Victor Dey – Observer https://observer.com News, data and insight about the powerful forces that shape the world. Thu, 08 Jan 2026 17:48:11 +0000 en-US hourly 1 https://wordpress.org/?v=6.8.3 168679389 As Chinese Tech Retreats From CES, Lenovo Claims Center Stage at the Vegas Sphere https://observer.com/2026/01/lenovo-ces-2026-sphere-tech-world/ Thu, 08 Jan 2026 00:49:43 +0000 https://observer.com/?p=1609348 The Sphere screen

Since the abrupt halt brought on by Covid-19, most Chinese companies that once dominated CES have retreated amid extended travel restrictions and rising geopolitical tensions. But Lenovo, the world’s largest PC maker, stands out as a notable exception, charging ahead in the U.S. market even as the presence of many of its Chinese peers has faded. This year, Lenovo claimed center stage at CES 2026, hosting its annual product launch event, Tech World, at Las Vegas’s coolest venue: the Sphere.

The two-hour event yesterday (Jan. 6) held the audience’s attention throughout. Lasers cut through the darkness as 16K video washed across the Sphere’s vast, curved screen, wrapping the audience in light and sound. The spectacle doubled as a demonstration of Lenovo’s partnership with Sphere Studios, which produces content for the venue. Behind the scenes, hundreds of Lenovo workstations, servers and services powered the ultra-high-resolution visuals, enabling real-time rendering for immersive live shows and cinematic-scale production.”

Unlike many of its peers, Lenovo loves co-marketing with other major tech brands, often inviting their senior executives to share the stage. Yesterday, Lenovo chairman and CEO Yuanqing Yang was joined by Nvidia’s Jensen Huang, AMD CEO Lisa Su and Intel CEO Lip-Bu Tan. Huang and Su had delivered their own CES keynotes earlier in the week.

Throughout the presentation, Lenovo unveiled a broad slate of products and platforms. The announcements centered on what Lenovo calls Hybrid A.I., led by Qira, a cross-device personal A.I. “super agent,” alongside a full stack of A.I. platforms and services designed for both consumers and enterprises.

On the hardware side, the company introduced new A.I. PCs across its Yoga, IdeaPad, ThinkPad and ThinkCentre Aura Edition lines, along with new Motorola flagship smartphones, including a FIFA World Cup 2026 special-edition Razr. Lenovo also showcased several rollable and wearable concept devices, as well as new ThinkSystem and ThinkEdge servers and an A.I. Cloud Gigafactory developed with Nvidia for large-scale A.I. infrastructure deployments.

the Sphere
Lenovo Group

The Sphere

Three people standing on a stage

Lenovo’s brand push in the U.S.

In recent years, Tech World has evolved into one of the industry’s more closely watched conferences, distinct from traditional product launches in both scope and ambition. Lenovo has used the event not only to introduce new devices but to frame its view of where its industry is headed.

Beyond technology, Lenovo has increasingly aligned itself with high-profile sports brands to strengthen its foothold in the U.S. market. It is an official partner of the FIFA World Cup 2026 in North America and a major sponsor of Formula 1. During yesterday’s event, Lenovo hosted a ticket raffle offering attendees a chance to win an all-inclusive ticket to FIFA World Cup quarterfinal matches (whose prices are at a historical high). FIFA President Gianni Infantino also appeared onstage during the presentation.

Lenovo sells personal computers, servers, monitors and smartphones, and Yang has long framed the U.S. as an important market. “We want to be a solid no.3 in North America,” he told Reuters in a 2021 interview. Today, Lenovo trails HP and Dell in the U.S. PC market. American consumers account for less than 20 percent of Lenovo’s total revenue, Yang told Reuters in August.

In the years leading up to 2020, Chinese companies were a dominant force at CES, often accounting for a third or more of all exhibitors. At CES 2018, more than 1,500 Chinese firms attended the show. By 2023, the first year after China fully reopened, that number had fallen to fewer than 500. While participation has rebounded since then, it remains below pre-Covid levels, and several major Chinese technology companies, including Huawei, DJI and Alibaba, were notably absent in recent years.

]]> 1609348
 Lisa Su Shows Off AMD’s High-End Chips Designed for A.I.’s ‘Yotta-Scale’ Future https://observer.com/2026/01/amd-ceo-lisa-su-lay-out-ai-future-ces/ Tue, 06 Jan 2026 18:03:19 +0000 https://observer.com/?p=1609278

At CES 2026, AMD CEO Lisa Su used the industry’s biggest stage to outline where the next era of A.I. is headed. The A.I. industry, she said during her keynote yesterday (Jan. 5), is entering the era of “yotta-scale computing,” driven by unprecedented growth in both training and inference. The constraint, Su argued, is no longer the model itself but the computational foundation beneath it.

“Since the launch of ChatGPT a few years ago, we’ve gone from about a million people using A.I. to more than a billion active users,” Su said. “We see A.I. adoption growing to over five billion active users as it becomes indispensable to every part of our lives, just like the cell phone and the internet today.”

Global A.I. compute capacity, she noted, is now on a path from zettaflops toward yottaflops within the next five years. A yottaflop is 1 followed by 24 zeros. “Ten yottaflops is 10,000 times more computing power than we had in 2022. There has never been anything like this in the history of computing, because there has never been a technology like A.I.,” Su said.

Yet Su cautioned that the industry still lacks the computing power required to support what A.I. will ultimately enable. AMD’s response, she said, is to build the foundation end-to-end—positioning the company as an architect of the next A.I. phase rather than a supplier of isolated components.

That strategy centers on Helios, a rack-scale data center platform designed for trillion-parameter A.I. training and large-scale inference. A single Helios rack delivers up to three A.I. exaflops, integrating Instinct MI455X accelerators, EPYC “Venice” CPUs, Pensando networking and the ROCm software ecosystem. The emphasis is on durability at scale, with systems built to grow alongside A.I. workloads rather than locking customers into closed, short-lived architectures.

AMD also previewed the Instinct MI500 Series, slated for launch in 2027. Built on next-generation CDNA 6 architecture, the roadmap targets up to a thousandfold increase in A.I. performance compared with the MI300X GPUs introduced in 2023.

Su stressed that yotta-scale computing will not be confined to data centers. A.I., she said, is becoming a local, everyday experience for billions of users. AMD announced an expansion of its on-device A.I. push with Ryzen AI Max+ platforms, capable of supporting models with up to 128 billion parameters using unified memory.

Beyond commercial products, Su tied AMD’s roadmap to public-sector priorities. Joined on stage by Michael Kratsios, President Trump’s science and technology advisor, who is slated to speak at CES later this week, she discussed the U.S. government’s Genesis Mission, a public-private initiative aimed at strengthening national A.I. leadership. As part of that effort, AMD-powered supercomputers Lux and Discovery are coming online at Oak Ridge National Laboratory, reinforcing the company’s role in scientific discovery and national infrastructure.

The keynote closed with a $150 million commitment to A.I. education, aligned with the U.S. A.I. Literacy Pledge—signaling that, in AMD’s view, sustaining yotta-scale ambition will depend as much on talent development as on silicon.

]]> 1609278
 Built by and for People With Paralysis, This ALS Tech Gives A.I. a Human Voice https://observer.com/2025/12/built-by-and-for-people-with-paralysis-this-als-tech-gives-a-i-a-human-voice/ Mon, 15 Dec 2025 21:24:18 +0000 https://observer.com/?p=1605963

When British roboticist Dr. Peter Scott-Morgan was diagnosed with ALS in 2017, he was told the disease would gradually take his voice, his movement and, eventually, his place in the world. But he refused to accept the idea that losing speech should mean losing identity. As his body weakened, the Scott-Morgan turned to technology, experimenting with voice synthesis, gaze interfaces and avatar-based communication. His public transformation earned him the description of the world’s first “human cyborg,” but the label masked a deeper ambition: to redefine how disability and technologies like A.I. can evolve together.

Following his death in 2022, the Scott-Morgan Foundation (SMF) carried forward his mission. The organization began translating Scott-Morgan’s philosophy of dignity-by-design into real-world technology. One of those efforts took shape through Bernard Muller, the Foundation’s chief technologist, who is fully paralyzed by ALS. Muller began architecting and co-developing what would become VoXAI.

“I built VoxAI letter by letter with my eyes. It’s slow, it’s stubborn work, but when your need is real, you just keep going,” Muller told Observer, responding through the VoXAI system. “I used A.I. agents as my ‘extra hands,’ breaking tasks into small steps, testing, refining and letting automation do what my body no longer can. Earlier tools were less intelligent and basically limited to typing letters—useful, but hardly empowering.” 

ALS currently affects tens of thousands of people in the U.S. and hundreds of thousands worldwide. As the condition progresses, up to 95 percent of patients eventually lose the ability to communicate through natural speech. Available solutions remain expensive and imperfect. High-end Augmentative and Alternative Communication (AAC) devices—especially those requiring specialized hardware like eye tracking—often cost between $10,000 and $15,000. For decades, that barrier has left millions effectively voiceless, relying on systems that flatten emotion and erase identity.

VoXAI was unveiled last week at the AI Summit New York. It’s a product of collaboration among Israeli A.I. startup D-ID, voice A.I. company ElevenLabs, Irisbond, Lenovo, Nvidia and several academic partners. D-ID’s real-time avatar engine animates facial expressions, micro-emotions and natural mouth movements; Irisbond’s hardware enables precise eye-tracking control; ElevenLabs’ voice synthesis recreates the user’s pre-illness voice; Nvidia GPUs provide the real-time A.I. performance required for near-zero latency; and Lenovo supplies the robust hardware environment that keeps the system stable and accessible.

Founded in 2017 in Israel, D-ID initially gained recognition for its privacy technology and became a pioneer in generative A.I. video in 2019. Its systems now power digital presenters, learning companions and interactive avatars for Fortune 500 companies and public institutions.

“When it comes to disability, the biggest blind spot is assuming it is too small or niche to matter commercially,” Gil Perry, co-founder and CEO of D-ID, told Observer. “We believe that expressive, real-time digital presence is becoming a new layer of communication infrastructure, and accessibility is where that value is most clear and most urgent.”

“For some people, expressive presence is a benefit; for others, it’s a lifeline,” Perry added. “Health care and assistive-tech providers needed a dependable expressive avatar layer that could plug into their systems and make communication feel truly human for those it matters most to.”

A screen of text prompts Leah Stavenhagen and her VoXAI avatar.

At its core, VoXAI is built on a simple yet transformative idea: assistive technology should not merely generate words on a user’s behalf; it should help them express themselves.

Leah Stavenhagen, an ALS advocate and early VoXAI trial participant, said the hardest part of losing her speech was the invisibility that came with it. She recently began using the tool as a beta tester and demonstrated her digital avatar onstage during the platform’s public debut at the AI Summit.

“When communication takes 30 seconds to several minutes for every response, conversations don’t wait. By the time you’ve composed your thoughts, the topic has already moved on,” she told Observer, responding through the VoXAI system. “People stop asking complex questions and start speaking about ‘you’ instead of ‘to you.’”

To communicate, users interact with a screen mounted in front of them via an eye-tracking device. As conversation unfolds around them, a microphone captures what others are saying, and the A.I. rapidly synthesizes three possible responses. The user selects one simply by moving their eyes. Once chosen, the avatar—displayed on a screen above the user or on a connected device—instantly delivers the response in the user’s own voice, complete with facial nuance and emotional expression. The avatar continues to learn through ongoing interaction, absorbing preferences, social cues and personal history. The goal is to preserve a continuity of self that many people lose as their illness progresses.

“The first time someone sees their avatar or hears their voice, there’s usually a moment of recognition because you’re giving back something lost,” LaVonne Roberts, CEO of the Scott-Morgan Foundation, told Observer. 

Assistive communication technology has long been defined by prohibitive pricing. SMF is attempting to upend that model by offering VoXAI free at the basic tier, with advanced features available for a $30 monthly subscription.

“Identity preservation will become its own category. Voice cloning exists now, but we’re moving toward comprehensive digital identities where your voice, your expressions, your communication patterns are preserved and protected,” said Roberts. “Ambient A.I. systems that listen and respond to context without explicit commands will transform caregiving, elder care, and offer more independence for people living with mobility limitations.”

]]> 1605963
 Google’s Gemini Rapidly Closes In on ChatGPT as Antitrust Scrutiny Mounts https://observer.com/2025/12/google-gemini-outgrow-chatgpt-antitrust-probe/ Wed, 10 Dec 2025 22:10:55 +0000 https://observer.com/?p=1605024

Google’s Gemini is in a hyper-growth phase and rapidly closing the gap with OpenAI’s ChatGPT. Between August and November, Gemini’s global monthly active users rose about 30 percent, roughly six times ChatGPT’s growth, according to web and mobile app traffic data from Sensor Tower, first reported by The Information. The surge suggests that Google has reached an inflection point following the release of Gemini 3, its latest A.I. model, which outperforms GPT-5 on several key metrics.

Gemini is gaining ground across both web and mobile. Global web traffic to Gemini doubled during the August–November period, compared with a 1 percent increase for ChatGPT, Sensor Tower found. Mobile downloads of Gemini rose at about twice the rate of ChatGPT during that period.

User engagement is also on the rise: Gemini users now spend around 11 minutes per day in the app, more than double the time logged in March. Sensor Tower estimates that twice as many U.S. Android users access Gemini through the operating system itself rather than the standalone app—a built-in distribution channel that ChatGPT lacks, despite its integration into Apple’s iOS.

ChatGPT, however, remains the most widely used chatbot. Sam Altman said in October that ChatGPT had more than 800 million weekly users, a figure that has since grown to nearly 900 million, according to The Information, which cited a source familiar with the company’s internal data.

Google’s Gemini 3 Pro model outperforms ChatGPT 5 Pro and the 5.1 “thinking” model in reasoning, multimodal comprehension and multiple benchmark tests. On Humanity’s Last Exam—a 2,500-question assessment spanning math, science, logic and history—Gemini 3 Pro scored 37.5 percent accuracy versus ChatGPT’s 26.5 percent. Third-party evaluations echo the trend: Vellum’s agentic A.I. leaderboard ranks Gemini 3 Pro ahead of GPT-5.1 in reasoning and high-school math, based on GPQA Diamond and AIME 2025 benchmark results.

But as Gemini rapidly expands its footprint, regulators are pressing Google on whether its advantage stems from technological innovation or from market power. Authorities are examining whether Google unfairly diverts traffic away from publishers and limits competitors’ access to comparable data.

Yesterday (Dec. 9), the European Commission opened a formal antitrust investigation into whether Google’s AI Overviews and AI Mode misuse publisher content without compensation, thereby bolstering the company’s market dominance.

The inquiry follows a July complaint from the Independent Publishers Alliance, which argues Google’s A.I. summaries siphon revenue from news outlets. The EU probe fits into a broader pattern of scrutiny under the Digital Markets Act, including earlier penalties tied to Android and Google’s advertising practices.

]]> 1605024
 Google’s New A.I. Chip Is Shaking Nvidia’s Dominance: What to Know https://observer.com/2025/12/google-ai-chip-tpu-nvidia-challenge/ Fri, 05 Dec 2025 22:29:04 +0000 https://observer.com/?p=1604111

Last week, The Information reported that Meta is in talks to buy billions of dollars’ worth of Google’s A.I. chips starting in 2027. The report sent Nvidia’s stock sliding as investors worried the company’s decade-long dominance in A.I. computing hardware now faces a serious challenger.

Google officially launched its Ironwood TPU in early November. A TPU, or tensor processing unit, is an application-specific integrated circuit (ASIC) optimized for the kinds of math deep-learning models use. Unlike CPUs that handle everyday computing tasks or GPUs that process graphics and now power machine learning, TPUs are purpose-built to run A.I. systems efficiently.

Ironwood’s debut reflects a broader industry shift: workloads are shifting from massive, capital-intensive training runs to cost-sensitive, high-volume inference tasks, underpinning everything from chatbots to agentic systems. That transition is reshaping the economics of A.I., favoring hardware like Ironwood that’s designed for responsiveness and efficiency rather than brute-force training.

The TPU ecosystem is gaining momentum, although real-world adoption remains limited. Korean semiconductor giants Samsung and SK Hynix are reportedly expanding their roles as component manufacturers and packaging partners for Google’s chips. In October, Anthropic announced plans to access up to one million TPUs from Google Cloud (not buying them, but effectively renting them) in 2026 to train and run future generations of its Claude models. The company will deploy them internally as part of its diversified compute strategy alongside Amazon’s Trainium custom ASICs and Nvidia GPUs.

Analysts describe this moment as Google’s “A.I. comeback.” “Nvidia is unable to satisfy the A.I. demand, and alternatives from hyperscalers like Google and semiconductor companies like AMD are viable in terms of cloud services or local A.I. infrastructure. It is simply customers finding ways to achieve their A.I. ambitions and avoiding vendor lock-in,” Alvin Nguyen, a senior Forrester analyst specializing in semiconductor research, told Observer.

These shifts illustrate a broader push across Big Tech to reduce reliance on Nvidia, whose GPU prices and limited availability have strained cloud providers and A.I. labs. Nvidia still supplies Google with Blackwell Ultra GPUs—such as the GB300—for its cloud and data center workloads, but Ironwood now offers one of the first credible paths to greater independence.

Google began developing TPUs in 2013 to handle growing A.I. workloads inside data centers more efficiently than GPUs. The first chips went live internally in 2015 for inference tasks before expanding to training with TPU v2 in 2017.

Ironwood now powers Google’s Gemini 3 model, which sits at the top of benchmark leaderboards in multimodal reasoning, text generation and image editing. On X, Salesforce CEO Marc Benioff called Gemini 3’s leap “insane,” while OpenAI CEO Sam Altman said it “looks like a great model.” Nvidia also praised Google’s progress, noting it was “delighted by Google’s success” and would continue supplying chips to the company, though it added that its own GPUs still offer “greater performance, versatility and fungibility than ASICs” like those made by Google.

Nvidia’s dominance under pressure

Nvidia still controls more than 90 percent of the A.I. chip market, but the pressure is mounting. Nguyen said Nvidia will likely lead the next phase of competition in the near term, but long-term leadership is likely to be more distributed.

“Nvidia has ‘golden handcuffs’: they are the face of A.I., but they are being forced to keep pushing state-of-the-art in terms of performance,” he said. “Semiconductor processes need to keep improving, software advances need to keep happening, etc. This keeps them delivering high-margin products, and they will be pressured to abandon less profitable products/markets. This will give competitors the ability to grow their shares in the abandoned spaces.”

Meanwhile, AMD continues to gain ground. The company is already well positioned for inference workloads, updates its hardware on the same annual cadence as Nvidia, and delivers performance that is on par with or slightly superior to equivalent Nvidia products. Google’s newest A.I. chips also claim performance and scale advantages over Nvidia’s current hardware, though slower release cycles could shift the balance over time.

Google may not dethrone Nvidia anytime soon, but it has forced the industry to imagine a more pluralistic future—one where a vertically integrated TPU–Gemini stack competes head-to-head with the GPU-driven ecosystem that has defined the past decade.

]]> 1604111
 Japan’s Bold Bid to Lead the Next Era of A.I. https://observer.com/2025/11/japan-ai-leadership-government-private-sector/ Fri, 28 Nov 2025 18:21:52 +0000 https://observer.com/?p=1602807

When OpenAI opened its first Asian office in Tokyo in April 2024, the company highlighted Japan’s strengths: a deep engineering talent pool, a corporate sector known for precision, and a government eager to advance A.I. Last month, Anthropic followed suit, launching its own Tokyo office. The company, best known for its Claude models, also signed a Memorandum of Cooperation with the Japan AI Safety Institute—making Japan only the second country after the U.S. where Anthropic works directly with national regulators on responsible A.I. development.

Rather than treating Japan as a simple export market for Western models, both OpenAI and Anthropic now view it as a strategic co-development hub. Their products are quickly embedding into Japanese society and industry.

Major companies such as Daikin, Toyota Connected and Rakuten have adopted ChatGPT to speed up data analysis, automate workflows and build custom assistants tailored to Japan’s business culture. Claude, meanwhile, is now fully localized for Japanese users, with adjustments for cultural nuance, linguistic complexity and local compliance rules. Panasonic, NRI and Rakuten have expanded their use of Claude for strategy, creative ideation and secure enterprise deployments.

These corporate moves align with a government pushing one of the world’s most ambitious A.I. agendas. Policymakers are knitting together domestic regulation, international partnerships, workforce initiatives and sovereign digital infrastructure to ensure A.I. becomes a catalyst for economic revival. Independent analyses estimate the technology could raise Japan’s GDP by as much as 16 percent.

In May, Japan passed the A.I. Promotion Act—a law that frames the technology as a national priority requiring structured oversight and rapid adoption. At the center of the effort is the A.I. Strategic Headquarters, a body led by Prime Minister Sanae Takaichi. Japan has also deepened its technology ties with India, agreeing at the G20 Summit this past weekend to broaden cooperation on A.I., critical tech, digital public infrastructure, semiconductors and cybersecurity.

Japanese LLMs challenge GPT and Claude

Most of today’s large language models are trained primarily on English text and struggle to serve non-English markets at scale. That gap has opened space for language-specific models designed for local accuracy and privacy. One of the most prominent is Tsuzumi 2, released last month by Japanese telecom giant NTT Inc.

“Frontier A.I. companies will never provide deeply localized, private Japanese-language models as part of their global roadmap. Tsuzumi 2 fills that gap,” Jan Wupperman, senior vice president of service assurance, data and A.I. at NTT, told Observer.

Tsuzumi 2 is also far more efficient than its Western counterparts. It runs inference on a single GPU rather than dozens. The 30-billion-parameter version operates on a laptop-grade GPU, while the 7-billion-parameter model can run with no GPU at all, Wupperman said. It performs on par with—and sometimes better than—models several times its size, including GPT-5 and Claude 3.5, for Japanese-language reasoning. The A.I. is multimodal, able to process text, images and voice in one workflow.

“We don’t aim to compete with GPT-5. Our philosophy with Tsuzumi 2 is to create small, task-optimized models trained across generic knowledge, industry knowledge and client-specific knowledge,” Wupperman said.

Merging quantum computing with A.I.

The next bottleneck for global A.I. growth is raw computing power. Data centers are straining electrical grids, GPU wait times have stretched to months, and silicon-based chips are hitting physical limits. Japan believes the solution lies in merging quantum computing with A.I.

Together with OptQC, NTT is developing optical quantum systems that operate at room temperature, thereby avoiding the massive cooling systems required by traditional quantum machines. This hardware aims to replace electrons with light, dramatically improving speed and energy efficiency.

“One of the greatest challenges in today’s quantum landscape is energy intensity—cooling, stability and thermal dissipation. Photonics gives us an architectural advantage: light generates almost no heat, enabling quantum processing at room temperature,” Wupperman explained. “This makes photonic quantum systems dramatically more compatible with large-scale A.I. workloads.”

These systems aim to accelerate molecular simulation, climate modeling, high-dimensional optimization and A.I. training tasks that remain out of reach for classical machines.

“Once quantum computing capacity reaches maturity, training an A.I. model will look entirely different. Instead of incremental improvements, this could shrink the training cycle of complex models from months to hours,” said Wupperman.

In the near term, he added, A.I. will continue to advance faster than quantum. But over the next five to ten years, that relationship is expected to flip, with quantum becoming a force multiplier for A.I.—and A.I. helping accelerate quantum hardware design in return.

]]> 1602807
 Anthropic’s Claude Expands Into Industrial A.I. Through Major Partnership https://observer.com/2025/11/anthropics-claude-expands-into-industrial-ai-ifs-partnership/ Fri, 14 Nov 2025 20:31:16 +0000 https://observer.com/?p=1600626
Dario Amodei, co-founder and CEO of Anthropic

Anthropic, long known in Silicon Valley for its high-performance A.I. models and API ecosystem used across tech, finance and major consumer brands, is now pushing into the industrial sector. The company this week announced a partnership with IFS Nexus Black—the innovation arm of enterprise software giant IFS, whose customers include Lockheed Martin, Exelon and Quanta Services—and unveiled Resolve, an industrial A.I. platform powered by Claude.

Anthropic already works with household names like BMW, L’Oréal, Sanofi and Panasonic. IFS is Anthropic’s first major customer in heavy industry, where split-second calls can halt production lines or shape how quickly field engineers respond to climate-driven disasters.

“The true test of A.I. is how it performs when the stakes are high,” IFS Next Black CEO Kriti Sharma told Observer. “The partnership allows us to bring advanced models into the physical world responsibly and at scale.”

Industrial A.I. bridges digital intelligence and real-world machinery. It predicts equipment failures, optimizes complex processes and reduces dangerous or repetitive work. Unlike consumer A.I. assistants, industrial systems must handle chaotic environments, inconsistent conditions and nonstop operational data.

“It’s about applying A.I. to environments surrounded by sensors and machinery, with people on the ground making high-stakes decisions every minute,” Sharma said. “Industrial A.I. can listen to a turbine and warn of a fault before it happens or ‘see’ subtle changes in a pipeline that would take a human hours to detect. It connects planners, technicians and assets in real time to improve yield, reduce costs and keep frontline operations running safely.”

Garvan Doyle, applied A.I. lead at Anthropic, said the platform also aims to demonstrate responsible A.I. in practice. “It’s about proving that frontier models (advanced, large-scale A.I. systems) can operate effectively where critical infrastructure is on the line,” he told Observer. Claude’s multimodal analysis and ability to synthesize disparate information “is a natural fit for industrial environments and what frontline workers need: an A.I. that can surface insights humans might miss.”

Resolve uses Claude to interpret video feeds, audio from rattling machinery, thermal or pressure readings and even technical diagrams. Workers can speak directly to Resolve, which transcribes notes, connects to documentation and creates an automatic decision trail.

The system’s goal is to reduce busywork, capture institutional knowledge and streamline the exchange between workers and the A.I. tools guiding them.

“Claude is trained to be honest about uncertainty, avoid confabulation and reason carefully through complex problems,” said Doyle. In industrial contexts, that means technicians can see why Claude surfaces a potential fault or recommends a repair and verify it against their own expertise. “Keeping humans in the loop is key, and it’s especially impactful when A.I. works as a force multiplier for skilled workers,” he added.

IFS says its customer, William Grant & Sons, the distiller behind Grant’s whisky and Hendrick’s gin, long struggled with fragmented data that forced 38 percent of repairs to be emergency fixes. With an early deployment of Resolve, IFS estimates the distillery will save roughly $11.05 million annually once the new workflows scale.

Severe weather is also driving demand for industrial A.I. Last year, 27 U.S. natural disasters each caused more than $1 billion in losses. IFS says utilities using Resolve can restore power up to 40 percent faster after storms, floods or wildfires. The system analyzes weather and grid data to predict outages, route crews and coordinate mutual aid.

“We’re solving the hard problems, not retrofitting generic A.I. into critical industries. The context, the data and the risk are completely different and we understand that at a very intimate level,” said Sharma. 

Doyle added that Claude’s broad reasoning ability matters in environments where problems “don’t come pre-labeled and edge cases are constant. A narrowly trained system breaks when it encounters something outside its training distribution,” he said. “Claude’s broad intelligence means it can reason through novel situations and synthesize information across domains even when the specific scenario hasn’t been seen before.”

Anthropic is entering an increasingly competitive industrial A.I. race as rivals invest heavily in infrastructure and sector-specific deployments. OpenAI is building its own network of industrial partnerships, including an alliance with Hitachi that embeds OpenAI models in energy, manufacturing and industrial data systems. Deployments like Mattel’s use of Sora 2 for toy design highlight its push into specialized workflows.

The IFS partnership gives Anthropic something its competitors lack: direct access to field operations, maintenance workflows and disaster-response systems where reliability is paramount.

In a sector where scale is often mistaken for capability, Anthropic is betting that trust, precision and resilience will matter most. If early deployments succeed, industrial A.I. could become one of Claude’s most tangible and consequential success stories.

]]> 1600626
 Sam Altman Projects OpenAI Revenue to Hit $20B—Where the Money Comes From https://observer.com/2025/11/openai-revenue-model/ Mon, 10 Nov 2025 19:43:41 +0000 https://observer.com/?p=1598858

Last week, White House A.I. adviser David Sacks sent a sharp message to Silicon Valley: the U.S. government won’t rescue A.I. companies that overreach. “There will be no federal bailout for A.I.,” he wrote on X. The statement came in response to comments by Sarah Friar, CFO of OpenAI, who had suggested the company might need government support as it spends heavily on A.I. infrastructure. Friar later clarified that OpenAI was not seeking a government “backstop” for its infrastructure commitments.

In reaction to Sacks’s remarks—and apparently in an attempt to reassure investors and calm concerns about an AI bubble—OpenAI CEO Sam Altman said the next day that the company is on a healthy path to profitability despite its large spending. OpenAI is on track to reach “$20 billion in annual recurring revenue by the end of 2025 and grow to hundreds of billions by 2030,” Altman wrote on X.

OpenAI disclosed in June that it had hit $10 billion in annual recurring revenue. Altman’s projection suggests that the company’s revenue will double in a matter of months. However, it continues to spend heavily. For all of 2024, OpenAI generated $5.5 billion in revenue, while posting a net loss of $5 billion, CNBC reported. 

OpenAI’s revenue comes from three main channels:

  • Consumer Subscriptions (55–60 percent): Consumer products, like paid versions of ChatGPT, account for roughly half of OpenAI’s business, according to third-party estimates based on OpenAI’s disclosures and news reports. The free tier acts as a broad funnel to drive users to the $20/month ChatGPT Plus and the $200/month ChatGPT Pro plans. 
  • Enterprise Solutions (25–30 percent): OpenAI products are used by millions of organizations, from startups to Fortune 500 companies. Plans range from ChatGPT Team ($25–30 per user/month) to Enterprise ($60/seat/month), with custom deployments for certain sectors, like education and health care. 
  • API and Developer Platform (15–20 percent): Developers embed OpenAI models into their products (for coding, automation, reasoning). Major firms such as Microsoft, Snowflake, HubSpot and Salesforce have been cited as users of the API.

While revenue is growing rapidly, so is spending. OpenAI is reportedly burning some $8 billion a year, and the rate could climb to $45 billion by 2028. The company has committed roughly $1.4 trillion over the next eight years to building data centers and inked massive GPU deals with Nvidia and AMD.

This pace inevitably raises memories of past tech bubbles, where spectacular growth masked underlying fragility. Altman himself recently warned that investor enthusiasm around A.I. has reached bubble territory, likening it to the late-1990s dot-com boom and calling some startup valuations “insane.”

Still, in his X post last week, Altman reaffirmed the company’s independent stance. “We do not have or want government guarantees for OpenAI data centers,” he said. “We believe that governments should not pick winners or losers, and that taxpayers should not bail out companies that make bad business decisions or otherwise lose in the market.”

]]> 1598858
 The Rise of the Chief A.I. Officer: A New Power Player in Corporate C-Suite https://observer.com/2025/10/chief-ai-officer-gain-importance-corporate-csuite/ Thu, 23 Oct 2025 20:33:10 +0000 https://observer.com/?p=1594480

When A.I. moved from academia to corporate America, it didn’t just change how companies operate—it reshaped what leadership looks like. A title that barely existed a few years ago is now spreading fast: the chief A.I. officer (CAIO). The role signals how deeply A.I. has become embedded in corporate strategy and identity.

According to IBM’s 2025 survey, 26 percent of global enterprises now have a chief A.I. officer, up from 11 percent two years ago. More than half (57 percent) were promoted internally, and two-thirds of executives predict that nearly every major company will have one within the next two years.

The title first appeared in the early 2010s, as deep learning began to take off, but it truly gained momentum after 2023 with the rise of generative A.I. The U.S. government cemented its importance in 2024 through Executive Order 14110, which required every federal agency to appoint a CAIO to oversee A.I. governance and accountability.

The private sector quickly followed suit. A.I. strategists began moving into the C-suite, marking a new kind of leadership role for the algorithmic age.

“A.I. was often a specialist function living under the CTO. Organizations realized A.I. was too strategic to be managed as a side project,” Baris Gultekin, software giant Snowflake’s vice president of A.I., told Observer. “In addition to CAIOs, we often hear that Snowflake customers now also have large internal A.I. councils made up of individuals across departments to strategically and effectively facilitate enterprise-wide A.I. adoption.” Gultekin reports through Snowflake’s product leadership to the CEO.

Some of the most influential chief A.I. officers are already reshaping Big Tech. At Meta, Alexandr Wang, former Scale AI CEO, took on the role in mid-2025, co-leading Meta Superintelligence Labs alongside Nat Friedman, former GitHub CEO. Microsoft’s Mustafa Suleyman, DeepMind co-founder and former Inflection AI CEO, now heads Microsoft AI, overseeing the company’s long-term infrastructure push. At Apple, veteran A.I. leader John Giannandrea, continues to guide the company’s A.I. direction, reporting directly to CEO Tim Cook.

Companies beyond tech are also joining the trend. Lululemon appointed Ranju Das as its first chief A.I. and technology officer in September to boost personalization and innovation. Consulting giant PwC recently appointed Dan Priest, former VP and CIO at Toyota Financial Services, as its first CAIO for the U.S. market. Even universities, such as UCLA and the University of Utah, have added CAIOs to coordinate campuswide A.I. strategy.

From CIO to CDO to CAIO

In the 1980s, chief information officers (CIOs) led the IT revolution; in the 2010s, chief data officers (CDOs) rose with big data; now, CAIOs embody the institutionalization of A.I.

“CAIOs are responsible for exploring what parts of the business can be safely delegated to A.I. agents, how teams can properly govern A.I. decisions, the types of infrastructure needed to serve context-rich data to A.I. systems, and much more,” Sean Falconer, head of A.I. at data streaming platform Confluent, told Observer. “CDOs ensure the data is clean, while CIOs ensure it’s accessible. CAIOs ensure data becomes actionable and capable of reasoning, predicting and taking autonomous steps on behalf of the business.”

In industries like banking, health care and retail, CAIOs often act as translators, turning complex A.I. potential into practical results. “They navigate complex legacy processes and cultural resistance, making upskilling and securing organizational willingness to change as critical as building the models themselves,” Snowflake’s Gultekin said.

The rise of the chief A.I. officer also parallels the growing influence of data engineers. A study by Snowflake and MIT Technology Review Insights found that 72 percent of global executives now view data engineers as essential to business success. More than half said data engineers play a major role in shaping A.I. deployment and determining which use cases are feasible.

“Businesses will always require a CIO, which has also evolved over the years into providing strategic guidance to the business rather than just simply an IT function. Where we see overlap (with CAIOs) are areas that are critical to a company, like governance, tech enablement and strategic alignment,” Bhaskar Roy, chief of A.I. & product solutions at business automation platform Workato, told Observer. “The mandate for CAIOs is clear: continuously push the boundaries of what’s possible with A.I., and ensure the organization remains at the forefront of technological change, all while listening to customers’ needs and concerns.”

]]> 1594480
 OpenAI Overhauls Copyright Policy After Sora 2’s Pokémon Mania Backfires https://observer.com/2025/10/openai-sora2-pokemon-video-copyright-issue/ Mon, 06 Oct 2025 21:21:33 +0000 https://observer.com/?p=1591966

When OpenAI’s Sora 2 launched on Sept. 30, it was billed as a playful experiment in A.I. video generation. Within 48 hours, the app, which lets users create short-form videos from simple text prompts, soared to the top of Apple’s App Store with more than 160,000 downloads. Notably, Japanese franchises such as Pokémon and Mario dominated Sora 2’s early outputs, flooding X and Instagram with A.I.-generated clips like Mario getting arrested for reckless driving and Pikachu cosplaying as Batman. The viral frenzy soon sparked backlash from intellectual property holders and lawmakers in Japan, prompting OpenAI to revise its content policy.

Nintendo, which owns the Pokémon and Mario franchises, said on X that it “will take necessary actions against infringement of our intellectual property rights.” Akihisa Shiozaki, a Japanese lawyer and member of Japan’s House of Representatives, urged immediate action on X to protect the nation’s content industry. “When I tried inputting prompts into Sora 2 myself, it generated footage of popular anime characters with a quality indistinguishable from the originals, one after another. Yet, for some reason, characters owned by major U.S. companies, like Mickey Mouse or Superman, didn’t appear,” he wrote on X. “This was clearly an imbalance and potentially a serious issue under copyright law.”

In a blog post on Oct. 3, OpenAI CEO Sam Altman announced policy changes following the surge of videos featuring Nintendo characters. “We’d like to acknowledge the remarkable creative output from Japan—we’re struck by how deep the connection between users and Japanese content is,” he wrote.

OpenAI’s previous “opt-out” policy, which allowed IP holders and creators to request removal of their works from the training data, was replaced with a stricter “opt-in” system. Under the new rule, the company must receive explicit permission from rights holders before Sora can generate content featuring their IP. Altman said the change would give rights holders “granular control over character generation,” aligning OpenAI’s approach with existing likeness and IP protection standards.

He admitted the company had underestimated how quickly users would push the boundaries, noting that “there may be some edge cases of generations that get through that shouldn’t,” and that refining the system “will take some iteration.”

Altman also suggested that creators could eventually earn royalties when their characters appear in Sora-generated videos. “We are going to try sharing some of this revenue with rightsholders who want their characters generated by users,” he wrote. “Our hope is that the new kind of engagement is even more valuable than the revenue share, but of course, we want both to be valuable.”

OpenAI is already entangled in a growing number of lawsuits from authors, media companies and other rights holders who accuse the company of using copyrighted material without permission to train its models. The New York Times sued OpenAI and Microsoft last year, alleging that ChatGPT reproduced many of its articles verbatim. A separate group of fiction writers, including George R.R. Martin, John Grisham and Jonathan Franzen, has filed a similar suit, arguing that OpenAI’s training methods violate copyright law by replicating their works.

OpenAI is trying to find a middle ground with the creative community, without much guidance from existing IP laws. “There are likely to be many gray areas, and the definition of fair use will need to be clearly defined. Past legal precedents are not very helpful,” Rahul Telang, an information systems professor at Carnegie Mellon University, told Observer.

A.I. companies “realize that, without giving more power to content creators, they cannot succeed,” Telang added. He said opt-in and revenue-sharing are a good start, but “will not be a permanent solution.”

One of OpenAI’s early backers, Vinod Khosla, came to the company’s defense. The billionaire venture capitalist hit back at critics of Sora 2, calling them “tunnel-vision creatives” who lack imagination.

“Let the viewers of this ‘slop’ judge it, not ivory tower luddite snooty critics or defensive creatives. Opens up so many more avenues of creativity if you have an imagination. This is the same initial reaction to digital music in the 90s and digital photography in the 2000s,” he wrote on X. “There will be a role for traditional video still, but many more dimensions of creative video [through A.I.].”

For now, Sora stands as both a marvel and a warning—a glimpse of how democratic storytelling might become possible through A.I. and how quickly those boundaries can be tested. The moment may have gone viral, but the ethical battle it unleashed is just beginning.

]]> 1591966
 Google Study Shows A.I. Writes Code, But Developers Still Don’t Fully Trust It https://observer.com/2025/09/google-study-ai-code-engineer-trust-issue/ Tue, 23 Sep 2025 15:32:29 +0000 https://observer.com/?p=1582384

For decades, software was built line by line by human hands. That process is changing fast because of A.I. According to Google’s latest annual DORA: State of A.I.-assisted Software Development report, released today (Sept. 23), 90 percent of technology professionals now use A.I. in their workflows, representing a 14 percent jump from last year. The survey of more than 5,000 software professionals and IT specialists found that developers rely on A.I. for tasks ranging from writing code snippets to running tests and reviewing security.

Despite higher A.I. adoption, however, trust in the technology remains low. While most say A.I. makes them faster and more productive, only 24 percent say they trust it “a lot” or “a great deal.” Nearly a third admit they trust it “a little” or not at all.

“Boardroom-level prioritization shows that this change is likely here to stay. Every organization is facing pressure to improve software performance even in the face of broad economic pressures and constraints,” Nathen Harvey, the study’s lead researcher and a developer advocate at Google Cloud, told Observer. “On an individual level, A.I. has captured the human imagination and inspired developers to find ways to drive both top and bottom-line improvements for businesses.”

The study  found that 85 percent of professionals say A.I. has made them more productive, though 41 percent call the improvement only “slight.” Fewer than 10 percent reported any decline in productivity. Developers now spend a median of two hours a day using these A.I. tools, and top-performing organizations report that A.I. is boosting throughput, allowing them to deliver applications faster and more reliably.

Code quality is where A.I.’s impact is most evident. Much of the software it helps create ends up running in production systems far longer than developers ever anticipated. That longevity shows A.I.–generated code can be more useful than expected, but it also raises the stakes. Readability and adaptability matter far more than quick fixes when judging code quality.

Software relies on constant code changes, such as tweaks, fixes and new features, to stay alive. Feedback loops from automated tools or users act like vital signs, signaling the system’s health. But Harvey cautioned that while A.I. speeds development, it can also make software delivery more unstable. “Even with the help of A.I., teams will still need ways to get fast feedback on the code changes that are being made,” he said. 

For now, developers are hesitant to give up control. Only a quarter in the survey say they have high trust in A.I.’s coding abilities. Google researchers call this the “trust paradox”: A.I. is a useful assistant, but not yet a true partner. That skepticism could slow progress toward advanced uses like autonomous testing or fully automated release management.

Harvey noted that developers treat A.I. output with the same healthy skepticism they apply to go-to resources, like coding solutions found on Stack Overflow—useful but never blindly trusted. “A.I. is only as good as the data it has access to,” he said. “If your company’s internal data is messy, siloed, or hard to reach, your A.I. tools will give generic, unhelpful answers, holding you back instead of helping.”

Harvey also noted that A.I. hasn’t eased burnout or reduced friction. While it boosts individual productivity, those challenges often stem from company culture, leadership and processes—problems technology alone can’t fix. “If leaders start expecting more because A.I. makes developers faster, it could even add to the pressure,” he added.

To address this gap, Google introduced the DORA A.I. Capabilities Model, a framework of seven technical and cultural practices aimed at amplifying A.I.’s impact. The model emphasizes user focus, clear communication and small-batch workflows—underscoring that success requires more than just new tools.

“Culture and mindset continue to be huge influences on helping teams achieve and sustain top performance. A climate for learning, fast flow, fast feedback, and a practice of continuous improvement are what drive sustainable success. A.I. amplifies the necessity for all of these and provides the catalyst to transform along the way,” said Harvey.

Ultimately, Google’s 2025 report argues the biggest barrier isn’t adoption but trust. Without stronger confidence in A.I.’s reliability, the future of software development will depend as much on winning developer faith as on improving the technology itself.

]]> 1582384
 20-Something Founders Debut $100M A.I. Assistant Approved by 6K Silicon Valley Users https://observer.com/2025/09/startup-interaction-launch-ai-message-assistant/ Mon, 15 Sep 2025 17:03:44 +0000 https://observer.com/?p=1580369

For more than a decade, tech companies have promised the arrival of the digital sidekick. Amazon built Alexa to anticipate household needs. Apple gave us Siri, and Google launched Assistant to answer questions and organize daily life. More recently, OpenAI’s ChatGPT showed the raw potential of generative A.I. Yet despite the hype, today’s assistants remain outside the natural flow of conversation. They often demand a separate device, a dedicated app or new user habits. And none have managed to plug into the chaotic center of modern life, where people juggle emails, reminders, travel changes, invoices and group texts. But what if using A.I. felt as effortless as texting a friend—no learning curve, no extra steps?

That’s the bet of Interaction, a Palo Alto startup whose A.I. assistant, “Poke,” slips directly into iMessage. Instead of asking users to download yet another app, the founders designed Poke to meet people where their attention already is: their messaging threads.

Marvin von Hagen and Felix Schlegel, the company’s co-founders, noticed how often people bounce between apps and saw an opportunity to pull A.I. into conversations at the exact moment it’s needed. Poke works like any other contact in iMessage or WhatsApp, instantly able to book a flight, summarize research, or suggest dinner options—all without leaving the thread.

The startup beta tested the idea among more than 6,000 users in Silicon Valley, from companies like OpenAI, Google, Stripe, Figma and Anthropic, who collectively sent more than 200,000 messages each month. Along the way, von Hagen and Schlegel uncovered surprising use cases: a parent asking Poke to generate math exercises for a child during homework, friends planning a Barcelona trip with bookings and itineraries, and even freelancers chasing overdue invoices. They realized that when an assistant blends into the natural rhythm of communication, people discover dozens of uses without ever being taught how.

“Users want easy access to A.I. the same way they text their partner, friends, parents and colleagues,” Schlegel, the company’s CTO, told Observer. He added that, when chatting with Poke, users train the A.I. architecture to learn their preferences through a personalized relevance engine.

“Beta testers shared personal stories and sought relationship advice, not just email-related tasks,” von Hagen, Interaction’s CEO, told Observer. “This taught us that personality, conversationality and emotional intelligence were as important as technical performance.”

Von Hagen, 23, and Schlegel, 25, first met studying computer science at the Technical University of Munich, where they built reputations for tackling outsized projects at unusually young ages. Before launching Interaction, they co-founded TUM Boring, a 65-person student team that twice won Elon Musk’s Not-a-Boring competition by designing and constructing a tunneling machine from scratch.

Von Hagen is also known in the A.I. safety world for exposing vulnerabilities in large language models, helping raise awareness of risks like prompt injection. His probing of Microsoft’s Bing famously uncovered its hidden alter ego, “Sydney,” along with a set of internal rules. He later pursued research on collective intelligence at MIT and interned at Tesla.

Schlegel started coding at 12, creating web and AR apps and presenting onstage at Apple’s WWDC before finishing high school. He went on to research simultaneous localization and mapping (SLAM) at the University of Cambridge and later joined Stanford as a researcher in machine learning and biodesign.

A $100 million bet

That mix of technical depth and early ambition helped persuade investors like General Catalyst. “Interacting with Poke over iMessage is like having an assistant that interacts in a tone almost like a close friend,” Yuri Sagalov, managing director at General Catalyst and one of the 6,000 beta testers on the app, told Observer. “It shows personality and demonstrates an understanding of you, making the experience feel both personal and useful.”

Investors see Poke’s advantage in its timing: messaging apps have become command centers of modern life, and an A.I. that can slip seamlessly into those threads could tap into nearly every corner of a smartphone. Banking on that belief, Interaction officially launched last week after months of private testing.

Alongside the launch, the company announced a $15 million seed round valuing Interaction at $100 million. The round was led by General Catalyst and Village Global, with participation from Earlybird VC and angel investors including PayPal co-founder Ken Howery, Dropbox co-founder Arash Ferdowsi, former Google VP Bradley Horowitz and OpenAI researcher Karina Nguyen.

The founders envision a future with fewer standalone apps, where A.I. proactively surfaces what matters—an upcoming deadline, an unanswered email—before the user even asks. To do this, Poke combines models from Anthropic, OpenAI, Voyage and Mistral AI with in-house, fine-tuned models, balancing speed and cost. The team hasn’t disclosed which models handle specific tasks.

Questions about privacy and scaling

Still, the question looms: will users trust an A.I. in their iMessage threads, with access to sensitive data like payments or work messages? Von Hagen and Schlegel say they’re addressing those concerns head-on. Interaction is building on SOC 2 compliance standards and planning integrations with financial services, scheduling and team tools.

“By default, our users are in ‘Maximum Privacy’ mode, which creates a complete data firewall, so even our engineers can’t access conversations or emails. We’re transparent about what data Poke needs to function and why, giving users granular control over what they share,” Schlegel said.

The real test, however, will be whether people outside Silicon Valley want an assistant that doesn’t live on the home screen. “We will prioritize localization to make Poke available in more countries, expand our phone number coverage with international codes, and build distributed infrastructure to reduce response times outside the U.S.,” Schlegel said. “A family in Mumbai might use Poke in Hindi on WhatsApp with intermittent connectivity, which presents very different challenges than serving Palo Alto founders on the latest iPhone.”

]]> 1580369
 OpenAI GPT vs. Google Gemini: the High-Stakes Battle Shaping the Future of A.I. https://observer.com/2025/09/openai-gpt-google-gemini-rival/ Mon, 08 Sep 2025 17:52:33 +0000 https://observer.com/?p=1579345

The race to build the leading A.I. model has narrowed into a high-stakes duel between OpenAI’s GPT and Google’s Gemini, each embodying a different vision for the technology’s role in society. OpenAI leans on a consumer-first approach and creative fluency, while Google emphasizes enterprise integration and technical scale. Their dominance is underscored by the fact that even rivals are adopting their models.

Apple, which is developing its own A.I. systems known as Apple Intelligence, is reportedly preparing to integrate Gemini into its digital assistant Siri. The project, codenamed “World Knowledge Answers,” according to Bloomberg, aims to transform Siri into a multimodal answer engine capable of handling unstructured text, image and video. The iPhone maker is expected to share more details at its annual product launch event tomorrow (Sept. 9). For years, Apple has avoided relying on external technology to power its core products. This move, if materialized, represents a major shift.

Meta, while building its in-house Llama models, uses both GPT and Gemini in its apps, integrating them in its flagship chatbot, Meta AI, across Facebook, Instagram and WhatsApp. These partnerships are seen as temporary until future versions of Llama—such as the upcoming Llama 5—can close the gap with industry leaders. The current versions (Llama 3 and 4) still lag GPT and Gemini in versatility and multimodal scale.

Gemini vs. GPT

Gemini is built for scale and technical precision. It offers a substantial one-million-token context window (soon to expand to two million) that allows enterprises to process entire codebases, lengthy regulatory filings or vast datasets in a single session. The model’s strength lies in structured reasoning, multimodal analysis across text, images, code, and increasingly video and audio.

OpenAI’s GPT, by contrast, excels in creative and conversational tasks. Powering ChatGPT, it is known for polished writing, imaginative output, and seamless real-time interactions with third-party tools.

While Gemini leads in reasoning and long-term memory, GPT continues to lead in adoption and consumer recognition. ChatGPT alone handles an estimated 2.5 billion prompts per day and is also popular in the workplace: about 55 percent of enterprises now use GPT-powered services, according to third-party surveys. In August, OpenAI secured a landmark federal contract to make ChatGPT Enterprise available across federal government agencies for just $1 each.

Google, however, is gaining ground through native integration. By early 2025, Gemini was embedded in 63 percent of enterprises, according to a U.K. survey, thanks largely to its presence in Google Workspace. From Gmail drafting to Meet transcription and Docs research, Gemini is becoming an office fixture.

The bigger picture

Beyond the GPT–Gemini duopoly, other players are gaining momentum. Anthropic’s Claude commands a 32 percent market share in compliance-heavy and coding-intensive sectors, according to a study by venture capital firm Menlo Ventures. Its clients include Amazon, Snowflake, Thomson Reuters, among others.

Meta’s Llama has become the leading open-source model, appealing to companies seeking control and customization. Elsewhere, DeepSeek dominates in China, while Cohere finds success with enterprises outside the U.S. that prioritize data sovereignty.

Increasingly, organizations are adopting a multi-model strategy: Claude for compliance, GPT for creativity, Gemini for productivity, and Llama for control. With Apple turning to Gemini for Siri and OpenAI expanding GPT into government and Fortune 500 firms, the choices made today could reshape the A.I. landscape for years to come.

]]> 1579345
 Jensen Huang’s China Balancing Act Is About More Than Selling Chips https://observer.com/2025/08/nvidia-china-chip-strategy/ Wed, 27 Aug 2025 17:24:46 +0000 https://observer.com/?p=1572867

Nvidia CEO Jensen Huang finds himself in an increasingly difficult balancing act in China. After months of effort persuading President Trump to ease restrictions and allow the company to resume A.I. chip sales to the country, Huang was hit with an unexpected reversal: Last month, Chinese regulators moved to block Nvidia’s H20 chips, forcing the company to halt production of its China-specific product line. In response, Nvidia has shifted to developing a new processor, called the B30A, Reuters reported last week..

Built on the company’s Blackwell architecture, the B30A will include NVLink and high-bandwidth memory but deliver only about half the performance of Nvidia’s flagship B300 chip, keeping it within U.S. export-control limits. Alongside it, Nvidia is also preparing a less advanced RTX6000D chip aimed at professional and inference workloads. Both chips are designed to comply with Washington’s tightening rules while keeping a foothold in the world’s second-largest economy.

The moves come amid broader political pressures. In May, Senator Tom Cotton and several House Republicans introduced the Chip Security Act, which would mandate that A.I. chips from companies like Nvidia include built-in location verification. While that bill has not advanced in Congress, Chinese authorities issued guidance earlier this summer urging local firms to avoid Nvidia products over alleged security risks, including the possibility of remote tracking or control.

Nvidia’s challenges are compounded by the revenue-sharing terms attached to its export license. The company now must send 15 percent of its China A.I. chip sales back to the U.S. government—a levy that some industry leaders warn could discourage innovation.

“President Trump imposing a 15 percent tax on our innovation companies sets a dangerous precedent,” Arnie Bellini, general partner at Bellini Capital, told Observer. Bellini estimates that China accounts for 15 to 20 percent of Nvidia’s global revenue (consistent with other third-party estimates), and existing H20 inventories may generate as much as $6 billion in sales, giving Nvidia a cushion as it pivots to new China products.

“Even with the tariff imposed, Nvidia will make significant revenue on any sales to China. The deal gives breathing room to meet the next two quarters,” Maribel Lopez, founder and principal analyst at Lopez Research, told Observer.

Nvidia’s share of revenue from China has actually been declining in recent years, dropping from 21 percent in 2023 to below 15 percent in the most recent fiscal year, according to Business Insider. Still, China’s importance isn’t just about sales; it’s about keeping developers there tied to Nvidia’s technology. “In order for America to have A.I. leadership is to make sure that the American tech stack is available to markets all over the world so that amazing developers, including the ones in China, are able to build on the American tech stack,” Huang said in an interview last month with CNN.

Nvidia will report second-quarter earnings after the market close today, with Wall Street expecting strong results on the back of Blackwell-driven data center growth. Blackwell already accounts for 70 percent of Nvidia’s data center revenue, a clear sign of Nvidia’s dominance over the A.I. market. Analysts will see whether the company’s new China-compliant chips can keep its momentum going.

]]> 1572867
 Sam Altman to Visit India Next Month, Announces OpenAI’s First New Delhi Office https://observer.com/2025/08/sam-altman-openai-india-office/ Mon, 25 Aug 2025 21:12:59 +0000 https://observer.com/?p=1572535

After a cooler-than-expected reception to GPT-5 and mounting pressure from rising training, compute and infrastructure costs, OpenAI is looking to India as a cornerstone of its global expansion strategy. On Friday, CEO Sam Altman announced on X that the company will open its first office in New Delhi later this year. He also said he plans to visit the country next month, writing, “A.I. adoption in India has been amazing to watch—ChatGPT users grew 4x in the past year—and we are excited to invest much more in India!”

India has become OpenAI’s second largest market for ChatGPT, trailing only the U.S., according to Altman. To appeal to local users, the company has rolled out ChatGPT Go, a $5 per month subscription pitched as a budget-friendly alternative to the Plus and Pro tiers ($20 and $200 per month, respectively). Marketed toward students and enterprises, ChatGPT Go promises access to premium features such as longer context memory, higher usage limits and advanced tools like editing custom GPTs to build A.I. tools tailored to specific user needs.

Altman has visited India multiple times in recent years, including a 2023 meeting with Prime Minister Narendra Modi, where he praised the country’s rapid adoption of A.I., saying it has “all the ingredients to become a global A.I. leader.” In June, OpenAI deepened its ties to the country by partnering with the Indian government’s IndiaAI Mission, an initiative to expand A.I. access nationwide.

But rivals are also circling the market. Google and Meta already operate major A.I. products and R&D hubs in India, while Perplexity AI, founded by Indian entrepreneur Aravind Srinivas, is seeing explosive growth. Perplexity’s monthly active users in India jumped 640 percent year-over-year in the second quarter of 2025, far outpacing ChatGPT’s 350 percent growth in the same period. While ChatGPT positions itself as a conversational assistant, Perplexity markets its tool as an A.I.-powered search engine that delivers cited answers, blending its own retrieval-augmented system with models from OpenAI and Anthropic.

In April, both OpenAI and Perplexity launched WhatsApp bots globally, aiming to integrate A.I.-powered chat and search into everyday messaging. Given WhatsApp’s ubiquity in India, the move could prove pivotal. “Perplexity on WhatsApp is super convenient way to use A.I. when in a flight. Flight WiFi supports messaging apps the best. And WhatsApp has been heavily optimized for this because it grew to support countries where connectivity wasn’t the best,” Srinivas wrote on LinkedIn in May.

OpenAI has been steadily expanding its global footprint, adding offices in London, Dublin, Paris, Brussels, Munich, Tokyo and Singapore over the past year. The company is headquartered in San Francisco and also maintains U.S. offices in New York and Seattle.

]]> 1572535
 Elon Musk Shuts Down Tesla’s Dojo Supercomputer As It Hits ‘Evolutionary Dead End’ https://observer.com/2025/08/elon-musk-shut-tesla-dojo-supercomputer/ Wed, 13 Aug 2025 18:23:18 +0000 https://observer.com/?p=1570651

Tesla has officially ended its Dojo supercomputer project, closing out a four-year effort to develop one of the world’s most powerful A.I. training systems and marking a major shift in Tesla’s A.I. ambitions. CEO Elon Musk announced the shutdown in a series of posts on X over the weekend. He also confirmed that the entire Dojo team of about 20 employees has been disbanded.

Dojo is powered by Tesla’s D1 chip, designed in-house (and manufactured by TSMC) to handle massive volumes of driving data for training the company’s Full Self-Driving (FSD) system. With Dojo shelved, Tesla is pivoting to a streamlined chip strategy, focusing on next-generation chips like AI5 (manufactured by TSMC) and AI6 (manufactured by Samsung) for both training and deployment. AI5 chip is built to power self-driving and robotics capabilities, while AI6 can additionally handle large-scale A.I. training tasks.

“Once it became clear that all paths converged to AI6, I had to shut down Dojo and make some tough personnel choices, as Dojo 2 was now an evolutionary dead end,” Musk explained on X.

First unveiled at Tesla’s AI Day in 2021, Dojo was pitched as a breakthrough that would reduce reliance on third-party chip suppliers such as Nvidia, while delivering greater bandwidth, lower latency and lower costs.

The decision follows months of internal turbulence in the Dojo division, including the loss of several key engineers. Musk noted that it no longer made sense for Tesla to divide its resources between two different A.I. chip product lines—one optimized for inference and another for training. Inference refers to the process of running an already-trained A.I. model to make real-time decisions, such as identifying objects on the road in a moving car. Training, by contrast, is the computationally intensive process of teaching an A.I. model by feeding it massive amounts of data until it can recognize patterns accurately.

“In a supercomputer cluster, it would make sense to put many AI5/AI6 chips on a board, whether for inference or training, simply to reduce network cabling complexity and cost by a few orders of magnitude,” Musk wrote on X.

The shutdown also comes as Tesla faces mounting challenges in its core electric vehicle business. In the latest quarter ended June 30, Tesla’s EV revenue dropped 16 percent from the previous year, while total revenue fell 12 percent year-over-year. The company’s U.S. market share has also slid sharply, falling to less than 50 percent from 75 percent in 2022.

]]> 1570651
 This AI Startup Founded by a Google Intern Bids $34.5B to Acquire Chrome Browser https://observer.com/2025/08/perplexity-ai-bids-google-chrome-browser/ Wed, 13 Aug 2025 15:43:58 +0000 https://observer.com/?p=1570546

Perplexity AI, the high-flying A.I. startup founded by former Google intern Aravind Srinivas, has made a bold all-cash offer of $34.5 billion to acquire Google’s Chrome browser, Reuters reported. The bid is nearly twice Perplexity’s own valuation and would require significant external financing. With more than three billion users worldwide, Chrome is one of Google’s most valuable strategic assets, serving as a dominant gateway to search traffic and user data in an A.I.-driven internet era.

The timing of Perplexity’s bid is notable. Google’s parent company Alphabet is already facing intense antitrust scrutiny in the U.S. Last year, a federal court ruled that Alphabet unlawfully monopolized online search. The Department of Justice has suggested that Alphabet be forced to divest Chrome as a potential remedy, with Judge Amit Mehta expected to rule on penalties later this month.

Perplexity’s proposal appears designed to leverage that regulatory pressure. According to multiple reports, the company has promised to keep Google Search as Chrome’s default search engine, maintain the Chromium open-source codebase, and invest $3 billion into browser development over the next two years. It also claims to have commitments from several unnamed investment funds to fully finance the deal if Alphabet agrees. Alphabet, for its part, has given no indication it intends to sell.

“The acquisition attempt is a signal that control over the browser is becoming one of the most valuable frontiers in the A.I. era,” Alon Yamin, co-founder and CEO of Copyleaks, an A.I.-based content verification platform, told Observer. “Whoever owns the gateway to the web holds immense influence over how information is accessed, prioritized and trusted.”

Earlier this year, Perplexity debuted Comet, its own A.I.-powered browser. Combining Comet with Chrome’s global reach could give the company unprecedented influence over how people access and interact with information online.

Perplexity is not the only player eyeing Chrome. OpenAI, Yahoo, and private-equity giant Apollo Global Management have all reportedly considered bids in recent years.

Founded in 2022 by Srinivas, Denis Yarats, Johnny Ho and Andy Konwinski, Perplexity made its name with a conversational search engine that delivers summarized answers instead of links.

Srinivas’s connection to Google is limited to a brief research internship from May 2020 to April 2021. The now 31-year-old engineer later worked as a research scientist at OpenAI and conducted graduate research under deep learning pioneer Yoshua Bengio at the University of California, Berkeley.

]]> 1570546
 UK Startup Led by Former F1 Engineer Reinvents Manufacturing with A.I. https://observer.com/2025/06/physicsx-startup-ai-manufacturing/ Thu, 26 Jun 2025 15:37:33 +0000 https://observer.com/?p=1562997

PhysicsX, a London-based startup applying cutting-edge A.I. to solve some of the toughest challenges in industrial engineering, has raised $135 million in a series B funding round. The company’s software uses large physics models (LPM) to simulate and optimize complex systems, promising faster innovation and performance gains in fields from aerospace to medical devices. The latest funding round was led by London VC firm Atomico, with additional backing from Siemens, Temasek (Singapore’s state-owned investment company), Applied Materials, and Berlin- and New York-based venture capital firm July Fund.

Founded in 2019 by Robin Tuluie and Jacomo Corbo, PhysicsX aims to tackle complex design and performance challenges in manufacturing. Tuluie previously led R&D for Formula One teams, including Renault (Alpine) and Mercedes. Corbo is a physicist who co-founded the A.I. firm QuantumBlack, which was acquired by McKinsey & Co. in 2015. Their experience in high-performance industries exposed a persistent challenge: as systems become more advanced, even experienced engineers struggle to navigate trade-offs and push performance forward using traditional tools and instincts.

“We’re setting out to change engineering, shifting physics simulation from physical testing to inference with A.I.,” Corbo told Observer. “The last few years have seen remarkable advances in A.I., allowing us to build models capable of predicting the complex physics and chemistry relevant to some of the most critical applications in advanced manufacturing industries.”

PhysicsX aims to solve industrial problems that conventional computer-aided design (CAD) and simulation tools cannot address. Its enterprise software platform combines A.I., deep learning, and domain expertise to enhance the entire engineering lifecycle—from early concept development to design optimization and deployment.

“Innovators often struggle with the choices, or hit a wall of performance that they cannot break through,” Tuluie told Observer. “Wouldn’t it be great to have a capability that could iterate incredibly fast, give answers to the optimum design and manufacturing implications, and last without breaking or wearing out? This superpower is exactly what we are seeking to provide to our users.”

PhysicsX uses large physics models (LPMs), or specialized algorithms built for physics-intensive research that draw on the power of large language models. These models can process any type of data and unify insights across the entire engineering lifecycle, significantly reducing the time spent on complex tasks that can otherwise take weeks. According to Tuluie and Corbo, the PhysicsX’s platform can unlock entirely new categories of hardware, from next-generation semiconductors to sustainable aircraft. While the company did not name specific clients, its A.I. platform is already being integrated into the workflows of leading manufacturing and engineering firms.

Tuluie said his company has already made significant breakthroughs, such as cutting the scrap rate of jet engine turbine blades by 70 percent for an aerospace client by integrating learned physics models into the quality assurance process. In another case, PhysicsX improved the energy efficiency and reduced blood damage of an artificial heart by 42 percent using the same approach, with generative A.I. used to design the pump’s blade shapes—achieving flawless performance in medical trials. He added that even more complex problems, once considered unsolvable, may soon be within reach.

“The manufacturing world knows it needs to reinvent itself,” Corbo said. “For decades, industries like aerospace and defense, automotive, semiconductors, materials and energy have been held back by the limitations of legacy tooling.”

He added that the most competitive hardware companies of the next decade will be those that operate more like software firms: agile, data-driven and continuously optimizing.

PhysicsX previously raised $32 million in a Series A round in November 2023. The company did not disclose its current valuation following the latest funding.

]]> 1562997
 Over 80% of Companies Embracing A.I. See No Real Gains Yet, McKinsey Finds https://observer.com/2025/06/mckinsey-study-business-ai-productivity/ Tue, 17 Jun 2025 20:06:33 +0000 https://observer.com/?p=1560499

Companies across all industries are rapidly adopting A.I., drawn by the promise of increased productivity and reduced costs. However, more than 80 percent of businesses using the technology are not yet seeing significant earnings gains, according to a new report by McKinsey & Company published last week. McKinsey calls this the ‘‘gen A.I. paradox’’: while the use of A.I. tools is widespread, their measurable business impact remains elusive. That’s because real-world A.I. adoption lags behind the progress of the technology itself. While “agentic A.I.” today dominates discussions at industry giants like Google and OpenAI, companies that don’t build their own A.I. systems still largely operate in the “copilot” phase.

Copilots, a term popularized by Microsoft since early 2023 (shortly after the launch of ChatGPT), are designed to assist users with specific, prompt-based tasks. These tools are helpful but require human direction and can only operate within narrowly defined use cases—think Microsoft Copilot in Word, which can rephrase sentences, or a sales tool that drafts follow-up emails. Because copilots require a human to initiate each task, their value is limited to the scope of the task, such as writing faster, organizing data or summarizing meetings. These benefits are real but often difficult to quantify and rarely scale across an entire organization.

In contrast, A.I. Agents can manage entire processes autonomously from start to finish. These systems can plan, adapt to evolving conditions and datasets within enterprise workflows, and even make decisions about the next best steps. In a customer service scenario, for example, an A.I. agent could handle a customer support ticket from initial input to resolution independently, even escalating issues to a human only when necessary.

Most companies are still stuck in “pilot mode,” running small-scale experiments in isolated teams, the report said. To reap the benefits of agentic A.I., organizations must “reset their A.I. transformation approaches from scattered initiatives to strategic programs, from use cases to business processes, and from siloed A.I. teams to cross-functional transformation squads,” the consulting firm advised. 

Among companies that have successfully embraced agentic A.I., the results are significant. For instance, Lenovo’s engineering teams experienced up to a 15 percent improvement in code quality and speed after implementing A.I. agents, McKinsey found in a case study. And in customer support, Lenovo’s A.I. agents resolved the majority of inbound queries, leading to a reduction in response times by up to 90 percent.

The cost of remaining in the copilot phase can be hefty. McKinsey warned that, in a world where competitors can condense a month’s worth of work into a single day using agentic systems, the cost of remaining in “pilot mode” may soon exceed the cost of doing nothing at all.

McKinsey did not disclose which specific companies were included in this report. However, it said the report was built upon an earlier study conducted this year that surveyed businesses with over 500 employees across various industries.

]]> 1560499
 Sam Altman Says Humans are Already Past ‘the A.I. Event Horizon’ https://observer.com/2025/06/sam-altman-says-humans-are-already-past-the-a-i-event-horizon/ Thu, 12 Jun 2025 21:00:06 +0000 https://observer.com/?p=1559800

OpenAI CEO Sam Altman believes we are already past “the A.I. event horizon,” he said in a new blog post yesterday (June 11), arguing that A.I. development is quietly reshaping civilization—even if the shift feels subtle. “The takeoff has started. Humanity is close to building digital superintelligence, and at least so far, it’s much less strange than it seems it should be,” he wrote.

According to the OpenAI CEO, 2025 marks a pivotal shift in A.I. capabilities, particularly in coding and complex reasoning. By next year, he expects A.I. systems to begin generating original scientific ideas, with autonomous robots functioning effectively in the physical world by 2027.

“In the 2030s, intelligence and energy are going to become wildly abundant. These two have long been the fundamental limiters on human progress,” he wrote. “With abundant intelligence and energy (and good governance), we can theoretically have anything else.”

One key driver of this shift is A.I. infrastructure, such as computing power, servers and data center storage. As it becomes more automated and easier to deploy, the cost of intelligence could soon be as low as electricity. And it will supercharge scientific discovery, enable infrastructure to build itself, and unlock new frontiers in health care, materials science and space exploration. “If we can do a decade’s worth of research in a year, or a month, then the rate of progress will obviously be quite different,” Altman wrote.

Altman also addressed a common question: how much energy does a ChatGPT query use? He revealed that a typical query consumes just 0.34 watt-hours of energy and 0.000085 gallons of water—roughly the same amount of power an oven uses in a second and as little water as one-fifteenth of a teaspoon.

While some fear that A.I. could render human labor obsolete, Altman believes that by 2030, A.I. will amplify human creativity and productivity, not replace it. “In some big sense, ChatGPT is already more powerful than any human who has ever lived. A small new capability can create a hugely positive impact,” he wrote.

However, Altman also acknowledged the dangers. He noted that alignment—the challenge of ensuring A.I. systems understand and follow long-term human values—is still unsolved. He cited social media algorithms as an example of poorly aligned A.I. systems—tools optimized for engagement that often result in harmful societal outcomes.

The real threat is not that A.I. will replace human purpose, but that society might fail to evolve the systems and policies necessary for people to thrive alongside increasingly intelligent machines. He urged global leaders to begin a serious conversation about the values and boundaries that should guide A.I. development before the technology becomes too deeply entrenched to redirect.

“The sooner the world can start a conversation about what these broad bounds are and how we define collective alignment, the better,” he wrote.

]]> 1559800