Amazon unveiled Alexa+ at its devices event in February 2025, positioning the upgrade as its answer to ChatGPT's conversational fluency. The service, powered by generative AI and priced at $19.99 per month (free for Prime members), has been rolling out to Echo Show devices since March. OpenAI, meanwhile, has been steadily upgrading its voice mode, integrating it directly into the ChatGPT interface in late November.
The timing is not coincidental.
The ChatGPT problem
When OpenAI released ChatGPT in late 2022, it exposed an uncomfortable truth about the voice assistants that had colonized millions of homes: they were fundamentally dumb. Siri, Alexa, and Google Assistant could set alarms and play music with reasonable reliability, but ask them to maintain a conversation or synthesize information across topics and they fell apart. Sean Saulsbury, a software entrepreneur in Oceanside, California, told Marketplace that Siri "still struggles with basic requests," noting his frustration when asking it to read an open email only to have it list recent message titles instead.
The gap wasn't subtle. While users could hold sprawling conversations with ChatGPT about everything from code debugging to relationship advice, their smart speakers were stuck parsing rigid command structures. Larry Heck, a professor at Georgia Tech who has worked on voice assistants for Microsoft, Google, and Samsung, noted that modern AI systems like GPT-4o can carry on spoken conversations naturally, but legacy assistants required users to phrase requests in specific ways to get results.
Amazon's response has been aggressive. Alexa+ can now understand context across multi-turn conversations, process what Amazon calls "half-formed thoughts," and tap into services like OpenTable, Ticketmaster, and Uber to complete tasks. At the February announcement, Panos Panay, Amazon's head of devices, demonstrated Alexa+ analyzing a camera feed of an audience and remarking that "those 250 folks look pretty fired up." Whether anyone needs their speaker to commentate on room vibes remains unclear.
OpenAI's hardware ambitions
OpenAI isn't content to remain a software company. The acquisition of Jony Ive's hardware startup io in May 2025 for $6.4 billion signaled a push into physical devices. Ive, who shaped Apple's industrial design for decades, confirmed in November that the resulting product will arrive "in less than two years" and will be pocket-sized and screenless, relying on cameras and microphones to interact with its surroundings.
The leaked ambitions are characteristically Altman-esque in scale. Internal communications reportedly referenced shipping 100 million devices "faster than any company has ever shipped 100 million of something new before." Sam Altman suggested the io acquisition could deliver an additional $1 trillion in value to OpenAI, a company that, by most accounts, isn't profitable.
The actual audio technology, though, has improved substantially. OpenAI's next-generation audio models, announced earlier in 2025, introduced gpt-4o-transcribe with roughly 35% lower word error rates on standard benchmarks compared to previous versions. The Realtime API now allows developers to build voice agents with response times averaging 320 milliseconds, approaching natural conversation speed. The December 2025 model snapshots added an 18.6 percentage point improvement in instruction-following accuracy for real-time agents, according to OpenAI's internal evaluations.
Those are OpenAI's own benchmarks, of course. Independent verification of voice AI accuracy remains sparse.
Apple's uncomfortable silence
Apple has taken a different path, which is to say it has taken almost no path at all. The company announced Apple Intelligence at WWDC 2024 with splashy demos of Siri juggling apps to help plan a lunch after a flight. Those capabilities were supposed to arrive in 2025. In March, Apple delayed them to 2026.
"We've also been working on a more personalized Siri, giving it more awareness of your personal context," an Apple representative said in a statement acknowledging the delay. "It's going to take us longer than we thought to deliver on these features."
The delay came after Apple had already been running advertisements for features that didn't exist. Bloomberg reported in June that Apple is now targeting spring 2026 for the upgraded Siri as part of iOS 26.4. Internal code reviewed by Macworld references the spring 2026 timeframe as well.
Apple's AI team has undergone significant restructuring. John Giannandrea, the AI and machine learning chief hired from Google in 2018, will retire in 2026. The company hired Amar Subramanya, who previously led engineering for Google Gemini, as vice president of AI. The reorganization suggests Apple recognized something wasn't working, though the company maintains its commitment to on-device AI processing for privacy reasons.
Gene Munster of Deepwater Asset Management summarized investor sentiment bluntly: "They basically said that this year, don't bother us about AI, and we'll blow you away by what we show next year."
The market projections
Market research firms are bullish on voice AI, though their estimates vary wildly depending on how they define the category. MarketsandMarkets projects the AI voice generator market will grow from $3 billion in 2024 to $20.4 billion by 2030, a compound annual growth rate of 37.1%. Grand View Research puts the conversational AI market at $11.58 billion in 2024, reaching $41.39 billion by 2030.
The broader voice and speech recognition market, according to MarketsandMarkets, should grow from $9.66 billion in 2025 to $23.11 billion by 2030. That's a 19.1% CAGR, considerably more modest than the voice generator segment.
These projections assume continued adoption, but consumer behavior has proven stubborn. Amazon has reportedly lost billions on its Echo devices business, with users predominantly setting timers and playing music rather than shopping through the platform as the company hoped. Whether an AI upgrade changes that calculus remains the core bet.
What's actually new
For all the marketing language about "natural conversations" and "getting things done," the practical improvements in 2025 are narrower than they appear. Alexa+ can book restaurant reservations through OpenTable and call Ubers. ChatGPT's voice mode now shows responses on screen while you talk, rather than forcing you into a separate interface with an animated blue orb. Amazon launched Alexa.com, a web portal for typed interactions.
These are incremental refinements to user experience, not fundamental breakthroughs. The underlying large language models have improved, enabling more flexible interpretation of requests, but the physics of spoken interaction haven't changed. You still need to get the assistant's attention, wait for processing, and hope it understood your intent.
Consumer Reports tested Alexa+ in December and found it "a much better assistant, but it's far from perfect." The Alexa app, which is supposed to tie everything together, "is still buggy."
The most interesting development may be the competitive pressure itself. Google is reportedly developing Gemini for Home, its own AI overhaul for Google Assistant. Apple's delayed Siri upgrade, if it actually ships, will need to justify the wait. And OpenAI's hardware gambit suggests the company believes voice AI needs its own device category, not just better software running on existing phones and speakers.
The voice AI market in 2026 will look different than it does today. Whether any of these companies can make smart speakers feel genuinely smart, rather than just incrementally less frustrating, is the question none of them have answered yet.
