Amazon CEO Andy Jassy on Why Amazon Is Investing $220 Billion in AI

Anthropic

Across the shows: 40 speakers in other episodes, 23 bullish and 42 bearish calls, past 30 days

• Andy Jassy · AmazonBullish

  Anthropic will be one of the most successful AI companies over the long term; Amazon invested in it.

  Jassy says Amazon's investment in Anthropic reflects the strength of the team and its technology and models, its fast top-line growth, and Amazon's belief in how transformative AI will be. He says Anthropic is one of two companies he expects to be among the most successful in AI over a long period, though not every AI company will win. The partnership is deep: many customers run Anthropic on Bedrock and are most successful doing so, the two companies have worked closely on chips for training, and Anthropic made a multiyear, multi-gigawatt commitment to Trainium.

  “we believe these are gonna be two of the most successful companies in the space over a long period of time. ... But we believe these will be two of them.”at 34:44

OpenAI

Across the shows: 20 speakers in other episodes, 9 bullish and 16 bearish calls, past 30 days

• Andy Jassy · AmazonBullish

  OpenAI will be one of the most successful AI companies long term; Amazon invested to bring it onto AWS.

  Jassy says Amazon's investment in OpenAI reflects the strength of the company's team, technology, and applications, and Amazon's belief that OpenAI will be one of the most successful AI companies over the long term. The investment also addressed customer demand: many AWS customers wanted OpenAI but couldn't use it on AWS because of prior business arrangements. Over months of conversations, the companies built a unique product, a stateful runtime environment on Bedrock, which Jassy believes is how all agents will be built, since agents need state.

  “We built a unique product together in this stateful runtime environment, which is how I think all agents are gonna be built moving forward.”at 36:36

Enterprise AI adoption & AI software

• Andy Jassy · Amazon

  Most enterprise applications will use AI inference pervasively within five to seven years; adoption is still early.

  Jassy describes a barbell in AI demand. On one end are AI labs consuming huge amounts of compute and a few breakout apps such as ChatGPT and Claude Code. On the other end, enterprises mostly use AI for cost avoidance and productivity, such as automating customer service, fraud, or business processes. The middle, enterprise production workloads, is where he expects the largest absolute segment of AI usage to come from. Most of those workloads are not yet pervasively using AI, though that is rising each year. Zooming out five to seven years, he expects quite a few, if not most, enterprise applications to use inference pervasively, and he says the growth runway is long.

  “if you just zoom out five to seven years, I think that quite a few, if not most, of the enterprise applications that we use and that exist today are gonna be using inference in a pervasive way.”at 22:34

• Andy Jassy · Amazon

  Slow incumbents will lose ground to AI-native startups and must rewrite their code to be agent-capable.

  Jassy observes that most new startups are being built agentically and are AI-native from the start. Larger established companies will have to work out how to build at the same pace. He predicts some will move too slowly and become more vulnerable to startups. He also predicts that companies will have to rewrite almost all of their code to be agentically capable. That means letting agents update different pieces of code, with built-in constraints, which is part of why firms that prioritise this will be better placed.

  “Some of them will move very slowly to figure it out and they will be more vulnerable, in my opinion, to startups. But I predict that companies will have to rewrite almost all of their code to be agentically capable.”at 41:43

Cybersecurity

• Andy Jassy · Amazon

  AI models expose far more vulnerabilities, making cybersecurity the immediate risk and forcing patching in minutes.

  Jassy argues the immediate, more likely risk from AI is cybersecurity, not speculative scenarios like evil botnets or malicious robots. He says models find far more vulnerabilities than humans previously did. Today's patching is slow: vulnerabilities are found, classified and patched, then most companies take a long time to deploy because of dependencies. In the future, applications must respond to imminent threats in minutes. That requires rewriting applications so agents can update code within set constraints, and companies that don't do this will have no chance in the new environment.

  “the more immediate term risks, which are really around cybersecurity. These models find so many more vulnerabilities than you did before ... the way that applications are gonna have to be able to respond to imminent security threats is gonna be in minutes.”at 42:51

• Ben Horowitz · Andreessen Horowitz (a16z)

  The software industry has poor security architecture, and the reckoning is here, creating opportunity for new companies.

  Horowitz says the industry did not architect security well, from large systems to embedded devices, and that the day of reckoning is here. He argues specialization has been the worst case for security. Security engineers are scarce and often do not work well with other engineers, so most software companies' security practices are weak for writing software. He sees the AI-driven threat as clear and present rather than theoretical, which he calls a big opportunity for new companies and efforts. He adds that tools now exist to help fix this.

  “we, as an industry, did not do a great job of architecting everything from the big systems to embedded devices with good security architecture. And the day of reckoning is here.”at 43:57

US–China AI & chip race

• Andy Jassy · Amazon

  The US cannot afford not to build data centers, or China will set the global AI norms.

  Jassy argues AI is the biggest technology transformation of his time, and the infrastructure build-out it requires is comparable to the US highway and bridge programs of the 1950s. He says the US, and the world more broadly, cannot afford not to build data centers, because otherwise another country like China would establish the norms. He does not want those norms set by China, either domestically or globally. He acknowledges the industry can build better, but argues the build-out is necessary.

  “cannot afford not to build data centers or we will find that another country like China will establish and all the norms there.”at 17:30

AI infrastructure & data centers

• Andy Jassy · Amazon

  Anti-data-center arguments are largely myths; data centers are public-private partners that bring tax revenue to communities.

  Jassy argues many public criticisms of data centers are myths or mistruths. Golf courses and almond farming use far more water than data centers: golf courses use 200 times more water than all Amazon's data centers, and more in two days than Amazon's data centers use in a year. Backup generators are counted at full capacity for permits, but run only about 0.1% of the year, roughly ten hours for maintenance. Energy rates have risen in some areas and fallen in others. Data center operators also pay large taxes: Amazon has paid about $3B in St. Joseph County, Indiana, which local communities can spend. Amazon has committed another $1B over five years to community programs, and he says this should be a public-private partnership.

  “Golf courses consume 200 times more water than all our data centers. And they consume more in two days than we do in a year in our data centers.”at 18:28

Power demand & the grid

• Andy Jassy · Amazon

  US underinvested in grid infrastructure; AI-driven build-out should leave power rates flat or falling after capacity comes online.

  Jassy says energy rates have risen in some data center regions and fallen in others, but the real issue is that the US has not invested in its grid infrastructure as needed. Meaningful load growth was going to require that investment, and AI's rapid growth is what pushed it over the line. Once data center capacity is built, he expects rates to flatline or fall in most regions, as they have in Indiana, Georgia, and Mississippi.

  “some of the energy rates have gone up. Some of the energy rates have gone down. But the real issue is that as a country, we have not invested in the infrastructure of our grids the way we've needed to”at 19:06

AI chips & semiconductors

• Andy Jassy · Amazon

  AI chip, memory, and power supply constraints will persist for several years, not just through 2027.

  Jassy says there is only so much data center and power capacity and only so much memory, and he does not think the memory shortage ends in 2027. Supply constraints will show up in different components for several years. Amazon has to make real trade-offs between allocating memory to HBM for AI accelerators and DRAM for its core business. He says Amazon is probably putting more capacity in than anyone else, but demand is incredibly high, so it will likely still fall short of demand.

  “I wish I could tell you that I thought that the supply constraints would be just a 2027 issue. I think we'll see them in different components for several years.”at 27:15

AI pricing & unit economics

• Andy Jassy · Amazon

  AI inference is still an order of magnitude too expensive; cutting its cost is needed for pervasive enterprise use.

  Inference costs have fallen considerably over the last few years but are still, in Jassy's view, an order of magnitude too high for AI to be used as pervasively as possible. Rewriting enterprise applications that run on CPUs to use AI accelerators is a big undertaking, so the cost gap has to be large enough to justify it. If inference cost drops, that bar falls and applications become better and more capable with AI accelerators and inference. He argues Amazon's chip business is a big part of driving that cost down.

  “The cost of inference has come down quite a bit in the last few years, but it's still an order of magnitude too expensive in my opinion for it to be used as pervasively as possible.”at 32:39

AI & jobs

• Andy Jassy · Amazon

  AI agents will mean fewer humans doing today's work, with smaller, more cross-functional teams, though new jobs will emerge.

  Jassy expects companies to employ fewer people than today as work becomes agentic. He cites coding being done agentically and automation of customer service, business intelligence, research, spreadsheet building and legal work. He expects teams to be much smaller, with people who work across functions because models get them about 80% of the way in many disciplines. As evidence he describes a four-to-five-person team that had an agent-built idea operating in an 80% shippable state by Sunday after an idea was agreed on Thursday. He adds that new jobs will arise, some known and some unknown, including from AI, robotics and space, but not enough to keep headcount at today's level.

  “we're gonna have fewer humans doing the work that they're doing today in the future than we have today. ... I think the teams will be much smaller than they are right now.”at 39:42

Venture capital & startups

• Ben Horowitz · Andreessen Horowitz (a16z)

  Some VCs buying old companies to cut costs with AI show few ideas; building new things is more exciting.

  Horowitz says AI lets firms cut costs, such as reducing accountants or lawyers, and that some venture capital firms now plan to buy old companies and cut their costs. For his firm this has never been interesting, because if that is the whole idea, the firm has very few ideas. He finds building something the world has never seen before much more exciting, and says a third path, rethinking a successful product around a new platform as Amazon did with Alexa, is also interesting.

  “there's actually venture capital firms that now are like, okay, we're going to go buy old companies and cut their costs. And that's never been that interesting to us because we're like, okay, if that's your idea, you have very few ideas.”at 46:29

Consumer AI products

• Andy Jassy · Amazon

  Universities should not fight AI in education; students must learn to use it while keeping critical reasoning.

  Jassy says AI use in education is already unstoppable, so universities would be wise not to fight it. He says his kids used AI to build applications, and that AI will bring everyone quickly to an advanced starting point on tasks like case reading and research. He still believes that putting the pieces together to build strategy, plans or products will mostly require human judgment, ingenuity and experimentation. AI should be used as a tool, but critical reasoning remains central.

  “you can't tell them not to use AI in education. ... And I think universities would be wise not to fight it. I just think it's a tool that people are gonna use.”at 53:27

AI: overall outlook

• Andy Jassy · Amazon

  AI, robotics and space are three of the biggest transformations in our lifetime, and Amazon is in the middle of all three.

  Jassy frames AI, robotics and space development as three of the biggest transformations in his lifetime, possibly ever. He says they will each substantially change many customer experiences. He points to Amazon's robotics in retail, AI tools that let builders create agents and applications, and its low Earth orbit satellite constellation as examples of Amazon operating across all three.

  “we are living in an era of three of the biggest transformations in our lifetime and maybe ever between AI, robotics, and space development.”at 57:06