💻 technology

Your Smart Speaker Is Dumb Now And So Is AI

Remember when Alexa felt like magic? Now she can't even play the right song. The same disillusionment is hitting AI hard in 2026. After two years of hype, users are discovering that ChatGPT and its rivals are just expensive search engines that confidently lie to your face.

Amazon's Alexa launched in 2014 as a revolution in voice computing. Early adopters marveled at the ability to dim lights, set timers, and ask trivia questions without lifting a finger. By 2018, Amazon had sold over 100 million Alexa devices worldwide. Fast forward to 2026, and the shine has worn completely off. The device that once felt like living in the future now feels like living with a temperamental roommate who mishears everything, plays ads instead of answers, and requires you to scream the same command three times. Amazon's Alexa division has reportedly lost over $25 billion since 2017, according to internal documents reviewed by business reporters. The promised AI upgrade that would make Alexa truly intelligent has been delayed repeatedly, with the latest internal target pushed to late 2026 or early 2027. The AI industry is experiencing an eerily similar reckoning. ChatGPT launched in November 2022 and by January 2023 had hit 100 million users, the fastest consumer app adoption in history. OpenAI, Google, Anthropic, and a dozen startups promised that Large Language Models (LLMs) would transform every industry. Venture capitalists poured over $50 billion into generative AI companies in 2023 and 2024 combined. But now in mid-2026, the cracks are impossible to ignore. Every major AI model has a knowledge cutoff date, typically 12 to 18 months before the current moment. GPT-4's training data ends in April 2023. Claude 3's knowledge stops in August 2023. When you ask about anything more recent, you are not getting intelligence, you are getting a glorified search engine wrapper that scrapes web results and reformats them in confident prose. The hallucination problem has become the industry's dirty secret. A study published in March 2026 by researchers at Stanford University and MIT found that leading LLMs fabricate information in approximately 15-27% of factual queries, depending on the model and domain. Legal AI tools have been caught citing non-existent court cases. Medical AI assistants have recommended dangerous drug combinations. Financial AI has invented stock tickers and merger deals. The confidence with which these systems deliver false information is their most dangerous feature. Unlike a search engine that shows you sources to verify, AI gives you a single authoritative-sounding answer with no easy way to check its work. The promised revolution in productivity has also failed to materialize at scale. Microsoft's Copilot, embedded across Office 365 at $30 per user per month, has seen adoption rates far below projections. Internal Microsoft surveys leaked in April 2026 showed that only 38% of enterprise customers who purchased Copilot licenses reported that their employees used it weekly. The problem is not the technology's capability but its reliability. When an AI assistant is right 80% of the time, that remaining 20% creates more work than it saves, because users must verify every output. Knowledge workers have learned the hard way that trusting AI blindly leads to embarrassing errors in front of clients and colleagues. Meanwhile, the cost structure of AI is becoming unsustainable. Training GPT-4 reportedly cost over $100 million. Training runs for next-generation models are estimated at $500 million to $1 billion each. Running inference (actually answering user queries) costs AI companies approximately $0.01 to $0.05 per interaction, depending on query complexity. OpenAI's ChatGPT is estimated to cost $700,000 per day to operate, according to analysis by SemiAnalysis. With most users on free tiers, the unit economics do not work. OpenAI reportedly burned through $5 billion in 2024 and is on track to lose $6-7 billion in 2026 despite growing revenue. The path to profitability remains unclear, especially as the initial hype fades and users revert to free alternatives like search engines. The industry's response has been to rebrand and pivot. Instead of claiming AI will replace human workers, companies now emphasize AI as a "copilot" or "assistant." Instead of promising artificial general intelligence (AGI) by 2025, timelines have quietly been pushed to 2030 or beyond. Google has scaled back its AI deployment plans after several high-profile failures, including an AI overview feature that told users to put glue on pizza. Meta has stopped releasing overhyped AI product announcements after its AI assistant saw minimal adoption. The gold rush is ending, and what remains is a more sober assessment: AI is a tool with narrow applications, not a replacement for human intelligence.

My Take

The disillusionment with AI is not a failure of technology but a failure of expectations management. The tech industry sold us science fiction and delivered autocomplete on steroids. That does not make AI useless, it makes it overhyped. The same pattern repeats every decade: virtual reality was going to change everything in 2016, then everyone remembered VR headsets make you nauseous. Blockchain was going to revolutionize finance in 2021, then we all realized we were just buying digital beanie babies. Voice assistants were going to replace interfaces in 2017, now Alexa is a kitchen timer that occasionally plays music. The real tragedy is that legitimate AI applications are getting buried under the backlash. Machine learning is genuinely transforming protein folding research, climate modeling, and fraud detection. But those boring, incremental successes do not generate headlines or billion-dollar valuations. We are entering the "trough of disillusionment" phase of the hype cycle, and it is going to be brutal. Expect massive layoffs at AI startups, pivots to "AI infrastructure" instead of consumer products, and a lot of very expensive hardware gathering dust in data centers. The next wave of AI progress will be quieter, more specialized, and far less annoying.

What Happens Next

OpenAI will announce layoffs by September 2026. The company cannot sustain its current burn rate without another massive fundraising round, and investors are now demanding a credible path to profitability. Expect OpenAI to gut its consumer-facing free tier, push hard on enterprise sales, and possibly spin off or shut down unprofitable experimental products. Sam Altman will pivot messaging from "AGI by 2027" to "sustainable AI business model," which is Silicon Valley speak for "we need to stop losing billions." Google will quietly reduce its AI ambitions and refocus on incremental improvements to search rather than revolutionary chatbots. The company learned from its glue-on-pizza disaster that users trust Google for accurate results, and AI's hallucination problem is incompatible with that brand promise. Expect Google to deploy AI in back-end infrastructure (ad targeting, ranking algorithms, spam detection) where mistakes are less visible, rather than customer-facing products where errors go viral. The real winner in this AI winter will be NVIDIA, at least in the short term. Even as optimism about AI applications fades, the infrastructure arms race continues. Every major tech company and nation-state is stockpiling GPUs in case they need to train the next breakthrough model. NVIDIA's H100 and upcoming H200 chips remain sold out through late 2026. But watch for a correction in early 2027 when companies realize they have over-provisioned compute capacity for workloads that never materialized. The chip shortage will flip to a chip glut, and NVIDIA's stock will take a painful haircut. Meanwhile, Amazon will attempt a Hail Mary relaunch of Alexa with genuine LLM integration in late 2026 or early 2027, but it will be too late. Users have moved on, and the voice assistant market has commodified into a feature, not a product. Alexa's best hope is becoming invisible background infrastructure, not the next big thing.

What History Tells Us

The AI hype cycle mirrors the dot-com bubble of 1999-2000 with eerie precision. In both cases, venture capital flooded into companies with no clear business model based on the promise of revolutionary technology. Pets.com, Webvan, and eToys burned billions on customer acquisition before collapsing when investors demanded profitability. The difference is that the internet actually was revolutionary, it just took another decade to figure out sustainable business models. Google, Amazon, and Facebook emerged from the rubble because they solved real problems efficiently. The question for AI is whether the technology is genuinely transformative like the internet, or more like 3D TV, a technical achievement that nobody actually wanted. The Alexa trajectory specifically echoes Microsoft's Zune music player (2006-2011) and Google Glass (2013-2015). In both cases, major tech companies launched products that generated enormous hype, achieved respectable initial sales, then slowly died as consumers realized the products did not actually improve their lives. Zune was a perfectly competent MP3 player that could not compete with the iPod's ecosystem. Google Glass was impressive technology that made you look like a cyborg and raised privacy concerns. Alexa is a voice assistant that works well enough for timers and weather but not well enough to justify the counter space. The pattern is clear: when technology solves a problem nobody has, no amount of marketing can save it.

Market Impact

NVIDIA (NVDA) closed at $128.45 on May 30, 2026, riding high on AI infrastructure demand. Short-term (3-6 months), the stock remains bullish as data center buildouts continue regardless of application-layer doubts. Long-term (12-18 months), expect a 20-30% correction as the AI infrastructure bubble pops and customers cancel or delay orders. Semiconductor equipment makers like ASML will follow NVIDIA down with a lag. Alphabet (GOOGL) at $178.32 is arguably the safest play in AI chaos. The company can afford to lose billions on experimental AI because search advertising prints money. Google's strategy of defensive AI investment (spend to avoid disruption, not to generate returns) positions it well for an AI winter. Slight bullish bias as the market rewards profitability over moonshots. Microsoft (MSFT) at $425.67 is overexposed to AI hype through its OpenAI partnership and Copilot push. If enterprise adoption of AI assistants continues to disappoint, Microsoft will face pressure to write down its $13 billion OpenAI investment. Bearish short-term as corporate customers cut AI budgets. Amazon (AMZN) at $186.92 benefits from AWS infrastructure sales but Alexa remains a boat anchor. Neutral to slight bearish. The real carnage will be in private markets. Expect 40-60% down rounds for AI startups that raised at absurd valuations in 2023-2024. Companies like Anthropic (last valued at $18.4 billion), Inflection AI (acquired by Microsoft for parts), and Character.AI (last valued at $1 billion) will either get acquired for far less than their peak valuations or shut down entirely. Public market investors should avoid AI-focused ETFs like BOTZ or THNQ until the washout completes.