supply chain for the neodymium used in permanent magnets for chip equipment sits outside the US, which is why the award targets rare-earth-free magnets.
novelty 1/3 · verifiable 2/3
Shalini Sharma (SandboxAQ)
18:41Semiconductor industry revenue is projected to go from about $800 billion last year to over $1.3 trillion this year (as stated; the jump looks too large).
novelty 0/3 · verifiable 3/3
Retail-investor valuation walk-through of Ouster using public Q2 2026 results; numbers are checkable but not new, and the thesis is a hold with no fresh information.
Ouster's Q2 2026 revenue grew 56 percent year over year to $55 million with GAAP gross margin of 49 percent.
OUST novelty 1/3 · verifiable 3/3 Ouster shipped over 17,000 sensors in Q2 versus 5,500 a year earlier while revenue rose only 56 percent, implying sharply falling average selling prices.
OUST novelty 1/3 · verifiable 3/3 Ouster guided Q3 revenue to a midpoint of about $56 million, roughly flat with Q2.
OUST novelty 1/3 · verifiable 3/3 Ouster's GAAP operating margin is about negative 36.7 percent and free cash flow margin about negative 28 percent.
OUST novelty 1/3 · verifiable 3/3 A December 2025 licensing payment from Hesai inflated Ouster's reported growth, and normalized growth is in the high 40 percent range (host's estimate).
4#31
The Builders · Tier 2 · 2026-10-22
Thiel interview is mostly philosophy and AI-politics; investable bits are an off-hand ~$2T Anthropic IPO value, a take on AI moat weakness and an Apple R&D-versus-buyback jab.
Thiel says Anthropic would be worth probably close to $2 trillion at its IPO, making it roughly the sixth or seventh largest company by market cap, built in six to seven years.
Anthropic novelty 2/3 · verifiable 1/3
Thiel says AI company pitches are weak on a monopoly story, that Anthropic may be in pole position over OpenAI, and that Google, Meta and Grok are not far behind.
OpenAIAnthropicGOOGLMETA novelty 1/3 · verifiable 1/3 Thiel says after Steve Jobs Apple did nothing new, with by far the lowest R&D spend and the largest share buybacks (as stated, no figures given).
AAPL novelty 1/3 · verifiable 3/3
4#32
The Builders · Tier 2 · 2026-10-20
Musk interview taped Aug 6, 2026 so dated; says xAI supplies compute to Anthropic and a 200 GW/yr US solar build (as stated); rest is opinion on AI and robots.
Musk says he is putting together 200 GW a year of US solar production capacity (as stated), which he says is much more than currently exists.
novelty 2/3 · verifiable 1/3
Musk says xAI is providing compute to Anthropic.
xAIAnthropic novelty 1/3 · verifiable 1/3
Musk estimates digital superintelligence would raise world output of goods and services by 20-30%, about $30T a year incremental.
novelty 1/3 · verifiable 0/3
Musk says the US national debt can only be solved with AI and robots, not by cutting spending.
novelty 0/3 · verifiable 0/3
Venture partners' vision talk; the only checkable item is a16z expanding its crypto fund into a broader Frontier mandate. Rest is opinion on AI, open source and media with little specific data.
a16z is expanding its crypto fund into a broader 'a16z Frontier' mandate to back founders working outside its existing verticals.
a16z novelty 2/3 · verifiable 3/3
a16z partner (unclear which)
41:47Open-weight models trail closed frontier models by roughly 6 to 12 months depending on how it is counted.
novelty 1/3 · verifiable 2/3
a16z partner (unclear which)
39:00Open-source models are growing faster than closed models in token consumption on OpenRouter data.
novelty 1/3 · verifiable 2/3
A chart from a16z's team puts AI at roughly 3 percent of measured enterprise use (host's approximate recollection, as stated).
novelty 1/3 · verifiable 1/3
Off-topic for markets (F1 management and talent); only investable content is Mercedes F1 team's GBP 4.6bn valuation, about 15x EBIT, with no listed ticker.
Mercedes F1 generates roughly $1 billion in revenue at about a 30% margin (about $300 million EBIT), so the valuation is around 15x EBIT.
Mercedes-AMG F1 novelty 2/3 · verifiable 2/3
Wolff says that when he sold part of the team last year the multiple supported the stated valuation, and he would probably get about 30% more today.
Mercedes-AMG F1 novelty 2/3 · verifiable 1/3
A recent transaction valued the Mercedes F1 team at 4.6 billion pounds, which the host puts at about a 7x increase from a few years earlier.
Mercedes-AMG F1 novelty 1/3 · verifiable 2/3
The F1 cost cap of a bit over $200 million a year counts all engineering spend, which Wolff says pushed out speculative long-term R&D.
Mercedes-AMG F1 novelty 1/3 · verifiable 3/3
Culture and education framing of AI with Mollick; few investable specifics. Newest items: Meta's Muse agent at a couple million users and Amazon blocking it, both as stated by the guest.
Meta's Muse personal agent has already been adopted by a couple million people with high stickiness (as stated, no source given).
META novelty 2/3 · verifiable 1/3 Amazon has banned Meta's Muse agent from its site.
The average best AI model changes every 70 days or less; GPT-4 is now the 181st best model and 19 phone-sized models beat it (as stated).
novelty 1/3 · verifiable 2/3
Only about 3% of individuals in the US personally pay for any AI subscription.
novelty 1/3 · verifiable 2/3
Business-history deep dive on Seinfeld syndication economics; only investor-relevant fact is the reported lower 2026 Netflix renewal (US and Canada only); otherwise entertainment history.
Netflix's 2026 renewal of Seinfeld streaming rights was reportedly slightly lower than the 2019 deal (reported well over $500 million for five years) and covers only the US and Canada, not worldwide.
The hosts estimate about $5 billion has gone to Seinfeld's cap table from all syndication, streaming and DVD deals, with Sony holding distribution rights and re-selling them every roughly five years.
SONY novelty 1/3 · verifiable 1/3 The hosts estimate the show cost about $250 million to make in total and has returned well above $7 billion including NBC's roughly $2 billion first-run ad profit, a roughly 30x return.
novelty 1/3 · verifiable 1/3
3#37
Better Offline · Tier 2 · 2026-10-21
Biographical profile of Larry Ellison; adds only that he holds about 40% of Oracle and a pulled share sale, little new on markets.
Tom Dotan (guest, journalist)
15:11Larry Ellison now owns about 40% of Oracle, up from the mid-20s a decade ago, after Oracle's stock buybacks.
ORCL novelty 1/3 · verifiable 2/3 Tom Dotan (guest, journalist)
42:22Oracle has been turned cash-flow negative by debt-funded capex for AI data centers, and it owns 15% of TikTok.
ORCL novelty 1/3 · verifiable 2/3 A planned Ellison sale of about $7B of Oracle shares was pulled back within a day, and he is instead taking another margin loan against the shares.
ORCL novelty 1/3 · verifiable 2/3 Oracle has done about $190B or more of stock buybacks since inception.
ORCL novelty 0/3 · verifiable 2/3
Founder-story interview with Kalshi's COO on its fight with the CFTC; private company, mostly narrative, a few checkable facts such as 80 to roughly 7,000 live contracts and a delayed clearinghouse application.
Luana Lopes Lara (Kalshi COO)
34:54Kalshi went from about 80 live contracts right before the election in question to roughly 7,000 live contracts now (figure as stated, with the speaker unsure of the exact number).
Kalshi novelty 2/3 · verifiable 2/3
Luana Lopes Lara (Kalshi COO)
32:20The CFTC took about two years to act on Kalshi's clearinghouse application against a six-month legal deadline, during Kalshi's lawsuit.
Kalshi novelty 2/3 · verifiable 2/3
Luana Lopes Lara (Kalshi COO)
33:36An appeals court declined to stay the ruling allowing Kalshi to list election contracts, clearing the launch for that election.
Kalshi novelty 1/3 · verifiable 3/3
Kalshi is reportedly valued at $22 billion.
Kalshi novelty 1/3 · verifiable 2/3
Kalshi had raised about $100 million of investor money by the time it decided to sue the CFTC.
Kalshi novelty 1/3 · verifiable 2/3
1#39
Odd Lots · Tier 1 · 2026-10-23
Biography of economist Maria Edgeworth; economic history with no investable content.
1#40
Odd Lots · Tier 1 · 2026-10-22
Political economy interview on market socialism with no companies, tickers or market data; not investable.
1#41
Odd Lots · Tier 1 · 2026-10-21
Academic interview on institutions and failed states with no companies, tickers or market data; not investable.