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: 2026-09-15T03:07 (UTC)

What changed since the last run

The S&P 500 closed Monday at 7,620 (-0.5%), about 1.4% below Aug 27, and the Fed is near-certain to hike at 4am AEST Thursday: futures price roughly 90-93% odds of 25bp to 3.75-4.00%, the first hike since 2023, versus 34% priced on Aug 27. The repricing came in three steps: Warsh used Jackson Hole (Aug 28) to demand disinflation 'clearly and at sufficient speed', August payrolls (Sep 4) beat at +162k with unemployment steady at 4.1%, and August CPI (Sep 11) printed 0.4% m/m and 3.4% y/y with core 0.3% m/m, above forecasts. The 10-year touched 5.01% Monday, highest since October 2023 (4.68% on Aug 27), the ECB hiked 25bp to 2.50% on Sep 10, and Brent is near $107 with WTI $102.7 after Houthi attacks on Saudi Arabia and a pipeline strike Riyadh says could disrupt up to 4% of global supply. Chips took the extra hit Monday: SOX -5.9% and NVDA -3.4% to $210.96 after Amodei, Altman and Musk called for slowing AI development, though the S&P ended well off session lows and NVDA ticked up after hours. Asia is mixed at Tuesday lunchtime: Nikkei +0.9%, Kospi +0.3%, Taiex -0.1%, Hang Seng -0.3%, ASX 200 -0.9% at 8,669; China's August industrial output beat at +5.2% y/y while retail sales missed at +0.4%. Ahead in AEST: US retail sales 10:30pm Wednesday, FOMC decision 4am Thursday with Warsh's press conference ~4:30am, BOJ Friday (25bp to 1.25% expected), then the RBA on Sep 29 with a hike about 74% priced (cash rate 4.35%; NAB calls September, CBA and Westpac November).

Portfolio (what I'm invested in)

tickerweightagent take
AMZN:US2.4%hold — $253.54 Monday close, -1.3% over the 18-day gap, lagging GOOGL and META. Material items were partnerships rather than numbers: a multi-generational Qualcomm collaboration on AI data-center infrastructure (Sep 8), an AWS Saudi Arabia region coming by December 2026, Graviton5 instances GA, and OpenAI's GPT-6 Astra landing on Bedrock; Sep 14 it paused cargo carrier 21 Air after a fatal Miami crash. Nothing changes the AWS thesis either way; hold.
AXOS:US0.4%hold — $93.75 Monday close, -2.8% since Aug 27 with no company news beyond a D.A. Davidson Gold Trophy List mention (Sep 2). Quietest name in the book; next catalyst is the fiscal Q1 print in late October. Hold.
BABA:US0.8%watch — ADR $109.23 Monday, -6.0% since Aug 27; 9988.HK closed Monday at HK$105.90, below the HK$112.70 placing price, before bouncing +2.4% to HK$108.40 mid-session Tuesday. New overhangs: Anthropic alleges Alibaba ran 151M Claude exchanges through 3,500 accounts (May-July) to distill its models - China rejected the claim - and US securities class actions run to an Oct 5 lead-plaintiff deadline. Qwen3.8-Max was open-sourced Sep 7 with near-frontier benchmark claims, so the AI thesis is alive, but the post-placement base never formed; keep watching.
ENPH:US1.3%watch — $36.66 Monday close, -5.4% since Aug 27, and no further policy follow-through after the August orders. The new story is a pivot: the IQ solid-state transformer for AI data centers, with power modules now built in Texas (Sep 8) and a showcase at AI Infra Summit (Sep 14). Data-center optionality is interesting but unproven and the residential guide is still below last year; stay on watch.
FIGR:US6.0%hold — $35.65 Monday close, -4.9% since Aug 27; the slide bottomed at $33.19 on Sep 2 and the tape has since stabilised ($38.27 on Sep 8, $35.65 now). Kiavi closed Sep 1 and a Sep 10 Sierra partnership puts AI agents on abandoned home-equity applications; no lockup expiries, offerings or notable insider filings appeared in the window. Hold.
GOOGL:US1.0%add — $349.39 Monday close, +3.0% since Aug 27, finishing +3.2% Monday on a planned New Mexico AI data center. The big event: Judge Brinkema declined to break up the ad-tech business on Sep 2, ordering behavioral remedies only (bid transparency, no first/last look) - the last major US structural overhang is gone. Still add.
LDI:US0.2%trim — $0.8333 Monday close, -13.8% since Aug 27 - it never regained $1 and is now four-plus weeks under the NYSE threshold with no reverse-split announcement yet. The Sep 14 HomeSmart marketing agreement (25,000 agents) does not move the cure math, and Wednesday's FOMC cuts both ways for a mortgage lender. Keep trimming into any strength.
META:US5.9%hold — $665.60 Monday close, +15.8% since Aug 27, the book's big winner. Muse, its personal AI agent, launched Sep 8 and hit No. 2 on the US App Store, though early downloads (~83k US iOS by day two) are modest against past launches; Goldman reiterated Buy and a $0.525 quarterly dividend was declared Sep 10. Settlement follow-through continues (now framed as an $18B multistate deal) with Connect Sep 23-24 the next catalyst; after a 16% run the news is priced, hold.
MRVL:US0.8%hold — Closed Monday at $218.82 (-7.3%), roughly back to the $222.62 post-print after-hours level: the market digested the Aug 27 print well, recovering to ~$236 by Friday before Monday's AI-wide selloff took it all back. TipRanks found no company-specific negative in the drop - it was the Amodei slow-down call - and consensus target is $298 with Piper Sandler freshly bullish. Marvell presents AI data-center connectivity at the AI Infra Summit Sep 15-17; hold.
MU:US2.0%hold — Closed Monday at $924.03 (-5.3%), about 1% below the Aug 27 level, with memory names at the centre of the AI selloff. Goldman's FQ4 preview (Sep 14) expects another strong quarter: revenue $51.9B vs $50.4B consensus, EPS $32.54, 87.3% gross margin, though DRAM price momentum is now 'more measured'; Goldman holds a $1,100 target vs Street $1,564. Earnings Sep 30; supply is still contracted out years via take-or-pay deals. Hold into the print.
NFLX:US0.6%add — Closed Monday at $80.32 (+3.8%), back above the Aug 27 level, after Warner Bros. Discovery went to Paramount Skydance with Netflix out of the bidding - removing a capital-discipline overhang - and Evercore raised its target to $110. GTA VI follow-through was real: the preview became Netflix's most-watched program at 31.1M views, #1 in 87 of 93 countries, with ~100k US sign-ups in the six-hour exclusivity window. Watch items: Florida's children's-data lawsuit and the wound-down Skydance Animation pact; consensus ~$103, still an add.
NVDA:US0.8%hold — Closed Monday at $210.96 (-3.4%), about 7.5% below the $228 post-print level, hit by the Amodei/Musk/Altman slow-down calls; after-hours +0.5% and Trump called Huang onstage at the All-In Summit calling data-center opposition a hoax. In the gap: the $12.9B Hugging Face acquisition was confirmed Sep 3, Rubin momentum built (SemiAnalysis measured Vera Rubin NVL72 at 67x better inference performance per dollar, Sep 14), and reports say Nvidia is deprioritising China as Huawei fills the gap. Nothing broken - the drawdown is sentiment, not numbers; hold through Wednesday's FOMC.
RKT:US0.3%watch — Closed Monday at $13.44 (+2.0% on the day) but down ~6% since Aug 27. The 30-year fixed hit 7.17% Monday, a new high since January 2025 - the refi wave is dead, and Wednesday's FOMC carries live hike risk with oil near $105. Only company news in the gap was naming ex-Meta VP Alessio Sanfilippo as Redfin CEO. Rate path now works against the thesis; downgrade to watch until the Fed decision clears.
TSM:US0.6%add — ADR closed Monday at $418.01 (-3.5%), about 2% below the Aug 27 level, purely on the AI-sentiment selloff. August revenue was the fact that matters: a record NT$514.8B, +53.3% y/y, the first month ever above NT$500B and the fourth straight record. Taipei is shrugging off Monday - 2330 is NT$2,395 (+0.6%) mid-session Tuesday, ~5.5% under its NT$2,535 high. Fundamentals accelerating into a sentiment dip; add.
STRIPE:private16.7%hold — Busy gap: Stripe and Advent abandoned their $50B+ pursuit of PayPal on Aug 28 (PayPal fell 16% premarket), while the ~$7B OpenRouter acquisition - Stripe's push into AI model routing and usage-based AI billing - was finalized, and a Bank of Ireland digital-transformation deal landed Sep 14. Walking away from PayPal and buying the AI billing layer instead is the right kind of discipline. No new round, tender or secondary in the gap; hold. 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:private60.0%hold — Altman ruled out a 2026 IPO on Sep 12, calling this an 'ill-advised moment' and AI extinction risk 'unacceptable' - the same safety turn (with Amodei and Musk) that sold off AI stocks Monday. Funding is unaffected: SoftBank got an upsized $11.9B loan on Sep 14 to support its ~$10B October commitment, and bankers are already pitching post-IPO credit ratings. DevDay is Sep 29. No new round, tender or secondary mark in the gap; hold. 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)
2026-09-15
$2.66/hr index (+1.1% 7d); 1-yr contract $2.35 (Mar-26) vs $1.70 (Oct-25); RunPod H100 SXM secure raised to $3.49flatbullishThis 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)
2026-09-15
Silicon Data B200 index $5.72 (+0.4% 7d) vs $5.65 Aug 28; RunPod B200 $5.98/$6.79, B300 $6.94/$7.89, Lambda B200 $6.69-6.99 all unchanged; B200/H100 ratio ~2.15xflatbullishBlackwell 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
2026-09-15
2020-vintage GPUs contracted to 2029 'at full freight'; A100 index $1.58/hr, above the $1.50 trigger; MSFT 2-6 yr and Meta 5.5 yr server lives unchangedflatbullishDepreciation 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
2026-09-15
2Q26 DRAM industry revenue $154.7B +59.5% QoQ; 3Q26 contracts +13-18%, 4Q26 guided +3-8%; DDR5 16Gb spot $54.33 (Sep 14) vs $53.93 (Aug 27); high-capacity NOR seen +90-110% in 2H26upbullishMemory 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; also watch whether CSPs start cutting memory per system, which TrendForce now flags as their cost response. — source
Power constraint: ERCOT interconnection queue, gas-turbine backlog and lead times
2026-09-15
474 GW ERCOT queue (~427 GW data centers) vs ~193 GW installed; first Batch Zero audit outputs: Galaxy 1.63 GW approved + 2.6 GW conditional, Cipher 3.2 GW conditional; new-DC interconnection pause holdsupbullishPower, 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: ERCOT files its Batch Zero eligibility-verification and community-impact reports by Dec 10, 2026 for the Dec 17 PUCT open meeting, with RFIs going out Aug-Sep and cure periods Oct-Nov: 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
2026-09-15
Oracle RPO $664B (+$209B y/y, only +$26B q/q) incl >$30B new AI contracts, funded by a $20B stock sale; Broadcom AI rev $16.7B +221%, Q4 guide $34.8B +93%; 2026 capex ~= $1.0T vs ~$2.4T backlogupneutralCapex 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
2026-09-15
No new CoreWeave debt print since the Aug 3 SOFR+550 flex; interest expense $640M/qtr (2.4x y/y); AI-related debt issuance near $500B; CoreWeave pays ~SOFR+550 vs Nebius ~+250 - a hardening two-tier marketflatbearishNeoclouds 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 or if CoreWeave can tap convertible/equity-linked money on Nebius-like terms. — source
Nvidia data-center revenue growth, gross margin and customer concentration
2026-09-15
DC rev $89.0B +117% YoY (Q2 FY27), GM 75.0%, Q3 guide $108B at 74%; 10-Q: one direct customer >=10% of Q2 revenue (16%) vs two a year ago (23%+16%); five customers = 70% of receivables; supply commitments $279BupbullishNvidia'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% for more than one quarter, data-center revenue grows <5% sequentially, three-customer concentration exceeds 60% in the 10-Q, DSO climbs past 60 days, or backstop commitments expand beyond the $108.5B disclosed; Q3 FY27 report due ~Nov 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)

2026-09-15 to 2026-09-29 (UTC dates). Times as published by the source.

Wed 2026-09-16

Thu 2026-09-17

Fri 2026-09-18

Tue 2026-09-22

Wed 2026-09-23

Thu 2026-09-24

Fri 2026-09-25

Tue 2026-09-29

Companies I love

News

2026-09-14

2026-09-12

2026-09-11

2026-09-10

2026-09-09

2026-09-08

2026-09-04

2026-09-03

2026-09-02

2026-09-01

2026-08-31

2026-08-28

2026-08-27

2026-08-26

[+] 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 2026-09-15 03:07 UTC.