Nvidia CEO Jensen Huang met with both Republican and Democratic lawmakers on Capitol Hill in late July 2026 to discuss what he framed as American leadership in artificial intelligence, touching on open-source AI policy and national security alongside Nvidia’s own manufacturing commitments. Huang pointed specifically to Nvidia’s pledge to produce up to $500 billion worth of AI infrastructure domestically in the United States — a figure he’s used repeatedly this year to position Nvidia’s own supply chain decisions as directly tied to the broader US-China competition over AI capability, rather than a purely commercial manufacturing choice.
The Capitol Hill visit came alongside an increasingly bullish public forecast from Huang about the scale of AI infrastructure spending still ahead. Speaking at Nvidia’s GTC conference earlier this year, Huang projected the company would collect $1 trillion in chip sales through 2027 — double an earlier $500 billion forecast for the same period through 2026 — and has separately projected global data center capital expenditure could reach $4 trillion by 2030, roughly a fivefold increase from current levels. Those numbers have become a genuine talking point among investors and analysts debating whether Nvidia’s market capitalization, which peaked near $5 trillion earlier this year before pulling back, could climb back to that level if Huang’s projections hold.
A New Bottleneck: Memory, Not Compute
In late July 2026, Huang made a notable shift in emphasis, telling investors and industry observers that memory chips — not raw GPU compute power, the metric that has dominated AI infrastructure discussion for years — are now the industry’s biggest practical constraint. The distinction matters technically: modern AI models are increasingly limited not by how fast a chip can perform calculations, but by how quickly data can move in and out of memory to feed those calculations, meaning memory supply and bandwidth are becoming as commercially significant to the AI buildout as GPU availability itself. It’s a real shift in where Nvidia — and its customers building AI infrastructure — will likely be looking for the industry’s next real supply constraint.
The Scale Behind Huang’s Public Confidence
Huang’s public bullishness is backed by real, current numbers. Nvidia’s market capitalization stood at roughly $4.77 trillion in late July 2026, making it the world’s most valuable publicly traded company, with Huang telling a podcast audience he sees a realistic path to a $10 trillion valuation over time. The company’s own supply commitments — contracted orders for future chip production — now stand at $119 billion, which Huang has described as part of “the largest infrastructure expansion in human history.” Part of that expansion is explicitly international: Huang signed Toyota, Fanuc, Kioxia, and other major Japanese industrial firms into Nvidia’s new Physical AI Coalition this year, a push to embed Nvidia’s chips and software into robotics and industrial automation specifically, not just data-center AI. Nvidia reports its next quarterly results on August 26, 2026, with analysts expecting revenue near $91-92 billion for the quarter — nearly double what the company reported in the same quarter a year earlier. Nvidia shares themselves have been comparatively muted for a company posting these kinds of figures, up roughly 11% year-to-date as of early August 2026 and trading around $190-211 a share through late July and early August — a smaller gain than the underlying revenue growth alone might suggest, reflecting how much of Nvidia’s continued AI-boom story is now already priced into the stock ahead of each individual earnings report, and how sensitive that price has become to any single data point that might cast doubt on the pace of the broader buildout Huang keeps describing in increasingly ambitious terms.
What This Means for Philippine Founders
Huang’s memory-bottleneck comments are a concrete, practical signal for any Philippine startup whose product roadmap depends on continued AI compute cost declines: if memory genuinely becomes the binding constraint on global AI infrastructure, the cost curve founders have gotten used to — compute getting cheaper and more available every year — may not move as predictably or as fast on the memory side, and product plans that assume unlimited, ever-cheaper inference capacity are worth stress-testing against that possibility now rather than after a real supply crunch hits. Huang’s Capitol Hill visit is also a reminder that the chip supply chain Philippine AI startups ultimately depend on — however indirectly, through whichever cloud provider or model API they use — is increasingly shaped by US industrial and national-security policy decisions being actively negotiated in Washington this year, not settled market dynamics a founder can simply assume will hold steady. It is also worth remembering that Nvidia’s own forecasts, however data-backed, are still projections from the company with the single largest financial stake in the AI buildout continuing at its current pace — a useful data point for any founder’s own planning, but not a substitute for a founder’s own independent read on how fast compute costs and availability are actually moving in their specific market.
Share this article