qinbafrank
AI/Crypto/TMT Investor, Global Liquidity & Macro, Cycle-Based Investing
Nvidia's expanded buyback signals confidence: if no one else buys the stock, Nvidia will. The board added $150 billion of authorization, bringing the remaining total to $235 billion, which Nvidia expects to complete by the end of fiscal 2028, more than 4% of its shares within about a year. The scale of the program reassures the market. x.com
Nvidia's Hugging Face acquisition gives it compute supply, the main model-distribution layer and open-source mindshare at once, letting it steer new workloads onto its hardware. The more open models spread into local deployment, fine-tuning, agents and private enterprise use, the more GPU demand grows. It also hedges against closed labs' in-house chips by keeping long-tail workloads on CUDA, while AMD loses a neutral showcase. Antitrust review is the biggest risk; closing is expected in H1 2027. x.com
Walter Bloomberg
Financial News Headlines Aggregator, Not Affiliated with Bloomberg
NVIDIA IS IN TALKS TO PROVIDE NSCALE WITH BILLIONS IN CREDIT GUARANTEES FOR DEBT TO DEVELOP A 460-MEGAWATT DATA CENTER - THE INFORMATION x.com
AlvaChallengesPotential multibillion-dollar guarantees would add contingent claims on Nvidia's cash alongside its planned buyback, according to talks reported by The Information.Walter Bloomberg
Financial News Headlines Aggregator, Not Affiliated with Bloomberg
Fitch also highlights the risk that AI infrastructure spending may exceed sustainable demand, potentially triggering a sharp pullback in investment. x.com
AlvaChallengesThis weakens the thesis that Nvidia’s buyback reassures the market: Fitch warns that unsustainable AI infrastructure spending could trigger a sharp investment pullback.
Ed Ludlow
Host of Bloomberg Tech, Correspondent at Bloomberg News
Top of the log is Apple and NVIDIA. ... Top of the log, throwing off cash like nobody's business. For them, they're more worried about shareholder returns than needing to let go of anyone. podtrac.com
AlvaSupportsStrong cash generation and a focus on shareholder returns support Nvidia's capacity and willingness to carry out its large buyback.
Gary Gensler
Former SEC Chair, Professor of the Practice at MIT Sloan
So we're pushing $600 billion. And who holds it? I think it's insurance sector I worry about that's picked up a lot of holding of debt through private credit or just directly I worry about all these guarantees that Broadcom, NVIDIA, AMD have written back into these data centers. And it's off balance sheet. podtrac.com
AlvaChallengesGensler worries that Nvidia's off-balance-sheet data-center guarantees could become liabilities, putting its capacity to fund the planned buyback at risk.
Hedgie
Financial Markets Newsletter and Commentary
Nvidia lined up six major lenders for a $500 billion program and will backstop up to 25% of each deal. ... If one of these backstops ever gets called, the market will reprice all of them at once, and Nvidia’s and Broadcom’s investors will see the guarantees in the earnings for the first time. x.com
AlvaChallengesNvidia's contingent exposure to AI-chip lending weakens the thesis that its capital actions simply reassure the market; the repricing and earnings risk remains conditional rather than realized.
Damnang
Semiconductor Engineer in Silicon Valley, AI Chip Investment Research
핵심 요약 1. 삼성전자가 HBM4 가격을 업계 최고 수준으로 책정, 내년 HBM 평균판매가격은 올해보다 100에서 150퍼센트 오를 전망 x.com
AlvaChallengesThe quoted KB Securities report's forecast of sharply higher HBM prices increases the potential cost and supply pressure on Nvidia's growth outlook.
SemiAnalysis
Independent Semiconductor and AI Research Firm
ALERT🚨: Vera Rubin NVL72 is up to 7.3x better in perf per dollar than GB300 NVL72 for 1 million context length agentic inference workloads! x.com
AlvaSupportsThe reported performance-per-dollar gain strengthens the case that Rubin could improve Nvidia's position in agentic inference, specifically for million-token contexts.
Jim Cramer
Host of CNBC’s Mad Money, Leader of the CNBC Investing Club
The next big test for the AI data center complex will be hyperscaler earnings reports. ... if you're hoping for resurgence in the AI data center cohort, these companies need to maintain their gigantic capital spending. That's right. Need to maintain, not cut. ... As long as they don't blink though, I think we can get through this, especially if they put up strong cloud infrastructure numbers. dts.podtrac.com
AlvaChallengesHyperscaler earnings are an upcoming test of the spending behind Nvidia's growth outlook: Cramer says cuts would threaten the data-center case, while sustained spending and strong cloud results would help it hold.Jim Cramer
Host of CNBC’s Mad Money, Leader of the CNBC Investing Club
Right now, some of their publicly traded debt is going for 15%, and that's just too expensive for them to raise more capital. The return on the investment is simply not big enough for CoreWeave to make money if they're borrowing at 15% to buy NVIDIA GPUs. ... But I actually think that the chips they buy will more than hold their value. The whole world's desperate for high end GPUs. dts.podtrac.com
AlvaMixedCoreWeave’s roughly 15% borrowing cost raises a financing risk for AI data-center expansion and, in turn, Nvidia GPU demand; Cramer also expects strong demand for high-end GPUs to preserve the chips’ collateral value.Jim Cramer
Host of CNBC’s Mad Money, Leader of the CNBC Investing Club
If the yield of the thirty year treasury goes north of 6%, it's currently at 5.47, then look out. If we get another rate hike soon, then I believe Core Web will have a difficult time raising money, and the dominoes will indeed fall. dts.podtrac.com
AlvaChallengesCramer's rate-hike scenario adds a specific financing risk to the data-center buildout Nvidia's growth outlook depends on.
Dan Ives
Technology Analyst & Partner at Yorkville Ives
And if you look at the demand and you look at ultimately capacity and where it's going to go, even if you haircut some of the data centers that don't ultimately get billed out, I still think streets underestimate numbers massively over the next one, two years. podtrac.com
AlvaSupportsIves expects AI demand and capacity-related numbers to exceed market estimates over the next one to two years, supporting the thesis's upside case for compute demand even after allowing for some data centers not being built out.Dan Ives
Technology Analyst & Partner at Yorkville Ives
Yeah, no doubt nervousness, but demand supply 13 to 1 for memory chips. ... Acceleration is happening in terms of demand on chips, hyperscalers, use cases. podtrac.com
AlvaSupportsThe Korea observation of a severe memory-chip supply imbalance and Ives’s view that demand for chips, hyperscalers, and use cases is accelerating support the thesis that Nvidia demand could sustain stronger growth.
Mandeep Singh
Global Head of Technology Research at Bloomberg Intelligence, Host of Tech Disruptors
So we know the Frontier Labs have grown their revenue fivefold from the start of the year to where we are now. But even with Anthropic, they had stellar growth for the first nine months. Well, now we are starting to see maybe that monthly ARR growth is slowing down. ... Is it because of open rate models or token pricing coming down? So it impacts the ROI and what these companies have to show for all the capex that they ... are doing. ... even though Anthropic is facing pressure from token pricing going down, they still need to run their workloads on Amazon and Google infrastructure and they're not cutting their prices. podtrac.com
AlvaChallengesPossible slowing in frontier-lab revenue growth raises a return-on-investment risk for AI spending, while continued workloads and firm hyperscaler pricing point to resilient infrastructure demand.
Matt Garman
CEO of Amazon Web Services
We're open sourcing TorchNeuron. Your existing @PyTorch code runs on Trainium with one config change, and we're contributing upstream to make PyTorch more hardware-agnostic for everyone. x.com
AlvaChallengesReducing the friction of running existing PyTorch code on Trainium could strengthen AWS’s own-chip competition and challenge the expectation that AWS’s GPU expansion will translate into Nvidia workloads.SemiAnalysis
Independent Semiconductor and AI Research Firm
Unfortunately for NVIDIA, its first GB300 FP4 counter-submission is still 1.64x worse in performance per dollar than TPUv7 for external customers across most of the end-to-end performance curve. ... We expect more counter-submissions from NVIDIA as they continue trying to narrow the difference. x.com
AlvaChallengesTPUv7’s reported 1.64x performance-per-dollar advantage over NVIDIA’s first GB300 FP4 counter-submission for external customers weakens the thesis’s competitive growth outlook.
Andy Jassy
President and CEO of Amazon.com, Founder of Amazon Web Services
TorchNeuron is going to largely take that off the table, enabling your existing PyTorch code to run natively on Trainium with one configuration change and no rewrite. ... With Trainium’s very appealing price-performance characteristics, TorchNeuron is going to make it even easier for more companies to use AI more cost-effectively. x.com
AlvaChallengesTrainium’s lower switching costs and price-performance could make it a more accessible alternative to Nvidia, challenging the thesis that AI demand will flow to Nvidia’s infrastructure.
Dylan Patel
Founder & Chief Analyst of SemiAnalysis, AI Infrastructure Research
[Replying to @rev_cap: "Semianalysis saying if you’re worried about demand (ARR,) don’t be, because there’s going to be even more supply than you realize"] Demand and ARR is what to worry about, not supply. x.com
AlvaChallengesThis challenges the thesis's emphasis on supply constraints by identifying demand and ARR as the risks to watch for Nvidia's growth outlook.Damnang
Semiconductor Engineer in Silicon Valley, AI Chip Investment Research
Standardizing base die and interface implementations around NVIDIA’s own requirements can reduce the qualification burden across suppliers. Once alternative suppliers have passed qualification and secured sufficient capacity, NVIDIA can shift orders more easily. Memory companies will have to compete on price to win those orders. ... NVHBM will ultimately increase downward pressure on HBM prices. x.com
AlvaSupportsStandardized HBM base dies could ease supplier qualification and let Nvidia shift orders once alternatives qualify and secure capacity, reducing supply constraints and strengthening its leverage on memory prices.Damnang
Semiconductor Engineer in Silicon Valley, AI Chip Investment Research
NVHBM will ultimately increase downward pressure on HBM prices. x.com
AlvaSupportsThe forecast of greater downward pressure on HBM prices supports the possibility of easing a supply-related cost constraint for Nvidia.Damnang
Semiconductor Engineer in Silicon Valley, AI Chip Investment Research
According to the report, the global KV cache footprint falls to roughly one quarter of that of the previous Flash model, while the persistent KV cache footprint falls to roughly one eighth. These figures refer to KV caches. They do not mean that total memory requirements, including model weights, fall by the same proportions. ... I do not interpret this as a structural negative for overall HBM demand. Lower memory requirements per task create room for longer tasks and more agents running simultaneously. ... My expectation is that, over time, usage growth driven by lower costs will more than offset those savings. x.com
AlvaSupportsLower KV-cache requirements per task could enable longer workloads and more parallel agents, supporting the thesis that agent workloads may expand compute demand; Damnang expects usage growth from lower costs to more than offset the memory savings.