Nov 11, 2026 · 1h 59m
Veterans, defense factories and chips reshape the AI debate
🇺🇸 Veterans Day 🇺🇸, The Great Hobby Debate, AI is BACK! | Dan Goldenberg, Darren Farber, Walter Goodwin
The episode connects veteran employment, defense manufacturing and AI hardware to the practical constraints shaping national security and technological competition.
- 1The Call of Duty Endowment treats veteran employment as a measurable outcome, emphasizing nonprofit accountability and cost per placement.
- 2Darren Farber argues that defense reindustrialization requires acquisitions, automation, skilled workers and commercial scale beyond venture-backed enthusiasm.
- 3Walter Goodwin sees memory bandwidth, rather than chip specialization alone, as Fractile’s durable advantage for increasingly demanding AI models.
Don't miss
Walter Goodwin explains why Fractile is prioritizing dramatically higher memory bandwidth through stacked DRAM as AI models and KV caches grow.
The brief
The hosts move from NVIDIA’s competitive position, Anthropic’s prospects and stock-market bubble theories to a playful argument over whether elite founders need genuine hobbies.
Dan Goldenberg explains how the Call of Duty Endowment replaced check-writing philanthropy with data-driven veteran employment programs, measuring nonprofit performance and cost per placement.
Darren Farber makes the case for rebuilding the defense industrial base through acquisitions, automation, trade schools and components designed to serve multiple weapons platforms.
The defense discussion links Ukraine’s battlefield lessons, drones, open-source AI and civilian data exposure to a military environment where commercial technology increasingly shapes operations.
Walter Goodwin presents Fractile’s chip strategy: use stacked DRAM to increase memory bandwidth, helping AI systems handle growing models, weights and KV caches faster.
Taken together, the conversations argue that AI competition will be decided not only by models and valuations, but by employment systems, manufacturing capacity and hardware architecture.
What was said on this episode
40 statements · 32 positive · 6 negative · 2 neutral
Brian Chesky’s Airbnb donation is San Francisco’s largest single donation.
“The $100 million donation from the founder and CEO of Airbnb is the largest single donation the city of San Francisco has ever received.”
Listen at 4:02
Brian Chesky’s donation may encourage similar donations.
“now I imagine it will spark. other donations like it.”
Listen at 5:25
Mythos reduced the time needed to run on AMD hardware to three days.
“it only took Mythos three days to figure out how to run itself on AMD hardware compared to six to 12 months.”
Listen at 13:11
Rapid software generation makes the CUDA moat vulnerable.
“you had to be bearish on CUDA”
Listen at 15:19
Building extinction-risk AI could violate fiduciary duties to shareholders.
“you have a fiduciary duty to maximize shareholder return, and you are violating that by building something that will destroy humanity.”
Listen at 19:08
Anti-AI shareholders can potentially sue an AI company to raise concerns.
“you can also file a lawsuit, hire a law firm”
Listen at 21:14
The S&P 500 rose above 7,600 as investors returned to technology stocks.
“The S&P 500 surges above 7,600 as dip buyers continue to rotate back into technology stocks.”
Listen at 24:54
Demand for computing capacity continues to grow.
“demand for compute continues to grow.”
Listen at 28:16
Long-term AI adoption trends are expected to continue.
“the long-term trends will just continue”
Listen at 28:47
AI models are expected to keep improving on the GDPX benchmark.
“I would expect that that continues and everyone hill climbs that over the next year.”
Listen at 29:18
Accenture could become a strong AI-auditing provider.
“It made me... Very optimistic about Accenture as an AI auditor”
Listen at 31:16
NVIDIA is expected to report strong earnings next week.
“I imagine they'll do very well.”
Listen at 42:07
- Dan Goldenbergon Call of Duty Endowment veteran placementsPositive1:10:50
The Call of Duty Endowment placed one veteran per $686 spent.
“our last year, every $686 placed a veteran in a job.”
Listen at 1:10:50
- Dan Goldenbergon Call of Duty Endowment veteran placementsPositive1:10:55
The Endowment’s placement cost is one-fourteenth of federal efforts.
“one-fourteenth of the cost per placement, one-fourteenth of the federal government's efforts.”
Listen at 1:10:55
- Dan Goldenbergon Call of Duty Endowment veteran placementsPositive1:11:13
The Endowment placed more veterans than the federal government for three years.
“for the last three years, we've placed more vets in jobs than the federal government has.”
Listen at 1:11:13
- Dan Goldenbergon Trust-based philanthropyNegative1:14:10
Trust-based philanthropy without accountability does not produce results.
“The data shows that does not work.”
Listen at 1:14:10
- Dan Goldenbergon Veteran corpsmen and medicsNegative1:15:18
Half of veteran corpsmen and medics seeking healthcare work cannot obtain healthcare jobs.
“Half of them who want to work in health care can't get jobs in health care.”
Listen at 1:15:18
- Dan Goldenbergon AI in nonprofitsPositive1:19:46
Nonprofits have greater potential upside from AI than commercial entities.
“I'll argue that nonprofits have far more upside with AI than commercial entities.”
Listen at 1:19:46
- Dan Goldenbergon AI in veteran employment counselingPositive1:20:16
AI should free nonprofit counselors for more direct veteran support, not replace them.
“I don't want a single counselor laid off because of AI. I want that counselor freed up.”
Listen at 1:20:16
- Dan Goldenbergon AI in veteran employment servicesPositive1:21:30
AI will transform nonprofit veteran employment services.
“I'm absolutely convinced this is going to be transformative.”
Listen at 1:21:30
Ignium aims to supply components across many weapon systems.
“we want to be kind of this permissive supplier to everyone's system at the end of the day”
Listen at 1:24:25
Defense manufacturing should adopt automation quickly.
“automation can't come too soon”
Listen at 1:26:24
Defense manufacturers should combine automation with mentor-protégé programs.
“our goal is a combination of leveraging the future through automation and then making sure the internals of the company have very strong mentor-protege programs”
Listen at 1:26:49
Modern warfare combines traditional and emerging technologies.
“war is primarily a combination of old and new.”
Listen at 1:28:23
- Walter Goodwinon Defense pyrotechnic manufacturingPositive1:29:56
Certain pyrotechnic manufacturing technologies will remain durable over long cycles.
“we fundamentally feel won't change over a much longer cycle.”
Listen at 1:29:56
- Walter Goodwinon Autonomous Ukrainian resupply systemsPositive1:34:38
Autonomous Ukrainian resupply systems use AI for battlefield navigation.
“the autonomous ones are using a form of artificial intelligence to navigate the battlefield and resupply the lines.”
Listen at 1:34:38
- Walter Goodwinon AI-enabled signals intelligenceNeutral1:35:22
AI is being used aggressively to extract signals intelligence from phones.
“AI is working on that pretty aggressively.”
Listen at 1:35:22
- Walter Goodwinon Defense venture investmentPositive1:39:04
Defense venture investment should produce several high-quality companies and technical leaders.
“my hope is that the reversion to the mean here isn't too much of a whipsaw and we get a few high quality. companies that can, that come out of this”
Listen at 1:39:04
- Walter Goodwinon Defense venture investmentNegative1:41:02
Without defense-sector exits and returns, venture investment will eventually stall.
“if there isn't, it will stall out ultimately.”
Listen at 1:41:02
- Walter Goodwinon U.S. defense manufacturing capacityPositive1:42:53
The United States could rapidly build massive defense factories if national priority demands it.
“if the country decides that this is important, you know, hang on because, you know. We'll take a whole part of the Nevada desert or something, and we will build a multi-million square foot factory to pump it out.”
Listen at 1:42:53
Dual-use defense investments need a path to commercial leadership.
“Unless I have a clear path to be the commercial world leader”
Listen at 1:43:34
Fractile aims to run the largest AI models orders of magnitude faster than competing chips.
“we've been building chips to run the world's largest models faster, orders of magnitude faster if we get this right than any other chip on the planet.”
Listen at 1:46:22
- Walter Goodwinon AI chip manufacturing capacityPositive1:46:48
Chip companies must secure manufacturing capacity now for 2027–2028 production.
“if you want to secure capacity on chip production that you might be doing in the second half of next year, even the first half of 2028, you've got to move today.”
Listen at 1:46:48
- Walter Goodwinon Hyper-specialized AI chipsNegative1:48:33
Hyper-specialized AI chips impose substantial product and amortization risks.
“you can't actually specialize without imposing a bunch of risks.”
Listen at 1:48:33
- Walter Goodwinon AI modelsPositive1:49:28
AI models will continue increasing in size.
“Models of whatever shape and size will keep getting bigger.”
Listen at 1:49:28
- Walter Goodwinon AI model memory bandwidthPositive1:49:32
Growing AI models will increasingly require higher memory bandwidth.
“they seem to be increasingly hungry for a particular thing that you can design your chips around, which is more bandwidth to memory”
Listen at 1:49:32
Fractile projects twenty times the HBM bandwidth of conventional GPUs or XPUs.
“On this first-gen chip, we're projecting 20 times more bandwidth than you would get to HBM on a GPU or XPU.”
Listen at 1:50:38
- Walter Goodwinon AI chip commoditizationPositive1:53:16
AI-chip commoditization could amplify the value of fundamental performance advantages.
“bring on the commoditization because what it means is you can juice so much more from these kind of core advantages.”
Listen at 1:53:16
Running frontier models twenty times faster could become vital for cyber defense.
“the ability to take a front-air model and run it 20 times faster than anything else can. becomes completely vital to that kind of counter-defense.”
Listen at 1:54:39
- Walter Goodwinon Faster AI hardwarePositive1:56:03
Faster hardware can shorten AI red-teaming cycles.
“you also want to be able to kind of compress that red teaming cycle by running on faster hardware, for example.”
Listen at 1:56:03
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.
Weekly email
State of the Pod, every Monday
The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.