NVIDIA (NVDA) on Solana

Nvidia returned $26.0 billion to shareholders through buybacks and dividends in fiscal Q2 2027 (ended July 26), exceeding the quarter's free cash flow of $21.3 billion and running at nearly ten times its capital spending of $2.7 billion. The pace has sharply accelerated from $10 billion in the same quarter a year ago, and the first half of fiscal 2027 alone delivered $45 billion in returns — already surpassing all of fiscal 2026's $41.1 billion. CFO Colette Kress has guided for returning 50% or more of free cash flow going forward, with year-to-date returns tracking at 60%.

The scale of the commitment signals deep board confidence in the durability of current earnings. In May 2026, Nvidia's board raised the quarterly dividend 25-fold from $0.01 to $0.25 per share and authorized an additional $80 billion in repurchases with no expiration date, leaving roughly $99 billion available after Q2. Analysts interpret the multi-year dividend obligation as evidence management expects today's profit levels — built on 106% year-over-year revenue growth to $96.2 billion and net income of $59.7 billion — to hold for at least three more years.

Nvidia has entered full production of its Vera family of CPUs, moving directly into a server CPU market previously dominated by Intel and AMD. The company's Grace CPU family already generated $5 billion in trailing twelve-month sales as of fiscal Q2 2027, and Nvidia projects data center CPU revenue will more than double from this fiscal year to the next. Nvidia claims Vera delivers performance and efficiency that significantly exceed competing data center CPUs in key metrics.

The strategic logic is straightforward: Nvidia controls more than 90% of the AI training GPU market and 60–75% of AI inference, giving it a built-in distribution channel to offer CPUs optimized for tight integration with its own GPUs. As more AI workloads shift onto CPUs and operational AI applications grow, Nvidia's incentive — and its ability to bundle CPU-GPU performance — positions it to gain share that Intel and AMD have historically held without direct competition from the GPU market leader.

Piper Sandler analyst David O'Connor initiated coverage of Nvidia on September 9 with an Overweight rating and a $300 price target, implying roughly 34% upside. On his fiscal year 2028 earnings estimate, Nvidia trades at just 14x — a PEG ratio of 0.3 that O'Connor described as "among the cheapest valuations in its AI universe" — underpinned by a projected revenue and earnings CAGR of 47% through FY30, with a FY30 earnings estimate of $30 per share. The company's most recent quarter reinforced the growth case: Q2 FY27 revenue hit $96.2 billion, up 106% year-over-year, while data center revenue climbed 117% to $89 billion on a GAAP gross margin of 75%.

The analysis frames the valuation as compelling only if execution holds. O'Connor noted that any meaningful slowdown in AI infrastructure spending, product execution, or profitability would make the 14x multiple "less compelling," putting the spotlight squarely on whether Nvidia can sustain the growth trajectory embedded in Piper's forecasts rather than on the headline multiple itself.

Nvidia is targeting up to 2 gigawatts of AI computing capacity in Australia by 2027 through its DSX AI factory platform, working with local Nvidia Cloud Partners and AI infrastructure providers to secure land, power, and data center shell capacity designed to host multiple generations of Nvidia hardware. The initiative extends a pattern of large-scale regional commitments — Nvidia is simultaneously building a 140-megawatt AI factory in Japan and deploying AI supercomputers across Europe — positioning the company as a full-stack infrastructure provider rather than a chip supplier alone.

The Australia buildout is strategically significant for the Asia-Pacific region, where sovereign AI capacity has become a policy priority for governments seeking to reduce reliance on overseas cloud infrastructure. By embedding its hardware, networking, and CUDA software stack into dedicated regional facilities, Nvidia aims to lock in multi-generational demand across training and inference workloads at a time when hyperscaler and enterprise AI spending in the region is accelerating.

Michael Burry, known for his prescient bet against the 2008 housing market, continues to hold short positions in Nvidia via put options, warning of a potential "1987-type fall" for AI stocks. His central concern is structural leverage: leveraged ETFs such as the GraniteShares 2x Long NVDA Daily ETF are mechanically required to purchase more Nvidia shares as the stock climbs, which Burry argues inflates the price beyond fundamentals. He also highlighted that margin debt rose roughly 50% year-over-year through June, amplifying the potential severity of any deleveraging event.

The short thesis faces a substantial counter-case. Nvidia delivered record revenue of $96.2 billion in its most recent quarter, and CEO Jensen Huang has guided for approximately 70% revenue growth by FY2028 as the Vera Rubin chip architecture ramps into volume production. Analysts note that real enterprise and hyperscaler demand—not purely speculative flows—underlies Nvidia's growth, distinguishing the current AI buildout from the dot-com bubble. Burry's track record on macro crash calls has been mixed; a recent short on AI infrastructure company Nebius Group moved against him after the company reported strong earnings. For Nvidia investors, the key question is whether leverage-driven froth poses a near-term dislocation risk, or whether the AI hardware supercycle is durable enough to absorb a deleveraging event without a structural drawdown.

Jensen Huang projected approximately 70% annual revenue growth for Nvidia in FY2028, significantly above the 44% consensus analyst estimate at the time of the Q2 FY2027 earnings call. Huang attributed the still-constrained outlook to ongoing supply headwinds, noting that demand continues to outpace what Nvidia can ship — a dynamic he framed as a ceiling on near-term growth rather than a sign of slowing demand.

Huang described the forthcoming Vera Rubin GPU platform as expected to deliver "the fastest product ramp in the company's history," representing a $40 billion revenue opportunity per gigawatt of deployed compute, compared with $25 billion for Blackwell processors. He also projected CPU revenue to more than double in the next fiscal year, signaling Nvidia's intent to become a leading server CPU supplier alongside its established GPU and networking dominance — a meaningful expansion of the company's addressable market beyond AI accelerators.

Nvidia (NVDA) shares climbed roughly 4% on September 2 after Dell Technologies delivered a top- and bottom-line beat and raised its fiscal 2027 earnings guidance by more than 40%, reinforcing the view that enterprise AI infrastructure spending remains on an upward trajectory. Dell's results were read by markets as downstream confirmation that demand for AI compute is still accelerating, with analysts characterizing the quarter as signaling it is "full steam ahead for the AI buildout."

The move builds on Nvidia's own recent record-setting quarter, in which the company reported approximately $96 billion in revenue and guided for roughly 70% revenue growth in its next fiscal year. With Broadcom set to report Q3 results — expected to show revenue nearly doubling year-over-year to around $29.4 billion — Nvidia's performance has raised the bar for the broader semiconductor sector. Broadcom has historically traded in sympathy with Nvidia, and analysts note it now faces pressure to deliver comparable results to arrest a significant stock skid.

OneRail has launched OmniSTAR, an AI-powered delivery decisioning platform built in collaboration with NVIDIA through a multi-year engagement via the NVIDIA Inception program. The platform combines NVIDIA cuOpt—a GPU-accelerated routing and decision optimization engine—with NVIDIA cuDF for data processing and NVIDIA's accelerated compute infrastructure, layered on top of OneRail's network of more than 12 million drivers and over 1,000 logistics partners. OmniSTAR evaluates every available delivery option for each order in real-time, spanning owned fleets, couriers, parcel carriers, and other modes, and selects the lowest-cost option that meets service requirements.

The platform compresses optimization computations that previously took around 20 minutes to under two minutes—a roughly 10x speedup—giving enterprise retailers, wholesalers, and distributors the ability to respond to changing conditions and improve delivery margins at scale. OneRail positions OmniSTAR as a way for mid-market retailers to match the last-mile economics of larger players like Amazon and Walmart. US Foods is among the early customers, using the platform to surface margin-eroding delivery configurations, such as shipping low-margin items long distances on expensive equipment, and then restructuring pricing and routing accordingly.