From 2017 to 2021, Amazon's devices division lost more than $25 billion, according to internal documents reviewed by multiple media outlets. At the center of this financial hemorrhage sat Alexa, the voice assistant that Amazon CEO Jeff Bezos once believed would become the operating system for consumers' lives.
By November 2023, the reckoning arrived. Amazon laid off more than 180 employees from the Alexa division as part of a broader restructuring that would shift hundreds of additional employees to different projects. A dozen current and former employees described to media outlets "a division in crisis," with one former engineer calling Alexa "a colossal failure of imagination" and "a wasted opportunity."
The man tasked with managing this crisis—and then pivoting Amazon's entire AI strategy—is Rohit Prasad, a soft-spoken engineer from Ranchi, India, who joined Amazon in 2013 and spent a decade building Alexa into the world's most widely deployed voice assistant. In August 2023, Amazon CEO Andy Jassy promoted Prasad from Senior Vice President and Head Scientist for Alexa to Senior Vice President and Head Scientist for Artificial General Intelligence, reporting directly to him.
The reorganization marked Amazon's acknowledgment that voice assistants, despite their ubiquity, had failed to become the transformative platform the company envisioned. Instead, generative AI—powered by large language models from OpenAI, Anthropic, and Google—was redefining what consumers expected from AI systems. Prasad's new mandate: catch up to competitors who had stolen a march on Amazon in the most important technology race of the decade.
The stakes could hardly be higher. Microsoft's $13 billion investment in OpenAI now appears worth over $90 billion on paper. Google has integrated its Gemini models across Search, Gmail, and YouTube, defending its core businesses while expanding into new AI applications. Meanwhile, Amazon—despite its $125 billion AI infrastructure spending in 2025—lacks a competitive consumer-facing AI product and depends heavily on partnerships with Anthropic and other external model providers.
Prasad's challenge is threefold: salvage what remains of Alexa's 500+ million installed device base, build credible foundation models to compete with OpenAI and Google, and position Amazon Web Services as the infrastructure layer for the entire AI industry. Whether he can succeed will determine not just Amazon's AI future, but also whether the company can maintain its position among technology's elite.
Rohit Prasad was born in Ranchi, a city in eastern India known more for cricket legend MS Dhoni than for producing Silicon Valley executives. His fascination with technology began early, leading him to pursue a bachelor's degree in Electronics and Communications Engineering at Birla Institute of Technology, Mesra—one of India's respected technical institutions, though not among the elite Indian Institutes of Technology.
Like tens of thousands of Indian engineering graduates, Prasad looked to the United States for advanced education and career opportunities. He enrolled at the Illinois Institute of Technology in Chicago, earning a Master's degree in Electrical Engineering. The program would prove consequential: decades later, Prasad would return to IIT as a distinguished alumnus, delivering keynote addresses on AI's future.
After completing his graduate studies, Prasad joined Raytheon BBN Technologies, a research and development company with deep roots in computer science history. BBN had helped build the ARPANET (the precursor to the internet) and maintained a world-class speech and language research division.
Prasad spent 14 years at BBN, rising to Deputy Manager and Senior Director of the Speech, Language and Multimedia Business Unit. During this period, he developed expertise in automatic speech recognition, natural language processing, and machine learning—precisely the technologies that would power the coming voice assistant revolution.
His academic contributions during this period were substantial. Prasad authored more than 100 scientific papers and earned multiple patents in speech processing and machine learning. This research pedigree would later prove critical when Amazon needed technical leadership capable of advancing Alexa's AI capabilities.
In 2013, Amazon recruited Prasad as Director of Machine Learning. The timing was deliberate. Amazon had been secretly developing a voice-controlled speaker codenamed "Doppler," which would eventually launch as the Amazon Echo in November 2014.
The Echo represented a massive bet by Jeff Bezos that voice interaction would become the next major computing platform after smartphones. Unlike Apple's Siri or Google's voice search, which lived inside existing devices, Amazon was creating dedicated hardware optimized for far-field voice recognition—the ability to hear and respond to commands from across a room.
Prasad's speech recognition expertise made him invaluable to the project. Early Echo devices struggled with accuracy, especially in noisy environments or when music was playing. Prasad's team developed beam-forming microphone arrays, acoustic echo cancellation, and machine learning models that dramatically improved Alexa's ability to understand natural speech.
The product struck a chord with consumers. By 2018, Amazon had sold over 100 million Alexa-enabled devices. The company's first-mover advantage in smart speakers gave it commanding market share: over 70% of the U.S. smart speaker market in early 2018, according to research firm Strategy Analytics.
In Alexa's early years, everything seemed to validate Amazon's voice-first vision. Developers built over 100,000 "skills" (third-party voice applications) for the platform. Major brands like Spotify, Uber, and Domino's integrated with Alexa. Amazon expanded the Echo lineup from the original cylindrical speaker to the Echo Dot (compact and cheap), Echo Show (with a screen), and Echo Auto (for cars).
The strategy appeared sound: sell Echo devices at cost or even a loss, then recoup the investment through increased Amazon purchases driven by voice commerce. Internal projections assumed that Alexa users would naturally evolve into heavy voice shoppers, reordering household staples, discovering new products, and shopping hands-free while cooking or doing chores.
Prasad, who had been promoted to Vice President and Head Scientist for Alexa by 2016, led the technical roadmap. His team focused on improving natural language understanding, adding support for multiple languages, and expanding Alexa's knowledge graph. In 2018, Amazon promoted him again to Senior Vice President, placing him among the company's senior technical leadership.
But the commercial reality diverged sharply from the vision. According to multiple reports citing internal Amazon documents, the most common Alexa interactions were:
Voice commerce—the fundamental business case for Alexa—never materialized at meaningful scale. Users found voice interfaces poorly suited for product discovery and comparison. Speaking credit card numbers aloud felt uncomfortable. Confirming orders without seeing product images proved cumbersome.
A dozen current and former Amazon employees told Business Insider that "just about every plan to monetize Alexa has failed." The company tried advertising (met with user backlash), premium subscriptions for enhanced features (minimal uptake), and partnerships with brands (limited success). None generated revenue remotely approaching the division's costs.
The scale of the failure became clear when internal documents revealed that Amazon's devices division—dominated by Alexa and Echo products—lost more than $25 billion between 2017 and 2021. These losses represented one of the largest failed bets in technology history, exceeding even Google's losses on its moonshot projects.
The economics were brutal. Amazon reportedly sold Echo devices at $10 to $20 below cost, expecting Downstream Impact (DSI)—additional Amazon purchases attributable to Alexa ownership—to compensate. But DSI remained far below projections. Most users treated Echo as an inexpensive music player and kitchen timer, not a shopping platform.
Making matters worse, operating costs kept rising. Maintaining cloud infrastructure for hundreds of millions of Alexa queries daily consumed enormous compute resources. Prasad's team employed thousands of engineers, scientists, and linguists to improve Alexa's capabilities across dozens of countries and languages. The division became Amazon's most expensive R&D project with the least clear path to profitability.
Meanwhile, competitors eroded Alexa's market dominance. Google launched Google Home in 2016, leveraging its superior search and knowledge graph to provide more accurate answers than Alexa. By 2018, Google had captured 30%+ of the smart speaker market, up from zero two years earlier. Amazon's share fell from over 70% to around 62%.
Apple entered with HomePod in 2018, targeting premium users willing to pay for superior audio quality and tight iPhone integration. Though HomePod sales disappointed, Siri remained the most widely used voice assistant due to its presence on hundreds of millions of iPhones and iPads.
More fundamentally, the entire voice assistant category stalled. After the initial novelty wore off, consumers' usage patterns plateaued. Smart speaker sales growth slowed dramatically after 2018. Industry analysts began questioning whether voice interaction would ever become the transformative platform that Amazon, Google, and Apple had bet on.
By 2020-2022, Alexa's capabilities had largely stagnated. Despite Prasad's team's efforts, core problems remained unsolved:
These limitations reflected the underlying technology. Alexa was built on earlier-generation natural language processing systems that relied heavily on pattern matching, intent classification, and hand-crafted knowledge bases. The system lacked the flexible reasoning and broad general knowledge that would soon define large language models.
On November 30, 2022, OpenAI released ChatGPT to the public. Within five days, the chatbot reached 1 million users. Within two months, 100 million. The system demonstrated capabilities that shocked even AI researchers: coherent long-form writing, code generation, creative problem-solving, and fluid multi-turn dialogue.
For Amazon's Alexa division, ChatGPT represented an existential threat. Users immediately noticed that ChatGPT could handle queries that stumped Alexa. Ask Alexa to "explain quantum entanglement in simple terms" and you might get a Wikipedia-style definition. Ask ChatGPT the same question and you'd receive a clear explanation with analogies, follow-up clarifications, and the ability to drill deeper through natural conversation.
The comparison brutalized Alexa in tech media and social networks. Users began connecting ChatGPT to their Echo devices through third-party workarounds, effectively replacing Alexa's brain with OpenAI's model. The message was clear: consumers wanted conversational AI, not pattern-matching voice assistants.
Microsoft, which had invested $1 billion in OpenAI in 2019 and another $2 billion in 2021, moved decisively. In January 2023, the company announced a new "multiyear, multibillion dollar investment" in OpenAI, later reported to be $10 billion. The deal gave Microsoft exclusive access to OpenAI's models for integration into Microsoft's products.
By February 2023, Microsoft had integrated GPT-4 into Bing search and launched Copilot across its Office 365 suite. Suddenly, Microsoft—long dismissed as a has-been in consumer technology—possessed the most advanced AI products in the market. The company's stock surged, adding hundreds of billions in market capitalization as investors bet on an AI-driven productivity revolution.
For Amazon, Microsoft's OpenAI partnership represented a strategic nightmare. Microsoft Azure became the preferred cloud provider for AI startups, as access to OpenAI's models drove customer acquisition. Amazon Web Services, despite its market-leading position in cloud infrastructure, lacked a comparable attraction for AI-focused customers.
Google, caught flat-footed despite having pioneered transformer architectures and founded DeepMind, rushed its own chatbot to market. The February 2023 launch of Bard (later renamed Gemini) stumbled badly when promotional materials contained factual errors. But Google's vast resources and technical talent quickly improved the product, and by mid-2024, Gemini rivaled ChatGPT in capabilities.
Meta took a different approach, releasing its Llama models as open-source alternatives. The strategy built goodwill with researchers and developers while positioning Meta as the counterweight to OpenAI's closed, proprietary approach. By 2024, Meta had invested over $70 billion in AI infrastructure and established a dedicated Superintelligence Lab led by Alexander Wang, recruited from Scale AI.
Against this backdrop, Amazon's AI positioning looked increasingly weak. The company had several scattered initiatives:
None of these efforts cohered into a unified strategy. More problematically, Amazon lacked a competitive consumer AI product that could match ChatGPT's utility or Google's integration across its product ecosystem.
Internal discussions grew heated. Some executives argued that Amazon should focus on infrastructure—providing AWS tools for others to build AI applications—rather than competing directly in foundation models. Others insisted that Amazon needed its own models to avoid dependence on competitors and capture value from the AI revolution.
Andy Jassy, who had succeeded Jeff Bezos as CEO in July 2021, faced the most consequential strategic decision of his tenure. In August 2023, he made his choice.
In August 2023, Andy Jassy announced that Rohit Prasad would transition from leading Alexa to heading a newly created Artificial General Intelligence (AGI) division, reporting directly to the CEO. The reorganization signaled Amazon's recognition that incremental improvements to Alexa would not suffice in an era defined by large language models and generative AI.
Prasad's new team would consolidate Amazon's scattered AI efforts under unified leadership. The AGI division absorbed researchers from Alexa AI, Amazon Science, and various AWS teams. Its mandate: develop Amazon's most capable foundation models and position the company to compete with OpenAI, Anthropic, and Google in the race toward increasingly general AI systems.
The Alexa product organization remained intact but reported to a different executive within Amazon's Devices division. Hundreds of Alexa team members transferred to the AGI division, while others shifted to hardware-focused projects. The message was clear: Amazon's AI future would be built on large language models, not on refining the voice assistant that had consumed $25 billion without producing a viable business model.
Even as Amazon assembled its internal AGI team, Jassy made another consequential bet. In September 2024, Amazon announced a $4 billion investment in Anthropic, the AI safety-focused startup founded by former OpenAI researchers Dario and Daniela Amodei. The deal made Amazon Anthropic's primary cloud provider and gave AWS customers access to Claude, Anthropic's flagship model.
The Anthropic partnership provided Amazon with immediate credibility in generative AI. Claude 3.5 Sonnet, released in mid-2024, matched or exceeded GPT-4's performance on many benchmarks. Anthropic's ARR surged from $1.4 billion to $4.5 billion in 2024, demonstrating strong enterprise demand for alternatives to OpenAI.
For Prasad, the Anthropic relationship created both opportunity and tension. On one hand, AWS Bedrock—the multi-model marketplace offering access to Claude, Llama, and other third-party models—differentiated Amazon from Microsoft's OpenAI-exclusive approach. On the other hand, relying on external models underscored Amazon's weakness in developing its own competitive AI.
Prasad's AGI team embarked on Amazon's most ambitious AI development program to date. Project Olympus, first reported in late 2023, aimed to train a foundation model with two trillion parameters—matching or exceeding the scale of GPT-4 and Google's largest models.
The project required massive computational resources. Amazon deployed thousands of its custom Trainium chips, designed specifically for AI training workloads. The company's willingness to dedicate such resources signaled the priority Jassy placed on competitive foundation models.
But Prasad recognized that single massive models developed over many months couldn't keep pace with competitors' rapid iteration. At Fortune's Brainstorm AI conference in December 2024, he articulated a new approach: "We are now moving away from a waterfall-style process of building one model at a time. Instead, we are focused on creating a 'model factory' designed to release a lot of models at a fast cadence."
This strategy manifested in Amazon Nova, a family of multimodal foundation models announced in late 2024. Nova included multiple variants optimized for different use cases: text generation, image understanding, video analysis, and code generation. The models demonstrated competitive performance while emphasizing price-performance efficiency—a characteristic advantage given AWS's scale and Amazon's custom silicon investments.
In June 2024, Amazon made another strategic move, hiring David Luan, co-founder and CEO of AI startup Adept, along with several other Adept team members. Amazon also licensed Adept's technology and IP.
Luan, who had previously led large-scale AI projects at Google Brain and OpenAI, brought valuable expertise in training large models and building AI agents—systems capable of taking actions on behalf of users rather than merely answering questions. Prasad appointed Luan to oversee "AGI Autonomy," a division focused specifically on developing AI agents that could navigate software interfaces, make decisions, and complete complex tasks.
The Adept deal reflected Prasad's belief that the future of AI lay not in static chatbots but in autonomous agents. "We are now moving from chatbots that just tell you things to agents that can actually do things," he told Fortune in December 2024.
Even as Prasad pivoted to AGI, the Alexa problem remained. Amazon had 500+ million Alexa-enabled devices in consumers' homes—an installed base that competitors would envy. Abandoning Alexa entirely would waste this asset and cede the voice interface market to Google and Apple.
The solution, announced in 2023, was "Alexa Plus" (also referred to internally as "Remarkable Alexa")—a rebuilt version of the voice assistant powered by large language models. Unlike classic Alexa, which relied on rigid intent classification and pre-scripted responses, Alexa Plus would offer ChatGPT-style conversational capabilities: fluid dialogue, contextual understanding, reasoning, and personality.
The plan called for a two-tier model. Classic Alexa would remain free, providing basic functionality for existing users. Alexa Plus would launch as a paid subscription service, priced between $5 and $20 per month according to various media reports. Premium features would include advanced conversational AI, home automation intelligence, personalized recommendations, and seamless integration with other Amazon services.
Internal projections assumed that even a small percentage of Alexa's massive user base would subscribe, potentially generating billions in high-margin recurring revenue. After years of losses, Alexa would finally achieve the sustainable business model that had eluded it since launch.
But Alexa Plus missed its target launch dates repeatedly. Initially planned for late 2023, then rescheduled to summer 2024, then October 2024, the service remained unavailable as 2024 drew to a close. Each delay eroded confidence that Amazon could deliver on its promises.
Multiple technical challenges plagued development. Integrating large language models into voice assistants proved more complex than anticipated:
Prasad's team experimented with various architectures: smaller, faster models for routine queries with escalation to larger models for complex requests; hybrid approaches combining classic Alexa's intent recognition with LLM-powered responses; edge deployment of compressed models to reduce latency and cloud costs.
Progress came slowly. Demos shown to Amazon executives reportedly impressed them with Alexa Plus's conversational abilities but also revealed concerning failure modes—instances where the system misunderstood commands or provided nonsensical responses.
Even more vexing than technical challenges was the business model question. Market research suggested that most Alexa users—accustomed to free service for a decade—would resist paying monthly fees. Amazon Prime members, who already paid $139 annually, especially balked at additional charges.
Internal debates raged over pricing strategy. Some executives argued for including Alexa Plus in Prime membership, absorbing the costs as a Prime benefit to drive retention and differentiation. Others insisted that giving away expensive AI capabilities would repeat Alexa's original mistake of subsidizing an unprofitable service.
Competitive dynamics complicated the decision. Google had not announced plans to charge for Google Assistant upgrades. Apple's Siri improvements would certainly remain free to iPhone users. If Amazon charged for Alexa Plus while competitors offered equivalent capabilities free, customers might simply switch assistants.
The Alexa Plus delays and Prasad's transition to AGI created organizational confusion. While Prasad formally handed day-to-day Alexa management to other executives, he remained involved in strategic decisions about the product's AI transformation. This dual role sometimes created unclear accountability.
The November 2023 layoffs of over 180 Alexa employees further demoralized the division. Engineers who had spent years building Alexa's infrastructure watched as resources and leadership attention shifted to the AGI team. Some departed for competitors; others grew cynical about Amazon's commitment to voice assistants.
In leaked emails reported by tech media, Prasad attempted to rally the team, emphasizing that Alexa remained strategically important and that integrating AGI breakthroughs would revitalize the product. But the message rang hollow to employees who had heard similar promises for years while watching Alexa lose ground to Google and fall further behind ChatGPT's capabilities.
By late 2024, Microsoft's AI advantages had compounded. The company's $13 billion OpenAI investment now appeared prescient, with OpenAI valued at $300 billion in a March 2025 funding round that valued Microsoft's stake at over $90 billion—a 6x return on paper.
More importantly, Microsoft had integrated OpenAI's models across its entire product portfolio:
This product integration created powerful network effects. Enterprises deploying Microsoft 365 Copilot often migrated more workloads to Azure. Developers using GitHub Copilot built applications on Azure. Microsoft was leveraging AI to strengthen its entire ecosystem.
Google, after its stumbling Bard launch, had regained competitive footing. Gemini 1.5 Pro, released in early 2024, offered superior multimodal understanding and a massive 1 million token context window. The model's pricing—just $0.15 per million input tokens compared to $3 for GPT-4—undercut OpenAI dramatically.
More critically, Google integrated Gemini across its core products:
Google's search dominance meant it controlled the highest-intent data flow on the internet. Every query revealed what users wanted to know, providing invaluable training data for improving AI models. Amazon's e-commerce data, while valuable for product recommendations, offered less insight into broad human knowledge and reasoning.
Against these integrated AI strategies, Amazon's approach appeared scattered. The company had investments and products across the AI stack, but they lacked cohesion:
Each product served specific use cases, but they didn't reinforce each other the way Microsoft's Copilot unified its offerings or Google's Gemini integrated across Search, Gmail, and Android. Amazon shoppers didn't encounter AI that made e-commerce dramatically better. AWS customers appreciated Bedrock's flexibility but many still preferred Microsoft's tighter OpenAI integration.
Most glaringly, Amazon lacked a breakthrough consumer AI product. ChatGPT had become synonymous with AI for hundreds of millions of users. Google's Search and Gmail integrations touched billions daily. Microsoft Copilot enhanced productivity for millions of enterprise workers.
Alexa, Amazon's most visible consumer AI product, remained stuck in its pre-ChatGPT paradigm. Despite Prasad's promises, Alexa Plus had not launched. Ordinary consumers could not experience Amazon's AI capabilities the way they could simply visit chat.openai.com or click Gemini in their Gmail interface.
This consumer invisibility created a perception problem. In surveys of AI awareness and usage, OpenAI, Google, and Microsoft dominated mindshare. Amazon barely registered, despite its massive AI infrastructure investments and Anthropic partnership.
At Fortune's Brainstorm AI conference in December 2024, Prasad articulated his vision for Amazon's AI strategy. His central theme: democratizing access to AI capabilities for developers and businesses of all sizes.
"The bar to build with AI has suddenly reduced," Prasad declared. "You don't need a PhD in machine learning or mathematics to build with AI." He predicted that the future workforce would focus on "coming up with prompts, rather than writing the code," with "more and more work at the application layer."
This philosophy aligned with Amazon's historical strength: providing infrastructure and services that allowed others to innovate. Just as AWS had democratized access to computing resources, Amazon's AI tools would democratize access to large language models and AI capabilities.
AWS Bedrock embodied this approach. Rather than forcing customers to use only Amazon's models, Bedrock offered choice: Anthropic's Claude, Meta's Llama, Cohere's models, AI21 Labs' offerings, and Amazon's own Nova and Titan models. Customers could select models based on their specific needs—choosing larger models for complex reasoning or smaller, faster models for high-volume applications.
Prasad's "model factory" concept represented a departure from the approach of OpenAI and Google, which focused on building increasingly large, general-purpose models. Instead, Amazon would produce multiple specialized models optimized for different tasks and price points.
Amazon Nova demonstrated this strategy. The family included:
This portfolio approach allowed customers to optimize cost and performance based on their use cases. Simple customer service chatbots could use Nova Micro at minimal cost. Complex analytical applications could leverage Nova Pro for deeper reasoning.
The strategy also hedged against model commoditization. If any single model became a commodity, Amazon had alternatives. The model factory could rapidly iterate, incorporating new techniques and architectural improvements across the portfolio.
Perhaps most significantly, Prasad emphasized AI agents—systems capable of taking actions, not just answering questions—as the future of AI applications. "We are now moving from chatbots that just tell you things to agents that can actually do things," he told Fortune.
This vision aligned with the Adept acquisition and David Luan's appointment to lead AGI Autonomy. AI agents could navigate software interfaces, complete multi-step tasks, and integrate across systems in ways that simple chatbots could not.
For Amazon, agents offered multiple strategic opportunities:
If Amazon could build superior agent capabilities, it might leapfrog competitors focused on conventional chatbots. The technical challenges were substantial—agents required robust reasoning, error handling, and safety mechanisms beyond what current LLMs provided—but the potential payoff justified the investment.
In his December 2024 Fortune interview, Prasad revealed that Jeff Bezos, despite stepping down as CEO in 2021, remained "very involved" in Amazon's AI efforts. This engagement mattered: Bezos had championed Alexa initially and maintained strong opinions about AI's strategic importance.
Bezos's involvement provided Prasad with top-cover for ambitious, long-term bets. The founder's legendary willingness to sustain losses for years in pursuit of transformative technologies aligned with the AGI division's mission. If anyone at Amazon could authorize the massive capital expenditures required to compete with OpenAI and Google, it was Bezos.
That said, Bezos's engagement also created pressure. The Alexa losses had occurred on his watch, representing one of his rare strategic failures. Prasad's AGI initiative represented Amazon's redemption opportunity—but also risked compounding the failure if it, too, failed to generate commercial returns.
As 2025 progresses, Amazon faces intensifying AI competition across multiple fronts:
Foundation Models: OpenAI's GPT-5, expected in 2025, promises another capability leap. Google's Gemini 2.0 and subsequent versions will leverage Google's massive computational resources. Anthropic, despite its Amazon backing, maintains model development independence. Amazon's Nova models must match or exceed these competitors to win customer mindshare.
Cloud AI Infrastructure: Microsoft Azure and Google Cloud aggressively court AI startups with compute credits, strategic partnerships, and tight model integrations. AWS's market share lead has narrowed as customers choose cloud providers based on AI capabilities, not just infrastructure fundamentals. Prasad's team must ensure that AWS remains the preferred platform for AI workloads.
Consumer Applications: ChatGPT's 200+ million users and Google's billions of Search and Gmail users create massive data flywheels for improving AI models. Amazon's e-commerce platform offers valuable data, but lacks the conversational interactions that train language models. Without a breakthrough consumer AI product, Amazon risks falling further behind in consumer AI mindshare.
Enterprise Adoption: Microsoft's 365 Copilot and Google Workspace's Gemini integration give those companies direct channels to enterprise workers. Amazon lacks comparable productivity software, limiting its ability to reach enterprise users outside of AWS customers. This disadvantage matters as AI adoption becomes a key enterprise purchasing criterion.
When (and if) Alexa Plus finally launches, it will serve as a crucial test of Prasad's ability to deliver consumer-facing AI products. Success requires:
Failure to deliver on any of these dimensions risks undermining Amazon's AI credibility. If Alexa Plus launches to lukewarm reception or another delay occurs, it will fuel narratives that Amazon has permanently fallen behind in consumer AI.
Beyond technology and strategy, Prasad faces organizational and cultural hurdles. Amazon's leadership principles emphasize frugality, customer obsession, and bias for action. These values drove AWS's success but can conflict with frontier AI development, which requires patient, expensive research with uncertain commercial timelines.
The November 2023 Alexa layoffs and subsequent restructurings created anxiety among AI researchers and engineers. Top talent has multiple options in 2025's heated AI labor market. OpenAI, Anthropic, Google DeepMind, and well-funded startups actively recruit from Amazon. Prasad must retain critical expertise while building a culture that can compete with AI-native organizations.
Additionally, Amazon's distributed, team-oriented structure can slow decision-making compared to more centralized competitors. OpenAI's relatively small size (compared to Amazon) allows rapid pivots and tight coordination. Google's AI efforts, while large, benefit from unified technical leadership under Demis Hassabis and Sundar Pichai. Prasad's AGI division must operate with startup-like speed despite being embedded in a 1.5 million person organization.
Prasad has articulated optimism about AI's continued advancement, pushing back against concerns that large language models have "hit a wall" in capability improvements. "Every time we come close to a wall, there's a new dimension," he told Fortune in December 2024.
This confidence reflects his decades of AI research experience. Speech recognition, computer vision, and natural language processing all experienced periodic plateaus followed by algorithmic breakthroughs that enabled continued progress. Prasad likely expects similar dynamics with large language models.
However, the innovation paradox cuts both ways. If AI capabilities continue advancing rapidly, Amazon's current models risk obsolescence before achieving market adoption. The model factory approach partially hedges this risk through rapid iteration, but fundamental architectural breakthroughs could still render Amazon's investments obsolete.
Conversely, if AI progress does plateau, Amazon's compute infrastructure and optimization expertise become more valuable. Competition shifts from model capabilities to deployment efficiency, cost optimization, and application integration—areas where Amazon's operational excellence shines.
Amazon's core strength has always been operational excellence: supply chain logistics, cloud infrastructure, cost optimization at massive scale. AWS succeeded not by inventing cloud computing (others pioneered the concept) but by executing better than anyone else—more reliable, more scalable, more cost-effective.
Prasad's AGI strategy essentially bets that this operational playbook can succeed in AI. Amazon may not have created the transformer architecture, ChatGPT's viral breakthrough, or the safety-focused approach Anthropic champions. But if Amazon can deliver superior price-performance through custom chips, efficient model architectures, and infrastructure optimization, that may suffice.
This strategy has precedent. Amazon didn't invent e-commerce but came to dominate it through superior logistics and customer experience. AWS didn't invent cloud computing but became the market leader through relentless operational improvement. Could the same pattern play out in AI?
The counterargument is that AI's winner-take-most dynamics differ from infrastructure markets. Foundation models exhibit network effects: more users generate more data, enabling better models, attracting more users. If OpenAI and Google's consumer products create such flywheels, Amazon's infrastructure advantages may prove insufficient to overcome the data disadvantage.
Amazon's $125 billion AI infrastructure spending in 2025 represents an unprecedented commitment. The capital goes toward custom AI chips (Trainium, Inferentia), data center buildouts, model training compute, and expanding AWS AI services.
This spending dwarfs the losses Alexa incurred. If Prasad's AGI division fails to generate returns, the financial consequences would far exceed the $25 billion Alexa write-off. Investors who tolerated Alexa losses as an experimental bet may not accept similar outcomes at 5x the scale.
Yet Amazon's core businesses—e-commerce and AWS—generate sufficient cash flow to sustain these investments for years. The company need not achieve profitability from AI quickly, giving Prasad time to develop competitive products and find sustainable business models.
The question is whether time alone suffices. If Microsoft, Google, and OpenAI continue pulling ahead in capabilities and market adoption, Amazon's catching-up timeline may extend beyond investors' patience. The longer Amazon remains behind in visible AI products, the more its competitive position erodes.
For Rohit Prasad personally, the stakes are equally high. He has spent over a decade at Amazon, rising from Director to Senior Vice President with direct CEO reporting. His technical reputation, built over 100+ published papers and successful Alexa technical deployments, is substantial.
The AGI role offers the opportunity to cement his legacy as the architect of Amazon's AI transformation. Success would place him alongside AWS founder Andy Jassy and Amazon device chief David Limp among Amazon's most impactful technical leaders. His influence would extend beyond Amazon, shaping how enterprises adopt and deploy AI at scale.
Failure, conversely, would tie Prasad to two of Amazon's most expensive mistakes: Alexa's $25 billion loss and an unsuccessful AGI initiative consuming even larger resources. The narrative would shift from visionary technical leader to executive who presided over strategic missteps.
This personal dimension likely drives Prasad's intensity. His December 2024 public appearances emphasized optimism and confidence—a leader rallying his organization and the broader AI community around Amazon's vision. Whether that confidence proves warranted will become clear over the next 18-24 months as Amazon Nova, Alexa Plus, and other AGI initiatives face market reality.
Rohit Prasad's journey from Ranchi to the leadership of Amazon's most strategic initiative embodies the opportunities and pressures of AI's transformative era. His technical expertise, honed over decades of speech recognition and machine learning research, positioned him perfectly to build Alexa into a household technology.
But Alexa's commercial failure—losing $25 billion despite 500+ million devices deployed—reveals the gulf between technical achievement and business viability. Building AI systems that users love proved easier than building AI systems that generate sustainable revenue. That lesson hangs over Prasad's AGI mission.
The challenges he faces are formidable:
Yet Prasad also possesses formidable advantages. Amazon's financial resources dwarf most competitors. Its AWS customer base provides built-in distribution for new AI services. Custom silicon investments in Trainium and Inferentia offer long-term cost advantages. The Anthropic partnership provides access to frontier models while Amazon builds its own. And Jeff Bezos's continued engagement signals long-term commitment that can outlast quarterly earnings pressures.
The AI race remains early innings. OpenAI's dramatic November 2022 ChatGPT launch occurred just over two years ago. Google's Gemini, Microsoft's Copilot, and Amazon's Nova emerged even more recently. The ultimate winners in foundation models, AI applications, and AI infrastructure remain far from determined.
Rohit Prasad's ability to navigate Amazon through this uncertainty will shape not just the company's competitive position but the broader structure of the AI industry. If Amazon's infrastructure-first, multi-model approach succeeds, it validates a different path than the vertically integrated strategies of Microsoft-OpenAI or Google. If it fails, the AI industry consolidates further around a handful of vertically integrated players controlling models, applications, and infrastructure.
For now, Prasad continues building, iterating, and evangelizing Amazon's AGI vision. His December 2024 declaration that AI hadn't hit a wall projected confidence that Amazon can still catch up and compete. Whether that confidence proves justified or becomes another expensive lesson in the limits of late-mover strategies will become clear soon enough.
The $25 billion Alexa loss taught Amazon painful lessons about commercializing AI. The question is whether Rohit Prasad and his AGI team learned those lessons well enough to avoid repeating them at an even larger scale—or whether Amazon's AI future will join Alexa as another cautionary tale of technical sophistication without business model viability.
## Continue reading - [100 Most Influential People in AI: 2025 Power List](https://digidai.github.io/2025/11/07/silicon-valley-ai-100-most-influential-2025/) - [Andy Jassy: Amazon CEO](https://digidai.github.io/2025/11/11/andy-jassy-amazon-ceo-aws-ai-deep-analysis/) - [Matt Garman: AWS CEO in AI Infrastructure Race](https://digidai.github.io/2025/11/15/matt-garman-aws-ceo-ai-infrastructure-race-deep-analysis/) - [Sam Altman: OpenAI CEO & AGI Race Leader](https://digidai.github.io/2025/11/08/sam-altman-openai-comprehensive-deep-analysis/) - [Dario Amodei: Anthropic CEO & AI Safety Pioneer](https://digidai.github.io/2025/11/08/dario-amodei-anthropic-comprehensive-deep-analysis/)