Dylan Patel
About
Founder of SemiAnalysis, semiconductor and AI hardware research firm
Cast within
No topic-region cast yet — this appears once Dylan Patel's compiled claims are aligned into a topic region's argument tree.
Claims by Dylan Patel (20 of 118)
Software-as-a-service underperformed due to AI COGS
SaaS companies have historically had very low incremental cost per user (high R&D and customer acquisition costs but cheap to serve), but the high COGS of AI completely breaks these business models, which is why SaaS companies (excluding Microsoft) have underperformed in the markets.
AI migration of legacy systems is the near-term value driver
The most immediately valuable AI use is migration—converting mainframe systems to cloud, Excel databases into real SQL databases, and Word/Excel artifacts into more programmatic, efficient forms—which could shrink the Office ecosystem's relative use, though even mainframes have continued growing for two decades despite such migrations.
Anthropic inference gross margins expanded despite competition
Over the course of the year, Anthropic's gross margins on inference went from well below 40% to north of 60%, expanding significantly at the model layer despite intensifying competition from Chinese open-source models, OpenAI, Google, and X/Grok.
Talent wars in AI are capital intensive
Building a leading AI lab is a capital-intensive talent game—evidenced by Meta spending north of $20 billion on talent and Anthropic and Meta poaching reasoning/post-training teams from Google—so a company spending $100 billion on infrastructure should commensurately spend on the people who make breakthroughs, keeping researcher-to-GPU ratios high.
Each technological revolution diffuses faster than the last
Each major technological transition—railroads, the internet, replaceable parts, industrialization, the cloud—has gotten faster in the time from discovery to economy-wide pervasiveness, and AI's three-year ramp to hyperscalers spending $500 billion of capex next year is unmatched in speed relative to prior revolutions.
OpenAI's aggressive deals beat Anthropic's caution
Anthropic was conservative on compute commitments to avoid bankruptcy risk, while OpenAI signed aggressive deals across many providers (CoreWeave, Oracle, SoftBank Energy, NScale); as a result OpenAI has far more compute access by year end, and Anthropic now must use lower-quality providers or pay revenue-share markups to catch up.
Nvidia secured TSMC capacity by ordering earliest
Nvidia is getting the majority of TSMC 3nm supply not by special favor but because it sent the strongest, earliest market signal—non-cancelable, non-returnable orders with deposits—far ahead of Google and Amazon, whose chips also faced delays, while TSMC independently verified downstream supply (PCB, memory) could support Nvidia's volumes.
ASML never raises price above capability gains
ASML is unusually generous: despite a monopoly on EUV with no competitor close, it has never raised tool prices more than it has increased capability (throughput, overlay accuracy)—tools went from ~$150M to ~$400M while capabilities more than doubled—always delivering net benefit to customers, unlike Nvidia or memory vendors who take the available margin.
700 EUV tools by 2030 enable ~200 GW of AI chips
TSMC's ecosystem already has ~250-300 EUV tools; adding ~70/80/100 per year reaches ~700 by decade's end, which at 3.5 tools per gigawatt could support ~200 gigawatts of AI chips per year—making Sam Altman's 52 GW/year goal a reasonable ~25% share, though some capacity still goes to mobile and PC.
Bandwidth per edge area, not bits per wafer, is the key metric
Switching accelerators from HBM to commodity DDR to save wafers fails because chip I/O escapes only on the edges; an HBM4 stack delivers ~2.5 TB/s per ~13mm shoreline versus only ~64-128 GB/s for DDR in the same area—an order of magnitude less bandwidth—and since FLOPS are gated by feeding weights and KV cache, the metric that matters is bandwidth per wafer edge, not bits per wafer.
Roughly 3.5 EUV tools support one gigawatt of AI
A gigawatt of Nvidia Rubin capacity needs ~55,000 wafers of 3nm (with ~20 EUV passes each, ~1.1M passes), plus 5nm and 170,000 DRAM wafers, totaling ~2 million EUV passes; at 75 wafers/hour and ~90% uptime, that requires about 3.5 EUV tools (~$1.2B) to underpin ~$50B of data-center CapEx and ~$100B of AI value.
Power is solvable through many inefficient sources
Power will not be the binding US constraint because beyond the three combined-cycle gas turbine makers there are 16+ gas power-gen vendors plus aeroderivatives, medium-speed reciprocating engines, ship engines, fuel cells (Bloom), and solar-plus-battery; even doubling power cost to ~$3,500/kW only raises a GPU's TCO a few cents/hour, and utility-scale batteries could unlock ~20% of the terawatt-scale US grid that sits idle outside peak hours.
Space GPUs delay deployment of the scarcest resource
Space data centers don't make sense this decade because they don't escape the binding constraint (chip supply, ~200 GW/yr by 2030) and add huge costs: testing then shipping unreliable GPUs (15% of Blackwells need RMA) to orbit could cost ~6 months of a 5-year useful life—the most valuable months—while inter-satellite networking requires expensive, unreliable space lasers instead of high-volume pluggable transceivers.
Destroying Taiwan fabs would collapse global compute growth
Even if every TSMC process engineer were airlifted out and the fabs destroyed in a Taiwan conflict, replicating that built-up capacity elsewhere would take years; incremental compute capacity would fall from hundreds of gigawatts/year toward just 10-20 GW across Intel and Samsung, massively shrinking (not just slowing) global GDP and leaving China with the strongest remaining vertically integrated supply chain.
Robot intelligence should be centralized in the cloud
For millions of humanoids, most intelligence should live in the cloud—a capable model batch-processing at high efficiency does planning and long-horizon tasks, pushing commands to robots that interpolate actions and handle only local sensing like weight and force—because on-device compute can't batch, can't be as intelligent, and would divert scarce leading-edge low-power chips from AI data centers.
Memory crunch will cut smartphone volumes drastically
Rising memory prices will push smartphone volumes down from ~1.1B toward ~800M this year and ~500-600M next year, concentrated in low and mid-range where memory is a larger share of BOM and margins are thinner; since cuts hit the low end disproportionately, DRAM is freed for AI while high-end phones (Apple) absorb only modest BOM increases (~$150-250 per iPhone).
Scale-up topologies differ: Nvidia all-to-all vs Google torus
A scale-up domain is the tight set of chips communicating at terabytes/second; Nvidia's NVL72 connects 72 GPUs all-to-all, Google's TPU pods reach thousands of chips but use a torus topology where each chip connects only to six neighbors (requiring bounce-through with resource blocking), and Amazon sits between them—while all three are migrating toward dragonfly topologies that mix fully- and partially-connected elements.
Advanced packaging scales chips per package on any node
Putting more dies per package (Blackwell's two, Rubin Ultra's four, Tesla's wafer-scale Dojo with 25, Huawei Ascend's progression) is a path to higher in-package bandwidth that works on older nodes too; Huawei leans on packaging scaling because it can't shrink process, and anything done on 7nm packaging can also be done on 3nm.
Fast timelines favor US, long timelines favor China
If AI takeoff is fast, the US wins because its labs are scaling compute to ~10 GW each, distillation from American models gets harder as work moves from visible reasoning chains to opaque automated white-collar output, and ROIC on its ~$1T data-center CapEx compounds; but if AI takes longer to reach high capability, China can catch up by building a fully vertical indigenized supply chain while the West's middling returns leave it exposed.
H100 prices have risen, not fallen
H100 rental prices have inflected upward, with some AI labs signing two-to-three-year deals as high as $2.40/hour for Hopper even though it costs ~$1.40/hour to build over five years—margins far above the original ~35%—because rolling-off short-term contracts let willing buyers crowd out other suppliers.
My Notes
Loading notes...