Google's custom AI chips draw billions from Meta and Anthropic, challenging Nvidia's dominance

Why Google's custom AI chips are shaking up the tech industry

Nvidia's position as the dominant supplier of AI chips may be under threat from a specialised chip pioneered by Google, with reports suggesting companies like Meta and Anthropic are looking to spend billions on Google's tensor processing units. The success of the artificial intelligence industry has been in large part based on graphical processing units (GPUs), a kind of computer chip that can perform many parallel calculations at the same time, rather than one after the other like the computer processing units (CPUs) that power most computers. GPUs were originally developed to assist with computer graphics, as the name suggests, and gaming. "If I have a lot of pixels in a space and I need to do a rotation of this to calculate a new camera view, this is an operation that can be done in parallel, for many different pixels," says Francesco Conti at the University of Bologna in Italy. This ability to do calculations in parallel happened to be useful for training and running AI models, which often use calculations involving vast grids of numbers performed at the same time, called matrix multiplication. "GPUs are a very general architecture, but they are extremely suited to applications that show a high degree of parallelism," says Conti. However, because they weren't originally designed with AI in mind, there can be inefficiencies in the ways that GPUs translate the calculations that are performed on the chips. Tensor processing units (TPUs), which were originally developed by Google in 2016, are instead designed solely around matrix multiplication, says Conti, which are the main calculations needed for training and running large AI models. This year, Google released the seventh generation of its TPU, called Ironwood, which powers many of the company's AI models like Gemini and protein-modelling AlphaFold. Technologically, TPUs are more of a subset of GPUs than an entirely different chip, says Simon McIntosh-Smith at the University of Bristol, UK. "They focus on the bits that GPUs do more specifically aimed at training and inference for AI, but actually they're in some ways more similar to GPUs than you might think." But because TPUs are designed with certain AI applications in mind, they can be much more efficient for these jobs and save potentially tens or hundreds of millions of dollars, he says. However, this specialisation also has its disadvantages and can make TPUs inflexible if the AI models change significantly between generations, says Conti. "If you don't have the flexibility on your [TPU], you have to do [calculations] on the CPU of your node in the data centre, and this will slow you down immensely," says Conti. One advantage that Nvidia GPUs have traditionally held is that there is simple software available that can help AI designers run their code on Nvidia chips. This didn't exist in the same way for TPUs when they first came about, but the chips are now at a stage where they are more straightforward to use, says Conti. "With the TPU, you can now do the same [as GPUs]," he says. "Now that you have enabled that, it's clear that the availability becomes a major factor." Although Google first launched the TPU, many of the largest AI companies (known as hyperscalers), as well as smaller start-ups, have now started developing their own specialised TPUs, including Amazon, which uses its own Trainium chips to train its AI models. "Most of the hyperscalers have their own internal programmes, and that's partly because GPUs got so expensive because the demand was outstripping supply, and it might be cheaper to design and build your own," says McIntosh-Smith. Google has been developing its TPUs for over a decade, but it has mostly been using these chips for its own AI models. What appears to be changing now is that other large companies, like Meta and Anthropic, are making sizeable purchases of computing power from Google's TPUs. "What we haven't heard about is big customers switching, and maybe that's what's starting to happen now," says McIntosh-Smith. "They've matured enough and there's enough of them." As well as creating more choice for the large companies, it could make good financial sense for them to diversify, he says. "It might even be that that means you get a better deal from Nvidia in the future," he says.

Google is relying on its own chips for its AI system Gemini. Here's why that's a seismic change for the industry

University of Portsmouth provides funding as a member of The Conversation UK. For many years, the US company Nvidia shaped the foundations of modern artificial intelligence. Its graphics processing units (GPUs) are a specialised type of computer chip originally designed to handle the processing demands of graphics and animation. But they're also great for the repetitive calculations required by AI systems. Thus, these chips have powered the rapid rise of large language models - the technology behind AI chatbots - and they have became the familiar engine behind almost every major AI breakthrough. This hardware sat quietly in the background while most of the attention was focused on algorithms and data. Google's decision to train Gemini on its own chips, called tensor processing units (TPUs) changes that picture. It invites the industry to look directly at the machines behind the models and to reconsider assumptions that long seemed fixed. This moment matters because the scale of AI models has begun to expose the limits of general purpose chips. As models grow, the demands placed on processing systems increases to levels that make hidden inefficiencies impossible to ignore. Google's reliance on TPUs reveals an industry that is starting to understand that hardware choices are not simply technical preferences but strategic commitments that determine who can lead the next wave of AI development. Google's Gemini relies on cloud systems that simplify the challenging task of coordinating devices during large-scale training (improvement) of AI models. The design of these different chips reflects a fundamental difference in intention. Nvidia's GPUs are general purpose and flexible enough to run a wide range of tasks. TPUs were created for the narrow mathematical operations at the heart of AI models. Independent comparisons highlight that TPU v5p pods can outperform high-end Nvidia systems on workloads tuned for Google's software ecosystem. When the chip architecture, model structure and software stack align so closely, improvements in speed and efficiency become natural rather than forced. These performance characteristics also reshape how quickly teams can experiment. When hardware works in concert with the models it is designed to train, iteration becomes faster and more scalable. This matters because the ability to test ideas quickly often determines which organisations innovate first. These technical gains are only one part of the story. Training cutting-edge AI systems is expensive and requires enormous computing resources. Organisations that rely only on GPUs face high costs and increasing competition for supply. By developing and depending on its own hardware, Google gains more control over pricing, availability and long-term strategy. Analysts have noted that this internal approach positions Google with lower operational costs while reducing dependence on external suppliers for chips. A particularly notable development came from Meta as it explored a multi-billion dollar agreement to use TPU capacity. When one of the largest consumers of GPUs evaluates a shift toward custom accelerators, it signals more than curiosity. It suggests growing recognition that relying on a single supplier may no longer be the safest or most efficient strategy in an industry where hardware availability shapes competitiveness. These moves also raise questions about how cloud providers will position themselves. If TPUs become more widely available through Google's cloud services, the rest of the market may gain access to hardware that was once considered proprietary. The ripple effects could reshape the economics of AI training far beyond Google's internal research. What This Means for Nvidia Financial markets reacted quickly to the news. Nvidia's stock fell as investors weighed the potential for cloud providers to split their hardware needs across more than one supplier. Even if TPUs do not replace GPUs entirely, their presence introduces competition that may influence pricing and development timelines. The existence of credible alternatives pressures Nvidia to move faster, refine its offerings and appeal to customers who now see more than one viable path forward. Even so, Nvidia retains a strong position. Many organisations depend heavily on CUDA (a computing platform and programming model developed by NVidia) and the large ecosystem of tools and workflows built around it. Moving away from that environment requires significant engineering effort and may not be feasible for many teams. GPUs continue to offer unmatched flexibility for diverse workloads and will remain essential in many contexts. However, the conversation around hardware has begun to shift. Companies building cutting-edge AI models are increasingly interested in specialised chips tuned to their exact needs. As models grow larger and more complex, organisations want greater control over the systems that support them. The idea that one chip family can meet every requirement is becoming harder to justify. Google's commitment to TPUs for Gemini illustrates this shift clearly. It shows that custom chips can train world-class AI models and that hardware purpose-built for AI is becoming central to future progress. It also makes visible the growing diversification of AI infrastructure. Nvidia remains dominant, but it now shares the field with alternatives that are increasingly capable of shaping the direction of AI development. The foundations of AI are becoming more varied and more competitive. Performance gains will come not only from new model architectures but from the hardware designed to support them. Google's TPU strategy marks the beginning of a new phase in which the path forward will be defined by a wider range of chips and by the organisations willing to rethink the assumptions that once held the industry together.

Alphabet's AI Chips Are a Potential $900 Billion 'Secret Sauce'

Alphabet Inc. investors are growing increasingly confident that the company's semiconductors could represent a significant driver of future revenue for Google's parent. The success of Alphabet's tensor processing unit, or TPU, chips is a primary reason for the stock's 31% fourth-quarter rally, which is the tenth best performance in the S&P 500 Index. The TPUs were always seen as a major strength internally, accelerating growth for the company's cloud-computing business. But there's rising optimism that Alphabet could start selling the chips to third parties, creating a new revenue stream that could ultimately be worth almost a trillion dollars.

Massive Meta-Google AI chip deal could flip the data-center market

Data-center operators face rising component costs across multiple hardware categories Meta is reported to be in advanced discussions to secure large quantities of Google's custom AI hardware for future development work. The negotiations revolve around renting Google Cloud Tensor Processing Units (TPUs) during 2026 and transitioning to direct purchases in 2027. This is a shift for both companies, as Google has historically limited its TPUs to internal workloads while Meta has relied on a wide mix of CPUs and GPUs sourced from multiple vendors. Meta is also exploring broader hardware options, including interest in RISC-V-based processors from Rivos, suggesting a wider move to diversify its compute base. The possibility of a multibillion-dollar agreement caused immediate market changes, with Alphabet's valuation climbing sharply, putting it close to the $4 trillion mark, while Meta also saw its stock rise following the reports. Nvidia's stock declined by several percentage points as investors speculated about the long-term effect of major cloud providers shifting their spending to alternative architectures. Estimates from Google Cloud executives suggest a successful deal could allow Google to capture a meaningful share of Nvidia's data-center revenue, which exceeds $50 billion in a single quarter this year. The scale of demand for AI tools has created intense competition for supply, raising questions about how new hardware partnerships could influence sector stability. Even if the deal proceeds as planned, it will enter a market that remains constrained by limited fabrication capacity and aggressive deployment timelines. Data center operators continue to report shortages in GPUs and memory modules, with prices projected to rise through next year. The rapid expansion of AI infrastructure has strained logistics chains for every major component, and current trends suggest that procurement pressures may intensify as companies race to secure long-term hardware commitments. These factors create uncertainty around the actual impact of the deal, since the broader supply environment may limit production volume regardless of financial investment. Analysts caution that the future performance of any of these architectures remains unclear. Google maintains an annual release schedule for its TPUs, while Nvidia continues to iterate on its own designs with equal speed. The competitive landscape may shift again before Meta receives its first large shipment of hardware. There is also the question of whether alternative designs can offer longer operational value than existing GPUs. The rapid evolution of AI workloads means device relevance can change dramatically, and these dynamics show why companies continue to diversify their compute strategies and explore multiple architectures. Via Tom's Hardware

Alphabet's AI chips are a potential $900 billion 'secret sauce' | Fortune

Alphabet Inc. investors are growing increasingly confident that the company's semiconductors could represent a significant driver of future revenue for Google's parent. The success of Alphabet's tensor processing unit, or TPU, chips is a primary reason for the stock's 30% fourth-quarter rally, putting it among the best performers in the S&P 500 Index. The TPUs were always seen as a major strength internally, accelerating growth for the company's cloud-computing business. But there's rising optimism that Alphabet could start selling the chips to third parties, creating a new revenue stream in a business that could ultimately be worth almost a trillion dollars. Shares were down 1% on Thursday. "If companies want to diversify away from Nvidia, TPUs are a good way to do it, and that means there's a lot of reason to be optimistic," said Gil Luria, head of technology research at DA Davidson. "The chip business could ultimately be worth more than Google Cloud. But even if it never sells a chip externally, the better chip means a better, more efficient cloud." Should Alphabet get serious about selling its TPUs, Luria estimates they could capture 20% of the artificial intelligence market over a few years, which would make it a roughly $900 billion business. Alphabet did not return requests for a comment. A Nvidia spokesperson pointed to a recent comment by CEO Jensen Huang about the company's competitive advantage: "As a company, you're competing against teams. And there just aren't that many teams in the world who are extraordinary at building these incredibly complicated things." In late October, Alphabet announced that it will supply tens of billions of dollars of chips to Anthropic PBC, which sent the stock on two-day rally of more than 6%. Then a month later, the Information reported that Meta Platforms Inc. is in talks to spend billions for access to TPUs, sparking another leap. TPUs are application-specific integrated circuit, or ASIC, chips. By definition, they're custom designed for a particular use, in this case to accelerate machine learning workloads. That makes them less flexible than the semiconductors made by Nvidia Corp., but it also makes them cheaper, which is a real benefit at a time when investors are questioning AI-related spending. "Nvidia chips are much more costly and hard to get, but if you can use an ASIC chip, Alphabet is right there, and it leads that market by far," said Mark Iong, equity portfolio manager at Homestead Advisers. "It won't control the entire market, but this is part of the secret sauce for the stock." The value of Alphabet's TPUs was corroborated by the launch of the company's latest Gemini AI model, which received glowing reviews and is optimized to run on the chips. "Alphabet is the only company with leadership in every layer of AI," Iong said, pointing to Gemini, Google Cloud, the TPUs and several other areas. "That gives it an incredible advantage." While it may be in Alphabet's best interest to sell the chips to third parties, it remains unclear how focused it is in doing so, Iong said. However, Morgan Stanley analyst Brian Nowak sees signs of a "budding TPU sales strategy" that could ultimately drive revenue. Nowak cited Morgan Stanley's Asia semiconductor analyst, who expects about five million TPUs to be bought in 2027, up roughly 67% from previous estimates, and seven million in 2028, 120% above prior estimates. While most of this will likely come from Alphabet's first-party use and Google Cloud Platform sales, it "speaks to the potential for GOOGL to sell more TPUs," he wrote in a note to clients on Dec. 1. Every 500,000 TPU chips sold to a third party data center could add about $13 billion to Alphabet's 2027 revenue, and 40 cents to its earnings per share, according to Morgan Stanley's estimates. Alphabet is expected to post roughly $447 billion in revenue in 2027, based on analyst projections, so adding $13 billion would boost sales by almost 3%. Consensus estimates for the company's 2027 revenue have risen by more than 6% over the past three months, according to data compiled by Bloomberg. Of course, high hopes for Alphabet's chips business could lead to disappointment if the upside fails to materialize, particularly with the stock's climbing valuation. The shares are trading at around 27 times estimated earnings, near the highest since 2021 and well above their 10-year average. But even at that level, Alphabet is still cheaper than Big Tech rivals like Apple Inc., Microsoft Inc. and Broadcom Inc. Allen Bond, portfolio manager at Jensen Investment Management, recently used the stock's rally to sell some of his stake. However, he remains positive on the company's overall position and prospects, including "a credible path to the TPUs becoming a driver of revenue." "Alphabet is showing tangible strength and progress with AI, and while that's increasingly appreciated by investors, the valuation still looks reasonable given growth expectations," he said. "The fact that we have increased evidence of AI momentum at a company that trades at a discount to Microsoft and Apple means it remains a core holding."

Could a new generation of dedicated AI chips burst Nvidia's bubble and do for AI GPUs what ASICs did for crypto mining?

A Chinese startup founded by a former Google engineer claims to have created a new ultra-efficient and relatively low cost AI chip using older manufacturing techniques. Meanwhile, Google itself is now reportedly considering whether to make its own specialised AI chips available to buy. Together, these chips could represent the start of a new processing paradigm which could do for the AI industry what ASICs did for bitcoin mining. In both cases, we're talking about chips broadly known as TPUs or tensor processing units. They're a kind of ASIC, or application-specific integrated circuit, that's designed at hardware level to do a very narrow set of AI-related tasks very well. By contrast, Nvidia's GPUs, even its AI-specific GPUs as opposed to its gaming GPUs, are more general purpose in nature. The parallel here is bitcoin mining, which used to be done on graphics cards before a new generation of dedicated mining chips, also ASICs, came along and did the same job much more efficiently. Could the same transition from more general purpose GPUs to ASICs hit the AI industry? The Chinese startup is Zhonghao Xinying. Its Ghana chip is claimed to offer 1.5 times the performance of Nvidia's A100 AI GPU while reducing power consumption by 75%. And it does that courtesy of a domestic Chinese chip manufacturing process that the company says is "an order of magnitude lower than that of leading overseas GPU chips." By "an order or magnitude lower," the assumption is that means well behind in technological terms given China's home-grown chip manufacturing is probably a couple of generations behind the best that TSMC in Taiwan can offer and behind even what the likes of Intel and Samsung can offer, too. While the A100 is an old AI GPU dating from 2020 and thus much slower than Nvidia's latest Blackwell GPUs, it's the efficiency and cost gains of the new Chinese ASIC that will be most appealing to the AI industry. As for Google, it has been making TPUs since 2017. But the more recent explosion in demand for AI hardware is reportedly enough to make Google consider selling TPUs to customers as opposed to merely hiring out access. According to a report on The Information (via Yahoo Finance), Google is in talks with with several customers, including a possible multi-billion-dollar deal with Meta, and has ambitions to "capture" 10% of Nvidia's AI revenue. Google's TPUs are also more narrowly defined and likely more power efficient than Nvidia's AI GPUs. Of course, there are plenty of barriers to the adoption of these new AI ASICs. Much of the AI industry is heavily invested in Nvidia hardware and software tools. Moving to a totally new ASIC-based platform would mean plenty of short term pain. But what with Nvidia said to be charging $45,000 to $50,000 per B200 GPU, companies arguably have every reason to suck up that pain for long term gains. Moreover, if more efficient ASICs do make inroads into the AI market, that could reduce demand for the most advanced manufacturing processes. And that would be good news for all kinds of chips, including GPUs for PC gaming. Right now, we're all effectively paying more for our graphics cards because of the incredible demand for large, general purpose AI GPUs built on the same silicon as gaming GPUs. If at least a chunk of the AI market shifted to this new ASIC paradigm, well, the whole supply-and-demand equation would shift. Heck, maybe even Nvidia would view gaming as important, again. We can but hope!

What's Going On With Alphabet Stock Thursday? - Alphabet (NASDAQ:GOOG), Alphabet (NASDAQ:GOOGL)

Investors are increasingly eyeing Alphabet Inc.'s (NASDAQ:GOOGL) (NASDAQ:GOOG) in-house artificial intelligence chips as a potential breakout business, betting that the company's tensor processing units could evolve from a behind-the-scenes strength into a major new revenue engine. Wall Street is increasingly betting that tensor processing unit (TPU) chips could become a major revenue driver for Google's parent company. The TPU's internal success has already fueled a 31% rally in Alphabet stock in the fourth quarter, ranking it among the top performers in the S&P 500. Also Read: Broadcom's New Google Chips Could Be 40% Cheaper To Run Than Nvidia's, Analyst Says Analysts highlight that selling TPUs to third parties could open a nearly $1 trillion market opportunity, while even internal use strengthens Google Cloud by boosting efficiency and AI performance. Wall Street sees Alphabet's chips as a strategic advantage. Alternative to Nvidia Gil Luria of DA Davidson told Bloomberg TPUs offer a viable alternative for companies seeking to diversify from Nvidia Corp (NASDAQ:NVDA), noting the business could eventually surpass Google Cloud in value. Morgan Stanley projects strong TPU adoption, estimating five million units sold in 2027 and seven million in 2028, adding roughly $13 billion to Alphabet's revenue. Alphabet has already made moves signaling potential external TPU sales, including a multibillion-dollar supply agreement with Anthropic PBC and reported talks with Meta Platforms Inc (NASDAQ:META). Meta's Strategic Shift Wellington-Altus Private Wealth's James E. Thorne argued that Meta's reported shift toward Google's TPUs reflects a tactical response to Nvidia's supply constraints, not a weakening of Nvidia's dominance. He says hyperscalers are using TPUs as a cost-effective hedge because Nvidia's Blackwell and Rubin graphics processing units face long wait times, but stresses that high switching costs and CUDA-related software friction prevent any broad industry move away from Nvidia. Thorne adds that TPUs provide extra capacity, not a true replacement, and calls the market sell-off in Nvidia a predictable "bearish hit" in an overheated environment. Tightening Competition in AI Chip Production SemiAnalysis founder Dylan Patel says competition is tightening as Google and Amazon.com Inc. (NASDAQ:AMZN) ramp up custom AI chip production, putting fresh pressure on Nvidia. He noted Google's TPUs are running at full capacity and argues that selling them directly, rather than limiting them to cloud rentals, could unlock massive market value and become Nvidia's biggest long-term threat. Patel added that the future balance of power will hinge on where AI development concentrates: custom silicon gains ground when a few tech giants dominate workloads, while broader industry demand still favors Nvidia's general-purpose GPUs. Price Action: GOOGL stock is up 0.65% at $321.70 premarket at last check on Thursday. Read Next: Palantir Targets America's AI Energy 'Bottleneck' With Chain Reaction And Nvidia Partnership Image via Shutterstock GOOGAlphabet Inc$321.620.31%OverviewGOOGLAlphabet Inc$321.000.43%AMZNAmazon.com Inc$232.760.16%METAMeta Platforms Inc$645.890.98%NVDANVIDIA Corp$180.390.45%Market News and Data brought to you by Benzinga APIs

NVIDIA's Partners Are Beginning to Tilt Toward Google's TPU Ecosystem, with Foxconn Reportedly Securing TPU Rack Orders

Foxconn, one of NVIDIA's largest supply chain partners, has reportedly received orders for AI clusters around Google's TPUs, marking a significant shift for the Taiwanese manufacturer. There's no doubt that the buzz around ASICs, especially after the release of Google's latest Ironwood TPU platform, has become increasingly mainstream. More importantly, Google's TPUs are rumored to be on the verge of adoption among several companies, with a notable name being Meta. This is why TPUs are evolving into a platform that is now targeting external adoption. According to a report by the Taiwan Economic Daily, Foxconn has received orders for Google's TPU compute trays and will also collaborate on Google's 'Intrinsic' robotics plans. Industry insiders say that Google's AI servers, built with its self-developed TPUs, are mainly divided into two racks: one rack for TPUs and the other for computing trays. According to Google's proposed supply chain, for every rack of TPUs shipped, Foxconn will ship one rack of computing trays, resulting in a 1:1 supply ratio. - Taiwan Economic Daily Well, if you are unaware, Google's 7th-generation TPUs aren't just limited to a chip configuration; rather, the firm has scalable 'rack' infrastructure in place, which it calls the 'Superpod'. It offers 9,216 chips per pod, resulting in a cumulative performance of 42.5 exaFLOPS in aggregate FP8 compute workloads, with the InterChip Interconnect (ICI). Google utilizes a 3D Torus layout for its TPUs, which enables high-density interconnect across large numbers of chips. The report claims that Foxconn will be responsible for producing computing trays for Google, based on the TPU rack orders it receives. Since inferencing is being a lot more dominant within AI workloads, companies are rushing to restructure their compute portfolio to get the best inference performance, within optimal TCOs, which is why Google's TPUs are being touted as the leading candidate in the AI application phase. Of course, this raises a whole debate about whether NVIDIA could be replaced by custom silicon from Big Tech, but for now, it's essential to note that the supply chain is witnessing immense interest in TPU solutions from Google.

Is Alphabet Really a Threat to Nvidia's AI Chip Dominance? | The Motley Fool

Alphabet's decade-long bet on custom silicon is finally paying off. Nvidia (NVDA 1.03%) looks unstoppable. The company has just posted $57 billion in quarterly revenue, with its data center business growing at a 66% annual rate. CEO Jensen Huang also discussed $500 billion in chip demand visibility through 2026. With a market share of around 90% in artificial intelligence (AI) accelerators, Nvidia has become the default infrastructure provider for the generative AI era. But Alphabet (GOOGL +1.26%) (GOOG +1.44%) has been quietly building an alternative. And it's starting to matter. Alphabet began designing its own AI chips in 2013 -- years before ChatGPT made "AI" a household term. The Tensor Processing Unit (TPU) originated as an internal project designed to meet the computational demands of Google's Search and Translate services. Today, it has evolved into a commercial platform that directly competes with Nvidia's data center GPUs. The latest generation, TPU v7 Ironwood, closely matches Nvidia's flagship Blackwell chips in raw compute power, as demonstrated in published benchmarks, while offering advantages in system-level efficiency for specific workloads. More importantly, Google Cloud now makes these chips available to external customers -- and some of the biggest names in AI are taking notice. Nine of the top 10 AI labs now use Google Cloud infrastructure. Apple trained its foundation models for Apple Intelligence on clusters of 8,192 Google TPU v4 chips -- not Nvidia GPUs. Anthropic, the company behind Claude, recently secured access to up to 1 million Google TPUs through a multibillion-dollar partnership. Reports suggest that Meta Platforms is in talks to deploy Alphabet's TPUs alongside its own custom silicon as early as 2027. These high-profile deployments are significant because they demonstrate that the TPU platform is effective at scale. If Apple -- arguably the most demanding engineering organization in tech -- chose Alphabet's chips for its flagship AI initiative, the technology is enterprise-ready. The real threat to Nvidia isn't in training frontier models. That market requires the raw horsepower and flexibility that Nvidia's GPUs excel at. The threat is in inference -- actually running those models to serve billions of users. Training is a capital expenditure. You do it once (or periodically) to create a model. Inference is an operational expenditure that runs constantly, and its costs compound as AI applications scale. By 2026, analysts expect inference revenue to surpass training revenue across the industry. This is where Alphabet's vertical integration shines. Reports indicate that for certain large language model inference workloads, Google's latest TPUs can deliver up to 4 times better performance per dollar than Nvidia's H100. Midjourney, the popular AI image generator, reportedly cut its monthly inference costs by 65% after migrating from Nvidia GPUs to Google's TPU v6e pods. For AI companies burning through venture capital, those savings aren't just efficient -- they're existential. For two decades, Nvidia's real competitive advantage wasn't silicon -- it was software. The CUDA programming platform created massive switching costs. Researchers wrote code in CUDA, universities taught CUDA, and enterprises deployed on CUDA. Leaving meant rewriting everything. That moat is eroding. Modern machine learning frameworks, such as PyTorch and JAX, increasingly abstract away the underlying hardware, allowing for more efficient and scalable computations. With PyTorch/XLA, developers can now run standard PyTorch models on TPUs with minimal code changes. That reduces the friction that once locked customers into Nvidia's ecosystem, even though CUDA still retains a larger and more mature developer community overall. This doesn't mean CUDA is irrelevant. But it does mean customers can now evaluate chips primarily on price and performance rather than software compatibility -- a shift that favors Alphabet's cost-optimized approach. Nvidia isn't going anywhere. The company will likely dominate model training for years, and its financial results reflect genuine, durable demand. However, the era of unchecked pricing power may be coming to an end. The clearest evidence: According to a recent industry analysis, OpenAI secured roughly a 30% discount on its latest Nvidia hardware order by raising the credible option of shifting more workloads to alternative hardware, such as Alphabet's TPUs. Even when customers stay with Nvidia, Alphabet's presence caps what Nvidia can charge. For Nvidia shareholders, this suggests margins may face pressure as competition intensifies. For Alphabet shareholders, it highlights an underappreciated growth driver. Google Cloud revenue jumped 34% last quarter to $15.2 billion, with AI infrastructure demand -- including TPUs -- cited as a key driver. The cloud backlog surged 82% year over year to $155 billion. Alphabet won't dethrone Nvidia overnight. But it has successfully positioned the TPU as the industry's credible second option -- and in a market this large, second place is worth hundreds of billions.

Google's TPU Advantage Could Shift the AI Cloud Race | Investing.com UK

Google's (NASDAQ:GOOGL) AI chips, called Tensor Processing Units (TPUs), are getting a lot of attention. They were used to train its newest gen-AI model, Gemini 3, which has been widely praised, and they're cheaper to run than Nvidia's (NASDAQ:NVDA) Graphics Processing Units (GPUs). The real reason Google created the TPU goes back to 2013. The company ran a forecast showing that if every Android user used voice search for just three minutes a day, Google would need to double its global data-centre footprint. Not because of video or storage, but because running AI on conventional chips was too expensive. So, Google built its own AI-focused processor. Fifteen months later, TPUs were already powering Google Maps, Photos, and Translate, long before the public knew the chip existed. TPUs matter because GPUs were originally built for gaming, not AI workloads. TPUs are purpose-built for AI with no unnecessary overhead. The result is better performance per dollar, lower energy use, and faster execution for many AI tasks. Each generation also brings a major performance jump. Even Nvidia's CEO, Jensen Huang, acknowledges the quality of Google's TPU program. So why don't more companies use TPUs? Most engineers are trained on Nvidia and CUDA, and TPUs only run on Google Cloud. Switching ecosystems is costly and disruptive. From Google's perspective, TPUs give its cloud business a major advantage. While AI workloads are pressuring cloud margins across the industry due to reliance on Nvidia hardware, Google controls both the chip and the software stack. That means lower costs, better margins, faster development cycles and a defensible position competitors can't easily replicate. Some experts argue TPUs now match or exceed Nvidia's top chips. In short, Google didn't create TPUs to sell hardware. It built them to handle its own AI growth. Today, TPUs may be Google Cloud's strongest competitive asset, and if Google opens them more widely to external developers, the AI infrastructure landscape could shift quickly. Alphabet has been the biggest driver of the S&P 500 this year, responsible for 19.4 percent of the index's total year-to-date gain. That's the result of adding about $1.3trin market value in eleven months. Nvidia is next with a 16 percent contribution, followed by Broadcom at roughly $520bnof added value, and Microsoft at about $380bn. Across the index, the top ten companies account for 59.4 percent of the S&P 500's gain this year. The remaining 490 companies together contributed only 40.6 percent. Source: The Kobeissi Letter, econovisuals Google's market share in AI models jumped from 5 percent to 14 percent even before the launch of Gemini 3. The Wall Street Journal reported that "Gemini 3's surge past ChatGPT and other competitors on benchmark tests has handed Google an elusive victory". OpenAI and ChatGPT still hold a large lead, but their share has been slipping. Behind Gemini, the next contenders are DeepSeek, Grok, Perplexity, and Claude, while Microsoft's Copilot is barely visible on the chart. Source: Josh Wolfe @wolfejosh HSBC built a model to figure out if OpenAI can actually pay for all the cloud compute it's contracted. The short answer is no. Actually, not even close. HSBC projects that by 2030, consumer AI will generate $129bn in revenue, mostly from search and advertising, and OpenAI's consumer market share will fall from 71% today to 56%. Enterprise AI revenue will reach $386bn, with OpenAI's share dropping from around 50% to 37%. Despite large projected revenues, HSBC argues that OpenAI faces a major funding shortfall. It estimates the company will incur $792bn in compute-related rental costs through 2030, rising to $1.4tn by 2033. Against this, OpenAI could generate about $282bn in cumulative free cash flow plus roughly $67bn from external liquidity, debt facilities, and investments, leaving a $207bn gap, or $217bn including a cash buffer. HSBC notes the assumptions are highly uncertain: gaining an extra 500M users could add $36bn in revenue, and converting 20% of users to paid plans could bring $194bn more. Adjustments in compute costs could also change the picture. The model does not account for the possibility of a breakthrough in Artificial General Intelligence (AGI). If revenues fall short or investors grow cautious, OpenAI may need to make difficult choices. With debt markets uneasy, Microsoft's support uneven, and SoftBank as the next largest shareholder, HSBC suggests OpenAI's least-bad option might be to exit some data-centre commitments early or at renewal. OpenAI has managed one of the most unusual financial manoeuvres in the tech sector: close to $100bnin debt is being taken on by its partners rather than by OpenAI itself. As the company accelerates its push toward AGI, firms such as SoftBank, Oracle (NYSE:ORCL), CoreWeave (NASDAQ:CRWV), and Blue Owl (NYSE:OWL) are borrowing massive amounts to build the computing infrastructure OpenAI relies on. More than $30bn has already been raised, another $28bnis tied to OpenAI-related agreements, and roughly $38bnin new financing is expected. In total, the debt connected to OpenAI approaches $100bn, while the company itself holds almost none of it. As one senior executive put it: "how does OpenAI leverage other people's balance sheets?" October recorded the largest US budget deficit on record at $284bn, and it happened without any underlying crisis. In 2020, huge deficits were justified by the pandemic. In 2025, with a functioning economy, the imbalance is harder to defend. Roughly 24 percent of every tax dollar now goes directly to interest payments. The government collects about $404bn a month, and around $100bndisappears before funding any actual programs. October's interest bill was $104bn, the highest ever. The driver is simple: interest rates rose from about 1.56 percent to 3.4 percent, and the government is issuing more debt at these higher costs. The trend is deteriorating. Interest is already the third-largest federal expense and could reach $1.8tr year within a decade. The cycle is clear: interest consumes more of the budget, the government borrows even more, and the interest burden grows further. The US is spending as if it were in a crisis, and the risk is that this behaviour ends up creating one. Polymarket data shows Kevin Hassett emerging as the frontrunner in the search for Trump's next Fed Chair, marking the first time any candidate has traded above 50 percent. Christopher Waller is currently the second-most likely pick.