Goldman Sachs Tech Trader: Optics Outperform Memory by Over 20%, Open-Source Model Costs Less Than 8% of Closed-Source
Taylor Wilson
Goldman Sachs tech trader Peter Callahan flags three structural shifts in 10 charts: AI-agent traffic now exceeds human traffic, optical stocks have outpaced memory by over 20 percentage points this quarter, and open-source models deliver better performance at under 8% of closed-source cost.
Bot traffic has overtaken humans — what does that mean?
Cloudflare disclosed that non-human traffic has surpassed human traffic; third-party data shows bots accounted for 53% of all web traffic in 2025.
Cloudflare projects that if the trend holds, non-human traffic will be 1,000× human traffic within five years — "humans will become a rounding error on the internet."
This means → the internet's primary "users" are shifting from people to AI agents. The structural change is present tense, not future.
In plain terms = the heaviest internet users going forward will be AI programs working on behalf of humans — whoever sells "the roads AI travels on" (bandwidth, optical interconnects) benefits first.
Are companies actually seeing returns on AI spending?
Booking said its AI investment has achieved positive ROI, with related costs still in the low single digits of total tech spending.
MercadoLibre spent roughly $80 million on AI this quarter, cut customer-service staff from 10,000 to 7,000, and now handles 90% of customer interactions without a human — while business volume tripled.
This means → AI is no longer just a "spend now, hope later" story. Leading companies are proving measurable cost savings with real financial data.
Why did optical stocks suddenly outpace memory by so much?
Both optical and memory stocks fell roughly 30% in July. Since August began, optical names have outperformed memory by over 20 percentage points.
Leaders: Coherent rallied for 7 straight sessions, up 73%; Lumentum +43%, Fabrinet +36%, Corning +34%, Ciena +24%.
NAND, DRAM, and HDD names gained just 0–20% over the same period — a clear lag.
This means → the market is repricing the compute-demand structure for AI inference. Demand for optical interconnects — moving data between chips with light instead of electricity — is surging faster than expected, and capital is flowing there first.
Open-source models at under 8% of closed-source cost — who should worry?
Pinterest reported that fine-tuning an open-source model on its proprietary data — retraining an existing open model with the company's own data — outperformed closed-source third-party models, at under 8% of the per-transaction cost.
This means → the pricing power of closed-source models is being undercut at the foundation. If a company can get better results for less than a tenth of the price, the premium logic for closed-source vendors becomes very hard to sustain.
In plain terms = open-source is like building your own PC; closed-source is like buying a branded pre-built. The DIY machine now performs better and costs a fraction — the brand premium collapses.
Where do SaaS and Nvidia stand right now?
Earnings season has widened the sentiment gap between data-infrastructure / dev-tools companies and traditional SaaS application vendors.
ServiceNow closed above its 200-day moving average for the first time in over a year. Atlassian has bounced roughly 160% off its intra-year low, yet remains down about 6% year-to-date — that contradiction itself captures where SaaS sits today.
Nvidia gained 11.5% last week, its best weekly performance since May 2025. Callahan sees the upcoming earnings report, the Vera Rubin architecture launch, and a packed conference season as potential catch-up catalysts.
What is Goldman's overall call on tech?
IT led the S&P 500 in Q2 earnings season: 82% of companies beat on earnings, 72% beat on revenue.
The Nasdaq 100 rose 5% last week; the VIX closed near 14, a year-to-date low.
Goldman characterizes the current market as "a return to normal, not a boom," and describes the earnings-growth slowdown as "a structural shift, not a reversal."
This means → Goldman maintains a constructive stance on tech, but the rationale is not "expect a surge." It is clean positioning, reasonable valuations, and solid fundamentals — a "can buy" call, not a "must chase" call.
Content is for reference only, not financial advice.