Jan 2, 2019 · 55m · a16z
a16z Podcast | From Research to Startup, There and Back Again
John Hennessy · 26m spoken Sonal Chokshi · 10m spoken Martin Casado · 8m spoken Marc Andreessen · 6m spoken
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of the a16z Podcast, Alphabet Chairman and former Stanford President John Hennessy joins Andreessen Horowitz partners Marc Andreessen, Martin Casado, and Sonal Chokshi to discuss the history of RISC processor architecture, the mechanics of commercializing university research, and key leadership lessons across academia and entrepreneurship.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The host holds 20.4% of the talking time here. How this is scored →
The host as informed peer 5.7 Guest teaching 4.7 Guest disagreement 1.1 The host pushing back 3.6
speaking balance: gold is the host, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 25:56 Hennessy rejects the premise that successful tech leaders cannot be humble
When Sonal challenges him on whether top leaders are ever truly humble, Hennessy forcefully reframes the concept, distinguishing humility from indecisiveness and citing Abraham Lincoln's resolute execution of the Emancipation Proclamation over cabinet objections.
Hardest push from the host ▶ 46:45 Marc challenges the mythologizing of historic corporate R&D labsMarc explicitly rejects the standard narrative praising ancient labs like Bell Labs and Xerox PARC, calling himself the skunk at the garden party and arguing they were tiny rounding errors dependent on monopoly profits before venture capital existed.
Biggest teaching moment ▶ 5:12 Hennessy explains data center power costs and RISC efficiency advantagesHennessy educates the hosts on hardware economics, pointing out that after physical server acquisition, electricity is the second largest expense in data centers, giving RISC architectures a decisive edge over CISC regardless of legacy software bases.
The host holds their own ▶ 46:45 Marc delivers an economic counter-analysis of corporate R&D vs venture capitalMarc demonstrates authoritative historical expertise, dismantling the romantic view of 20th-century corporate research labs by detailing their dependence on monopoly status and demonstrating how modern venture-funded innovation operates at far greater scale.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The host as informed peer | Guest teaching | Guest disagreement | The host pushing back | Why |
|---|---|---|---|---|---|---|
| Podcast Title Transition and Branding Sequence | 2 | 4 | 0 | 0 | Sonal provides a standard introduction and synthesizes Hennessy's explanation of RISC architecture using the English language analogy. Hennessy educates the hosts on how simpler instruction sets enable faster and cheaper processing. The interaction is warm, collaborative, and informational. | |
| The Timeline and Mainstream Adoption of RISC | 4 | 4 | 1 | 1 | Marc shows good industry context regarding desktop vs mobile processor adoption. Hennessy details how architectural fragmentation in the 1980s allowed Intel's x86 to maintain dominance until smartphones emerged. The discussion is a collegial exchange of historical market insights. | |
| Technical Comparison: RISC vs. CISC and Power Efficiency | 6 | 5 | 1 | 3 | Martin draws on his experience working under Pat Gelsinger to challenge whether CISC is genuinely dying or serving a distinct market. Hennessy clearly delineates the difference between legacy software base lock-in and fundamental hardware power efficiency. Sonal adds context linking early game console chips to modern GPU acceleration. | |
| Founding MIPS Technologies: From Academic Research to Startup | 4 | 4 | 0 | 1 | Hennessy shares his experience as a reluctant academic entrepreneur founding MIPS after established East Coast companies ignored RISC research. Marc extends the thought by citing the classic industry adage about needing to bludgeon people to adopt good ideas. The tone remains light and mutually agreeable. | |
| Translating Academic Prototypes into Production Tech | 7 | 3 | 0 | 2 | Marc and Martin bring significant founder experience to the discussion, explaining how Mosaic/Netscape and Nicera had to fully rewrite academic code for production. Martin articulates the counterintuitive insight that customers value software more when paying for it than when receiving it for free. Hennessy nods along to the hosts' domain expertise. | |
| Crisis Management in Startups and Stanford University | 2 | 5 | 0 | 0 | Hennessy recounts early startup crisis management at MIPS and explains how those painful lessons guided his decision to execute swift budget cuts at Stanford during the 2008 financial crisis. Sonal asks guiding questions that let Hennessy elaborate on decisive leadership. The dynamic is purely illustrative and non-confrontational. | |
| Startup Evolution: Product Utility vs. Viral Market Adoption | 5 | 4 | 0 | 1 | Martin analyzes how bottom-up developer adoption has made enterprise software selection behave more like viral consumer adoption. Hennessy agrees and notes how executive leadership must become comfortable making calls in gray, uncertain environments. | |
| Evaluating Faculty Co-Founders and Entrepreneurial Commitment | 7 | 5 | 2 | 6 | Martin states a16z's firm stance requiring full-time founder commitment, while Marc plays devil's advocate by asking if encouraging faculty entrepreneurship depletes the university's core educational mission. Hennessy acknowledges the risk of eating seed corn in AI/ML, but defends faculty industry experience as beneficial for teaching and recruiting. | |
| Replicating Silicon Valley Globally and in China | 6 | 5 | 2 | 5 | Marc contrasts cynical and optimistic visions for replicating Silicon Valley globally. Hennessy rejects simplistic binaries, explaining why China will successfully build major tech centers while US regional hubs have failed to catch up to Silicon Valley's widening lead. | |
| Infrastructure Challenges, Local Ecosystems, and Talent Density | 5 | 4 | 0 | 3 | Marc asks if local infrastructure, housing costs, and state government hostility risk strangling Silicon Valley's success, which Hennessy strongly confirms. Sonal highlights middle-management talent density as Silicon Valley's true defensible moat, supported by Hennessy's example of a UIUC startup moving to the Valley. | |
| Leadership Principles: Humility, Decisiveness, and Inverted Org Charts | 6 | 6 | 3 | 7 | Sonal directly challenges Hennessy's book thesis by noting that very few top tech leaders appear humble. Hennessy reframes humility as the willingness to solicit advice and admit mistakes rather than indecisiveness, citing Abraham Lincoln's firm execution of the Emancipation Proclamation despite cabinet opposition. | |
| Institutional Empathy, Fairness, and Financial Aid Policy | 8 | 6 | 3 | 8 | Marc challenges physical university expansion with a sharp analogy, asking if expanding elite campuses is like adding balcony seats to the Globe Theater rather than inventing television. Hennessy responds by explaining Baumol's cost disease in higher education and advocating for technology-enabled learning to bend the cost curve. | |
| Diploma Signaling vs. Skill Certification Models | 7 | 5 | 2 | 6 | Marc brings up Bryan Caplan's sheepskin effect to argue that degree value lies mostly in signaling rather than skill accumulation. Hennessy agrees in part, reinterpreting the signal as perseverance, while pointing out that post-graduate professional education is rapidly shifting toward direct skill certification. | |
| Blockchain, Machine Learning, and Academic Domain Integration | 5 | 5 | 1 | 2 | Sonal references the debate between Vitalik Buterin and academic professors regarding specialized blockchain degrees versus general computer science fundamentals. Hennessy extends the concept to machine learning, stressing the urgent need for interstitial specialists who bridge deep CS knowledge with specific scientific domains. | |
| Real-World Enterprise AI/ML vs. Academic Elegance | 8 | 5 | 0 | 5 | Martin critiques startup pitches that treat AI/ML as the end of theory, emphasizing that domain expertise and clean data are necessary for meaningful results. Hennessy agrees, noting unsupervised learning's limited real-world scope, while Martin contrasts academic elegance with industry's dirty heavy-tail engineering reality. | |
| Startup Focus vs. Academic Breadth and Corporate R&D History | 5 | 5 | 2 | 4 | Sonal pushes back against the distinction between academic ideas and corporate implementation by pointing to Xerox PARC's mission-driven applied research. Hennessy clarifies that legendary corporate labs benefited from long investment horizons, whereas universities now serve as the primary engine for public tech transfer through graduating students. | |
| The Shift of Fundamental R&D to Tech Giants and Venture Capital | 9 | 5 | 2 | 9 | Martin and Marc vigorously reframe industrial R&D history. Marc argues that iconic corporate research labs like Bell Labs and Xerox PARC were small anomaly offshoots of government-sanctioned monopolies before venture capital existed, whereas today's corporate and VC landscape funds R&D at vastly superior scale. | |
| Evolving Tech Talent Dynamics: Diversity, Productivity, and Art | 6 | 4 | 0 | 1 | The hosts and guest review shifting talent dynamics in computer science. Hennessy celebrates increasing female enrollment and higher mathematical proficiency among incoming freshmen, while Martin reflects on the transition from passion-driven research to professionalized industry careers. |
Statements from this episode (47)
Chokshi: John Hennessy pioneered the microprocessor architecture used in 99% of devices
“But first, we begin with John's own history as a startup founder, based on pioneering the microprocessor architecture used in 99% of devices today.”
Hennessy: Hardware breakthroughs take much longer to show impact than software
“Many times when you find a fundamental breakthrough, its importance may take a really long time to emerge, particularly in the hardware sector. It moves so much slower than software.”
Hennessy: Mobile and IoT growth made RISC processor architecture crucial
“And in this case, with the explosion of the mobile world and Internet of Things, Efficient process architectures became really crucial, and that really changed the world, and that's why our work has had such great impact over time.”
Hennessy: IBM and DEC missed microprocessors by building increasingly complex machines
“They were building machines which were getting increasingly complicated rather than simpler. And they missed the whole importance of the microprocessor and VLSI and how we completely changed the industry.”
Hennessy: Early RISC fragmentation allowed Intel to retain microprocessor market dominance
“Rather than the industry converging on one risk architecture, they converged on three or four. That gave Intel a real lead up because they didn't have to beat anyone. They had to kind of beat these little three or four.”
Hennessy: The Apple iPhone was the true tipping point for RISC dominance
“Yeah, the iPhone was really the taking-off point. Some of the earlier Nokia phones began to use the technology, but then when the iPhone came along... Boom.”
Hennessy: Global RISC processor chip volume exceeds 50 billion units
“Oh, more than that. Probably fifty billion.”
Hennessy: Power is the second largest cost in large data centers
“The fascinating thing people don't realize is that after the cost of the physical servers themselves, The second biggest cost in a large data center is power.”
Hennessy: CISC chip architectures lag far behind in energy efficiency
“So you care about energy efficiency even in these large data centers, and when it comes to that measure, the CISC architectures are far behind.”
Hennessy: Early RISC commercial adoption was driven by gaming and networking
“Yes, that was one of the earliest breakthroughs for the RISC people in the embedded space were games, high-end network switches, places where there was really high-end color printers, where there was really a fair amount of performance demand, but also conside…”
Hennessy: Foundational tech startups must target underdogs seeking a competitive advantage
“Companies are always a little reluctant to take a risk on a startup, particularly with something like a new architecture, which really is a long commitment. So what you had to do was find companies who felt like they needed a leg up. Over the other players in …”
Andreessen: Truly good ideas require forceful persuasion to gain market adoption
“Everybody worries about protecting their idea. But if your idea is actually any good, you're going to have to bludgeon people to adopt it.”
Casado: Selling technology drives greater adoption than giving it away for free
“It's hard to give something away. It's much, much easier to sell it, especially if you want to have impact afterwards.”
Andreessen: Charging enterprise customers higher prices leads to successful software implementations
“The more you charge, the more successful the implementation is. Right, because the more painful it's going to be for them to write it off.”
Hennessy: Leaders facing a crisis should execute tough cuts quickly
“If you have a crisis and you've got to take a tough step, do it quickly, get it over with, and move through. Reset the clock so you can Then charge ahead.”
Hennessy: Stanford lost roughly 28% of its endowment during the 2008 crisis
“About 28% of the endowment vaporized in a six-month period.”
Casado: Enterprise software success often depends on developer popularity, not core technology
“Whether or not a company is going to do well is, Somewhat independent of technology often, and somewhat independent of the approach they take, and it's more like, you know, do they become the popular one that they use?”
Hennessy: Top decisions in complex organizations are always gray and ambiguous
“In a complex organization, all decisions are gray when they get to the top. And so you've got to get comfortable making decisions, making calls in that situation.”
Hennessy: Professors who start companies become better teachers and researchers
“My experience is the faculty members I know at Stanford have gone out and started companies, are better researchers, they're better teachers, they're all around better because they have a wider range of experience.”
Hennessy: The drain of AI faculty to industry will harm long-term innovation
“I think we're in a tricky position right now, especially around the machine learning AI area, where there are lots of faculty who are leaving, and that will hurt the industry in the long term, because that means we're eating the sea corn.”
Hennessy: Commercialization flexibility gives universities a major recruiting edge for faculty
“So you're a young person, you've got multiple faculty offers, you might be interested someday in taking your technology out. Where's the place to come? Well, it's pretty obvious where the place to come is, and that's a big benefit to the university in terms of…”
Hennessy: A true equivalent to Silicon Valley will undoubtedly emerge in China
“So it is going to happen in China. I have no doubt about it.”
Hennessy: Silicon Valley's dominance has widened despite expectations of rival U.S. hubs
“If you had asked me 1520 years ago, will there be another Silicon Valley in the U.S., I would have said yes, for sure. In fact, just the opposite has happened. The Valley's lead has gotten bigger.”
Hennessy: California and San Francisco would economically collapse without the tech sector
“You know, our cities and the state have such dramatic issues, and yet you pull out the high tech sector. I mean, the state and the city of San Francisco will collapse.”
Hennessy: Startups relocate to Silicon Valley due to executive talent scarcity elsewhere
“I remember a startup founded at Mark's alma mater, University of Illinois, and great group of people, they could hire great young engineers because it's one of the best engineering schools in the country, but they couldn't get the kind of middle and upper leve…”
Hennessy: Stanford set zero tuition for families earning under $100,000 a year
“We decide we need a very simple message, right? Your family makes less than a 100,000 dollars a year. Your tuition at Stanford is zero.”
Hennessy: Stanford required full financial aid recipients to work campus jobs
“Even though your tuition is zero, you have to work for the university 10 hours a week during the year and 20 hours a week during the summer and contribute that to your education.”
Hennessy: For-profit higher education fails to deliver real value to students
“The for-profit industry, unfortunately, in the higher education space doesn't deliver a lot of value, so you end up with lots of students who are not able to use their education to get ahead.”
Hennessy: Technology is the sole path to lower education costs
“We've got to figure out how to deliver a high-quality education, not decrease the quality in order to just get the cost down, but hold the quality up while reducing the cost, and the only way I know how to do that is by using technology.”
Andreessen: The economic value of a college degree comes from diploma signaling
“If you take an undergrad who's completed seven out of eight of their semesters, right, so they're three and a half years into their program, and they drop out, you might think that they would get seven eighths of the income in their first job as somebody who d…”
Hennessy: Post-bachelor's education will shift toward skill-based certification models
“We're moving very quickly towards a certification type model where you take a course or a sequence of courses, right? So you go and take the sequence of courses on cryptography and blockchain, and you become an expert on that, and by demonstrating that you've …”
Hennessy: Interdisciplinary education cannot substitute for deep core domain knowledge
“I don't believe that multidisciplinary or interdisciplinary things are a substitute for some deep domain knowledge. I'm a firm believer that you start with deep domain knowledge and then you build on top of that.”
Hennessy: Algorithmic thinking is a fundamental meta-discipline everyone must learn
“It is unique, and it is this meta-discipline, I mean, I think, and it's become the new meta-discipline that everybody needs to learn. Because algorithmic thinking is such a fundamental thing about how the world operates these days.”
Hennessy: Machine learning is driving scientific breakthroughs across biology, chemistry, and astrophysics
“You just see breakthroughs in biology and chemistry, in astrophysics. Coming out of various forms of machine learning. So all of a sudden it becomes this tool that is applicable to a whole range of things and is changing those fields.”
Hennessy: Scientific innovation requires a new breed of interdisciplinary machine learning experts
“And this is a big gap right now because the senior people in the field, it's highly unlikely that most of Most of them are going to take a year or two out and go back and learn a bunch of things about computer science and statistics and machine learning ideas.…”
Hennessy: Machine learning is the ultimate garbage-in, garbage-out technology
“ML is the ultimate garbage in, garbage out technology, because if the data isn't good and properly validated and the learning process isn't You're going to get assumptions and outputs that are ridiculous.”
Hennessy: Most interesting unsupervised learning applications are in natural sciences, not business
“And the space where that works, sort of unsupervised learning, is such a small part of the giant ML space. It's relatively small. And most of its interesting applications are in the natural science world, not in real world applications.”
Casado: Successful startups master messy real-world complexity rather than seeking theoretical elegance
“What you learn about starting a company is it's actually the opposite, which is almost every solution is dealing with the heavy tail of complexity, and it's a bunch of patches and the real world and everything else, and so mentally you've got to go from, I'm g…”
Hennessy: Startups must aggressively eliminate any projects that aren't potential home runs
“And that was good advice about how to think about a research career, but it doesn't work in a company. You've got to get rid of those things that are not the home runs.”
Hennessy: Bell Labs and Xerox missed major economic gains from key inventions
“Bell Labs and AT&T were not the major beneficiaries of the discovery of the transistor. Xerox was not the major beneficiary of the discoverer of modern personal computing, right?”
Hennessy: Universities naturally transfer technology to society via student graduation
“That's why universities are the ideal place to do this kind of work, because society benefits. Universities do technology transfer in a very natural way. It's called graduation.”
Casado: Fundamental research and development is shifting from academia to tech companies
“Some of the most fundamental research contributions are actually happening in industry today, and not only that, that, you know, the academic system has actually moved towards short-termism, especially in incremental publishing.”
Andreessen: Legendary corporate R&D labs required monopolies and a pre-VC era
“They were basically like, they were rounding errors on everything. Like, there weren't 10, there weren't 20, there weren't a hundred, there were three or four. And there were two preconditions for them. One is, they all were offshoots of monopolies... And then…”
Casado: Cisco is uniquely successful at acquiring startups and integrating technology
“They've really, I think, are probably the top companies in making those acquisitions successful and doing spin-ins. I mean, there are very, very few companies you can point in that have been so successful in acquisitions. It's basically a core competency.”
Hennessy: Computer science now attracts top academic talent over biological sciences
“I thought 10 years ago that computer science was going to become second to the biological sciences in terms of getting the best students, and that Everybody, the really best students were going to go do the biologicals, biotech, things like this. Well, that's …”
Andreessen: Software engineers are vastly more productive today than 20 years ago
“I think the big one I see that I think is probably under remarked on is engineers are so much more productive today, especially in software than they were 2030 years ago. The tools are so much more sophisticated and powerful, all the infrastructure technologie…”
Casado: Software engineers are significantly more mercenary today than 20 years ago
“Kind of on the negative spectrum, there's, you know, people are a lot more mercenary about it than they were before. And on the positive end, I do think we have a lot of framing around it. What does it mean to have a workforce in computer science that will com…”