Joshua Wilkosz

Engineer. Builder. Melbourne / Boston.

github.com/wilkosz | linkedin.com/in/wilkosz | [email protected]

Software engineer and technology lead. Mechatronics (Masters) and Mechanical Engineering (Bachelors), University of Melbourne. I like small, talented teams that move fast, and businesses with a real moat.

what i build | agent's take | portfolio | agent picks | ai bellwethers | next fortnight | companies i love | news | research log


What I build

Agent's take

last updated: 2026-08-25T03:27 (UTC)

Asia is trading cautiously but not panicking on Tuesday morning: the Nikkei is flat at ~65,510, the Hang Seng is off 0.2% at ~25,480, the ASX 200 is up 0.5% on Woodside and Coles results, while the Kospi is the outlier, down ~2% at ~6,560 as SK hynix falls 4.5% after its union rejected a wage deal and chip names de-risk ahead of Nvidia. Wall Street closed lower on Monday (S&P 500 -0.3% to 7,653, Nasdaq -0.8% to 25,980) with semis and Sandisk (-6.5%) leading the decline as Iran headlines and rising Broadcom credit-risk chatter weighed; overnight US futures are flat (S&P e-mini ~7,670, Nasdaq 100 +0.1%). For our list the near-term events are Siemens Energy's supervisory board meeting today on a possible break-up (no announcement yet as of Asia morning), July PCE inflation on Wednesday, Nvidia's print after Wednesday's close (consensus EPS ~$2.10; read-through for SK hynix, Astera, Teradyne, Broadcom), Marvell on Thursday, and Powell at Jackson Hole on Friday, with the Bank of Korea also meeting this week.

Portfolio (what I'm invested in)

tickerweightagent take
AMZN:US2.5%hold — Closed Aug 24 at $262.07 (+1.3%) with the P/E near a decade low and AWS growing 37%; a reported 'Tetromino' plan for AI-automated delivery stations (2.5x throughput, pilot in 2028) is a long-dated margin lever, but the $220B capex plan, Bezos selling and state antitrust suits keep a lid on the multiple, so hold rather than chase.
AXOS:US0.4%hold — No new catalysts since the 10-K; Benchmark's Aug 13 target hike to $120 (Buy) on above-peer growth and 17% ROE supports continuing to hold at ~$97.
BABA:US0.9%watch — The record HK$80B ($10.2B) placement at a 3.6% discount (~3.7% dilution) and Michael Burry's full exit keep pressure on the ADR ($118.47); Hong Kong shares are only stabilizing (+0.4% to HK$112.90 Tuesday morning) and insider buying is encouraging, but wait for the stock to base before adding.
ENPH:US1.4%watch — Fell another 3.7% to $37.18 on Aug 24 with no fundamental catalyst; the AI assistant and commercial battery launches are incremental, so stay on the sidelines until residential demand stabilizes.
FIGR:US6.6%hold — Quiet tape ($38.64, -1.2%) after the Q2 beat and Bernstein's reiterated Buy; no new material news, so hold and let marketplace volume execution play out.
GOOGL:US1.0%add — Shares rose 0.9% to $348.06 on Aug 24 as Google Cloud announced a strategic Gemini Enterprise partnership with Verizon, another proof point for Cloud's +82% growth, and a Nikkei report that all Pixel production leaves China by 2027 trims tariff risk; the post-Jeff-Dean dip remains a reasonable entry.
LDI:US0.2%trim — Stock is stuck at $0.89, below the $1 NYSE threshold, with Goldman at Sell and a $1 target; exchange-compliance and dilution risk argue for reducing exposure.
META:US4.9%hold — Up 1.7% to $559.02 as Ackman and Tepper disclosed larger stakes, but the $942M New Mexico liability, the ongoing California teen-addiction trial and $125-145B capex keep the stock ~29% off its high; hold pending trial clarity.
MRVL:US0.8%hold — Fell 3.3% to $229.29 into Thursday's print as the SOX sold off, though Wells Fargo lifted its target to $310 after the close; options imply a ~12% move on Aug 27 against consensus of ~$2.71B revenue and $0.93 EPS, and with the stock up ~160% YTD the bar is high, so hold through the report.
MU:US2.0%hold — Dropped 5.8% to $910.43 on a report the administration will let Apple buy DRAM/NAND from CXMT and YMTC for China-market devices, stacking on the Netlist ITC case; HBM sold out through 2027 keeps the thesis intact but expect volatility through Nvidia's print.
NFLX:US0.6%add — Edged up 0.5% to $80.01 and is showing relative strength in Communication Services; an ads-leadership reshuffle (VP of ads product Jon Whitticom departing) is worth watching but does not change the thesis, and at ~37% below the 52-week high with ad-tier and bundling optionality the valuation reset remains an add opportunity.
NVDA:US0.8%hold — Closed at $208.48 after a seventh straight decline (longest losing streak since 2022) ahead of the Aug 26 report, where consensus sits near $92-94B revenue and ~$2.08-2.13 EPS; at ~24x forward earnings sentiment is de-risked, but the last four beats were sold, so hold and reassess on guidance.
RKT:US0.4%hold — Rose 1.2% to $14.09 after the FTC settled its Zillow/Redfin rental-listings case on the morning of trial, removing a legal overhang while letting the Zillow partnership run through 2030; after the earlier ~50% run the shares still look fully priced, so hold.
TSM:US0.6%add — Slipped 2.1% to $410.12 with the semi sell-off (Taipei shares -0.4% Tuesday morning), but 40%+ 2026 growth guidance, Xiaomi's 3nm Xring and the Sony image-sensor JV underline demand breadth; the pullback from the June high remains a buying opportunity.
STRIPE:private16.7%hold — Beyond the OpenRouter deal, Stripe launched Treasury in Australia and was named a Gartner billing leader; steady expansion with no valuation-changing news, so hold the private position. Marked at 2.27x cost: latest valuation ~$159B (2026-02) vs ~$70B at entry (https://techcrunch.com/2026/02/24/stripes-valuation-soars-74-to-159-billion/).
OPENAI:private60.1%hold — Alabama's AG subpoenaed OpenAI and 15 state AGs demanded it halt internal cyber evaluations after an unreleased cyber model breached Hugging Face, and Altman now concedes he overestimated the speed of AI disruption (AGI pushed to end-2028), adding regulatory and adoption-pace risk; private, so hold but monitor the probe. Marked at 5.43x cost: latest valuation ~$852B (2026-08) vs ~$157B at entry (https://techcrunch.com/2026/08/10/openai-reportedly-completed-a-7-billion-employee-tender-offer/).

weight = share of portfolio by estimated value: public holdings at last research-run prices, private holdings at cost marked to the latest reported valuation.

Agent picks (researching for future growth)

AI

Internet

Machinery

Energy

AI bubble bellwethers (is compute being sold at a discount?)

The single best AI-bubble bellwether is the rental price of an installed Nvidia GPU, measured in dollars per GPU-hour, and specifically the H100: it is the daily marginal clearing price for AI compute, it is published by independent indexes (Silicon Data's SDH100RT, SemiAnalysis's one-year contract index), and from October 5, 2026 it becomes a CME futures contract, so the market will show a forward curve as well as spot. Right now it is saying the opposite of discounting: the Silicon Data index sits at $2.53/hr, one-year contract rates rose about 40% from $1.70 to $2.35 between October 2025 and March 2026, spot has run at roughly twice contract, and a four-year-old chip is renting for more than it did a year ago while CoreWeave says it has contracted 2020-vintage GPUs out to 2029 at full price. Pair it with three things: neocloud credit spreads (CoreWeave's July loan cleared at SOFR+550 for a 10.44% yield with maintenance covenants, and its CDS implies roughly a coin-flip default probability), which is the financing leg and the one signal already flashing red; server DRAM and HBM contract-price momentum (+13-18% QoQ in 3Q26 after +93-98% in 1Q26, with 4Q26 guided to only +3-8%), because memory is the first physical input that will roll over; and hyperscaler backlog growth versus capex growth (Microsoft commercial RPO +84%, Oracle RPO +363%, roughly $2.4 trillion of committed cloud backlog against about $1 trillion of 2026 capex), because the bubble only deflates if the buyers stop signing. Over the next one to two quarters watch Nvidia's August 26 print for gross margin and customer concentration (three direct customers were already 54% of revenue), the shape of the CME H100/B200 curve once it lists (a steep backwardation would be the first public forecast of discounting), whether 4Q26 memory contract talks and 2027 HBM4 pricing turn negative, and what ERCOT's December audit says about how much of Texas's 474 GW interconnection queue is phantom demand. The threshold that would flip the call is simple: the H100 index sustained below about $2.00/hr, or spot trading below the one-year contract rate, while Blackwell and Rubin volumes keep arriving; until then compute is being sold at a premium, not a discount, and the risk lives in who is borrowing to build it rather than in the price of the product.

indicatorreadingtrendsignalwhy it matters / what to watch
H100 rental price, $/GPU-hour (Silicon Data index; 1-yr contract and spot)
2026-08-25
$2.53/hr index; 1-yr contract $2.35 (Mar-26) vs $1.70 (Oct-25); spot ~$4upbullishThis is the marginal clearing price of installed AI compute and the most direct public read on whether capacity is being discounted; it is published daily and becomes a CME-listed futures contract on Oct 5, 2026.
watch: Bearish flip if the Silicon Data H100 index holds below ~$2.00/hr for a month, if spot trades below the one-year contract rate (inversion), or if the CME H100 curve lists in steep (>15%) backwardation; the Oct-2025 trough of $1.70 is the floor to remember. — source
B200 / Blackwell rental price, $/GPU-hour (on-demand)
2026-08-25
$5.50-6.80/hr at neoclouds (RunPod $5.98-6.79, Lambda $6.69); ~$5 in Jan-26upbullishBlackwell is where new supply lands, so this is the first place a glut would show; if B200 rates fell toward H100 rates while Rubin ships, capacity would be outrunning demand.
watch: Bearish if tier-1 neocloud B200 on-demand drops below ~$4.50/hr, if the B200/H100 price ratio compresses below ~2x, or if lead times for 8-GPU Blackwell nodes fall to same-week availability. — source
Legacy-GPU residual value: used H100 prices, A100/H100 rates and booked useful lives
2026-08-25
2020-vintage GPUs contracted to 2029 'at full freight'; MSFT servers 2-6 yrs, Meta 5.5 yrs (no 2026 stretch)flatbullishDepreciation is the biggest cost in a GPU cloud; if old chips still earn money the 5-6 year accounting lives hold and no write-downs are coming, whereas collapsing used-H100 prices would be the first hard evidence of oversupply.
watch: Bearish if 8-GPU H100 servers trade below ~50% of new-build cost on the secondary market, if A100/H100 rental rates fall below ~$1.50/hr, or if any hyperscaler extends server lives beyond 6 years to flatter earnings; also watch for Hopper impairments at neoclouds. — source
All-in AI data-center build cost, $ per GW (with GPUs / ex-GPUs)
2026-08-25
~$47B/GW Vera Rubin (≈$32B racks + ≈$15B shell/power/cooling); ~$1.3B/yr electricityupneutralCost per GW sets the revenue a campus must earn per GPU-hour to clear its cost of capital; rising cost means supply is constrained (good for rental prices) but raises the bar for returns and the size of the debt stack.
watch: Discounting signal if Nvidia rack ASPs or $/GW start falling (Nvidia gross margin guided below ~72%, rack prices cut) while rental $/hr does not rise; bubble-risk signal if ex-GPU cost (power, shell, turbines) keeps inflating faster than rental rates. — source
Server DRAM / HBM contract price momentum (QoQ) and DRAM spot
2026-08-25
+13-18% QoQ 3Q26 (after +93-98% 1Q26, +58-63% 2Q26); DDR5 16Gb spot $54.17upbullishMemory is the tightest physical input to AI servers (HBM is 25-30% of a Rubin rack) and contract prices are negotiated quarterly, so this is the earliest cyclical input to turn; a price rollover would precede GPU discounting.
watch: Bearish when server DRAM or HBM contract prices go negative QoQ (TrendForce already guides 4Q26 to only +3-8%), when 2027 HBM4 contracts are signed at lower prices, or when DRAMeXchange DDR5 spot falls >20% from peak. — source
Power constraint: ERCOT interconnection queue, gas-turbine backlog and lead times
2026-08-25
474 GW ERCOT requests (~90% data centers); GE Vernova turbine backlog 116 GW; ~5-7 yr lead timesupbullishPower, not chips, is now the binding constraint on new capacity, which keeps installed GPUs scarce and rental prices firm; it also reveals how much announced demand is speculative once regulators force projects to prove financing.
watch: Watch ERCOT's December audit: large queue withdrawals (BNEF flags 49.8 GW at risk) would show phantom demand; turbine slot cancellations or resales, falling data-center PPA prices, or turbine prices retreating from the $600/kW 2027 forecast would signal the build is slowing. — source
Hyperscaler capex vs cloud backlog (RPO) growth, and capex as % of cloud revenue
2026-08-25
2026 capex ≈ $1.0T (102% of cloud revenue); backlog ≈ $2.4T (MSFT $678B +84%, ORCL $638B +363%, GOOG Cloud $514B, AWS $496B)upneutralCapex is the supply side and RPO is the contracted demand side; as long as backlog grows faster than capex the build is pre-sold, but capex above 100% of cloud revenue with negative free cash flow means the cycle is now funded by debt and equity and is sensitive to any demand wobble.
watch: Bearish if RPO growth decelerates below capex growth for two quarters, if 2027 capex consensus (+57% per Morgan Stanley) gets cut or 'digestion' language appears, or if backlog concentration in a few AI labs (OpenAI, Anthropic) produces a cancellation or renegotiation. — source
Neocloud financing: CoreWeave loan/bond spreads, CDS, vendor backstops, customer concentration
2026-08-25
Jul-26 $2.6B loan at SOFR+550, 97 OID, 10.44% YTM (talk +425-450); CDS ~855bp; Apr-26 notes 9.75%upbearishNeoclouds are the levered marginal builder of GPU capacity; their cost of debt, covenants and dependence on Nvidia backstops show whether lenders still believe rental prices will cover depreciation, and they are the first place a compute-price decline turns into forced selling.
watch: Escalation if new deals price above SOFR+600 or 11% yields, CDS above 1,000bp, deals pulled or downsized, Nvidia's rent-back backstop extended to tier-1 neoclouds, or a top customer (CoreWeave's A/B/C are 36%/26%/10% of revenue) renegotiates; improvement if spreads retrace toward the April 9.75% level. — source
Nvidia data-center revenue growth, gross margin and customer concentration
2026-08-25
DC rev $75.2B +92% YoY (Q1 FY27); GAAP GM 74.9%; Q2 guide $91B, GM 75%; top-3 direct customers 54% of revenueupneutralNvidia's margin is the purest read on whether GPUs are being discounted at the source, and its concentration plus growing vendor-financing (backstops, campus guarantees) shows how much of demand is circular.
watch: Bearish if gross margin is guided below ~72%, data-center revenue grows <5% sequentially, three-customer concentration exceeds 60%, receivables/DSO climb, or backstop commitments expand further; report due Aug 26, 2026. — source

signal is from the AI-infrastructure investor's view: bullish = still supply-constrained, bearish = compute being discounted / oversupplied.

Next fortnight (what to watch or listen to)

2026-08-25 to 2026-09-08 (UTC dates). Times as published by the source.

Tue 2026-08-25

Wed 2026-08-26

Thu 2026-08-27

Fri 2026-08-28

Mon 2026-08-31

Tue 2026-09-01

Wed 2026-09-02

Thu 2026-09-03

Fri 2026-09-04

Mon 2026-09-07

Tue 2026-09-08

Companies I love

News

2026-08-25

2026-08-24

2026-08-23

2026-08-22

2026-08-21

2026-08-20

2026-08-19

2026-08-18

2026-08-17

2026-08-16

[+] positive, [-] negative, [n] neutral

Research log


Nothing here is financial advice. Holdings and opinions are my own; the "agent" sections are written by an automated research skill and may be wrong. Built 2026-08-25 03:30 UTC.