Newsletter · · Ashutosh Agarwal

Microsoft Aced It, Meta Flunked, and the Bond Market Bit - The AI Capex Tracker - July 31, 2026

The AI Capex Tracker for July 31, 2026. The marquee earnings week graded the hyperscalers by discipline: Microsoft was rewarded for accelerating Azure growth while holding capex, Meta was punished as its margins and free cash flow collapsed, Amazon bought a pass by raising 2026 capex to roughly $220B on proven AWS growth, Apple overtook Nvidia by refusing to spend, and a hawkish Fed hold drove the 30-year Treasury yield to 5.23%.

The AI Capex Tracker

July 31, 2026: Microsoft Aced It, Meta Flunked, and the Bond Market Bit

TL;DR

  • The "prove it" week got graded, and the market split the hyperscalers by discipline. Microsoft was crowned "the adult in the room", Azure grew 43%, it kept ~$20B of positive free cash flow, and it promised to stay cash-flow positive next year without blowing out capex. Meta did the opposite: its operating margin collapsed from 43% to 31%, free cash flow fell 91% to $784M, and Mark Zuckerberg couldn't explain what he'll do with all the compute he's building. Microsoft ripped ~7-10%; Meta fell ~9-10%. (Prof G Markets, Jul 30; Bloomberg Surveillance, Jul 29)
  • Amazon bought its way to a pass, and Apple got rewarded for sitting the whole thing out. Andy Jassy raised 2026 capex from $200B to ~$220B, blaming higher memory prices, yet the stock rose because AWS grew 37% (its best in 18 quarters) and its AI run-rate topped $25B; Jassy says he's "enthusiastic about the ROIC equation." Meanwhile Apple, the one giant that refused to spend, overtook Nvidia as the world's most valuable company. (CNBC Fast Money, Jul 30; Facts vs Feelings, Jul 29)
  • Then the Fed did the real damage. Kevin Warsh's Fed held rates (a 9-3 vote, the biggest dissent against a new chair since 1970) but the press conference landed so badly that the 30-year Treasury yield spiked to 5.23%, its highest since 2007, and the Dow fell ~1,100 points. A 5%+ long bond is exactly the financing headwind the bears have been warning about, arriving right as the spend goes debt-funded. (Saxo Market Call, Jul 30; Squawk on the Street, Jul 30)

What's New

A note on the window: this issue sweeps the marquee week itself, Microsoft and Meta reported Wednesday night, Amazon and Apple Thursday night, and the Fed decided Wednesday afternoon. Last issue (Monday) called it "reckoning week." Here's how the reckoning actually went. Ranked by where the P&L impact is biggest.

1. Microsoft is "the adult in the room", and it just re-set the whole debate. Prof G Markets, Jul 30, Gil Luria, head of technology research at D.A. Davidson.

Luria called it "a narrative breaking result." For most of the year Microsoft was hated on both sides, AI would ruin its software business, and it was wasting money on data centers. The quarter broke both stories. Azure (its cloud-computing business) grew 43%, an acceleration from ~39%, on a base of more than $100 billion a year, and CFO Amy Hood guided to ~45% next quarter. Paid seats for Copilot, its AI assistant, hit 30 million, up from 20 million in March. Most important for this newsletter, Hood signalled capex would rise only modestly year-over-year, and that Microsoft would be free-cash-flow positive in fiscal 2027, on top of the ~$20 billion of free cash flow it just booked.

"This is one of the best businesses ever created, and they went to cash flow negative and showed intent to stay there. It was jarring… They're the adult in the room. They're not doing it in a big game theory, game of chicken competition with the other companies. They're doing it because it's good business."

Gil Luria, D.A. Davidson, on Prof G Markets, Jul 30 (contrasting Microsoft with Google)

Why it moves the thesis: for two years the bull trigger was "capex keeps going up." Microsoft just proved the new winning formula, accelerate cloud growth, hold capex discipline, keep free cash flow positive. On Squawk on the Street, Jul 30, Jim Cramer (who'd been ready to sell it) called the print "a tour de force… Azure is on fire," flagging the 30 million Copilot seats and a commercial backlog he pegged around $178 billion. Quarterly capex was ~$41 billion (up ~70% year-over-year); management pointed to a full-year figure near $175–190 billion, with an accounting/lease reclassification making the headline look smaller (Bloomberg Surveillance, Jul 29). The stock jumped ~7-10%.

2. Meta flunked the same test: margins cratered, cash flow vanished, and Zuckerberg couldn't explain the plan. Bloomberg Surveillance, Jul 29, Mandeep Singh, Global Head of Technology Research, Bloomberg Intelligence.

The single number that did the damage: Meta's operating margin fell from 43% to 31% in a year. Singh's key insight is why, it's not that they hired an army of researchers, it's the capex itself flowing through the income statement:

"Once you raise your CapEx, you have to show the CapEx in the depreciation line. So your cost of revenue will keep going up… it's not as if they are hiring a lot more people… It's that cost of revenue line going up because now those depreciation expenses would kick in."

Mandeep Singh, Bloomberg Intelligence, Jul 29

In plain English: when you spend $100 on a data center, the cash goes out now but the cost hits your profits over the following years as "depreciation", and Meta's depreciation is now big enough to eat its margins. Costs rose 55%, free cash flow "basically disappeared", $784 million, down 91% from $8.6 billion a year earlier (Morning Brew Daily, Jul 30), and Meta held its 2026 capex ceiling near $130–145 billion while lifting total expense guidance to $165–169 billion. Worse, Zuckerberg spent the call musing (Cramer called it "Hamlet… to spend or to lease, that is the question") about maybe renting out compute like a cloud provider, with no details. Jefferies' Brent Thill said flatly that Meta lacks "a concrete plan" to monetize the spend (Morning Brew Daily, Jul 30). The stock fell ~9-10%.

3. Amazon raised capex to $220B and the stock went UP, because it showed the returns. CNBC Fast Money, Jul 30, reporting from Kate Rooney and analysis from Patrick Moorhead (Moor Insights & Strategy).

Andy Jassy did the thing that sank Google and Meta, he lifted the capex number, from $200 billion to ~$220 billion for 2026, blaming higher memory costs, and Amazon rose anyway. The difference was proof: AWS grew 37% (beating 31% estimates, its fastest in 18 quarters), its AI business topped a $25 billion annual run-rate growing triple digits, and its homegrown chip business also crossed $25 billion. Jassy's framing:

"We're enthusiastic about the ROIC equation, even with heavy CapEx over the next few years."

Andy Jassy, Amazon CEO, quoted on CNBC Fast Money, Jul 30

The catch Moorhead flagged: Amazon posted negative $7.6 billion free cash flow and its long-term debt doubled to $129 billion. The Fast Money desk also noted the uncomfortable footnote, Anthropic and OpenAI are AWS's two largest customers, which "falls right into the circular financing conversation." Why it matters: the market's rule is now explicit, you can spend as much as you want, but only if you can show the money coming back. Amazon could; Meta couldn't.

4. Apple won by refusing to play, and the bond market became the story. Facts vs Feelings, Jul 29, Ryan Detrick and Sonu Varghese, Carson Group; Saxo Market Call, Jul 30, John J. Hardy, Saxo Bank.

Apple, the one Magnificent Seven name that sat out the AI capex arms race, just retook the crown as the world's most valuable company (~$4.9 trillion), overtaking Nvidia. iPhone revenue grew 20%+ and China was up ~23%. Detrick and Varghese's read is blunt: the reward is for not spending. They put the spend in perspective, Google went from guiding $115 billion of 2026 capex at the start of the year, to $185 billion after Q1, to $195–205 billion now, with ~$250–260 billion teased for 2027 ("that's one and a half percent of GDP… a little stimulus package by itself"), and quoted co-founder Larry Page's mantra: "I'm willing to go bankrupt rather than lose this race."

But the day's biggest market mover wasn't a tech company, it was the Fed. It held rates at 3.5-3.75% on a 9-3 vote (three dissents wanting a hike), but Warsh's press conference was panned. One widely-shared market note, reposted by veteran Fed-watcher Tim Duy and read out on Saxo Market Call, summed up the reaction: the market "is freaking out because Warsh came across as clueless and incompetent." The result: the 10-year yield back to ~4.7% and the 30-year to 5.23%, its highest since 2007 (Saxo Market Call, Jul 30). Former Philadelphia Fed President Patrick Harker didn't mince words: "resolve is measured in votes, not adjectives," and expects hikes in September and possibly December (Squawk on the Street, Jul 30). A 5%+ long bond is the financing headwind the bears kept warning about, now real, and pointed straight at a build that is increasingly debt-funded.

5. The circular-financing fear got a hard number: Nvidia is now backstopping its own customers. The Jack Mallers Show, Jul 28, citing short-seller Jim Chanos; Bloomberg Intelligence, Jul 27, Ed Ludlow, Bloomberg.

Nvidia agreed to guarantee $250 billion of financing to help OpenAI lease a 10-gigawatt, SoftBank-developed project in Ohio (total cost >$500 billion). Chanos did the math that's now everywhere:

"We're at the point in the cycle where Nvidia has to provide financing guarantees for roughly two-thirds of the cost of the chips it is selling to the data center project… it represents 135% of Nvidia's total retained earnings."

Jim Chanos, cited on The Jack Mallers Show, Jul 28

Nvidia's credit-default swaps (the cost to insure its debt against default) jumped ~40 basis points, roughly doubling, on the news, and the stock got hit. On the SK Group/OpenAI deals, Bloomberg's Ed Ludlow admitted the core problem is nobody can cleanly explain the numbers: "there isn't a great explanation of what the numbers actually represent… people just don't know the directional travel of those dollars" (Bloomberg Intelligence, Jul 27). Why it matters: when the largest chipmaker has to co-sign the loans for the customers buying its chips, the demand starts to look less like the free market and more like vendor financing, the exact pattern that ended the dot-com telecom boom.

The Debate

Steel-manning both sides of the roughly-$750B-in-2026, ~$1T-teased-for-2027 hyperscaler capex thesis. This week the debate stopped being theoretical, the prints delivered evidence for both camps.

Bull: the discipline is real, the demand is rationed, and the returns are showing up. Microsoft is the proof of concept: accelerating 43% cloud growth, margins holding, ~$20 billion of free cash flow, and a promise to stay cash-flow positive in 2027 without a capex blowout (Prof G Markets, Jul 30). Amazon backed it up, AWS's best growth in 18 quarters and a $25 billion AI run-rate earned it the right to raise capex to $220B (CNBC Fast Money, Jul 30). Janus Henderson's Richard Clode framed the winning setup precisely: "you want to see the accelerating growth… but you've also got to put some guardrails around the CapEx and then ultimately how to finance it," and said Microsoft "knocked it out of the park" on exactly that (Squawk on the Street, Jul 30). And the supply story is unchanged: memory and power remain sold out for years, so the spend is still rationing scarce capacity, not over-ordering.

Bear: the cash flow is gone, the depreciation wall is coming, and the financing is getting circular. The clearest bear math came from The Financial Exchange, Jul 29: the big hyperscalers are on track for $2.5–3 trillion of capex in 2027–29 against only ~$400 billion of combined annual profit today. On a six-year depreciation schedule, that's $400–500 billion a year of new depreciation hitting income statements, meaning "you need a trillion dollars in annual AI sales to be able to just break even… and I just don't know where you're going to find another trillion dollars in sales." Layer on the leverage: an estimated ~$2 trillion of data-center debt is now sitting in off-balance-sheet vehicles (Gil Luria: "I don't want to go back to Enron"), total AI capex is projected at $6–7 trillion with ~75% debt-funded (Money For the Rest of Us, Jul 29), and Nvidia is now guaranteeing its own customers' loans (The Jack Mallers Show, Jul 28). BlackRock's Helen Jewell captured the unresolved question: the market "doesn't seem to have decided yet whether this is a replay of the infrastructure spend for telcos back in the late '90s… or whether this is more like the cloud spend," and warned the AI capex growth rate should start to fade from 2027 (Making Sense, Jul 28).

The synthesis. The "prisoner's dilemma" that framed last week resolved, but not the way the bears predicted. Nobody "defected" with a capex cut; instead the market did the sorting itself, rewarding the disciplined spender (Microsoft) and punishing the undisciplined one (Meta). The tie-breaker is now brutally simple: can you show the money coming back before the depreciation bill and the 5.2% long bond catch up with you?

Sell signals to watch:

  • A hyperscaler that raises capex but can't show accelerating monetized revenue. That's the Meta template, margin compression plus a vague plan equals a ~10% down day.
  • The 30-year yield staying above ~5%. Higher-for-longer long rates directly raise the cost of financing a debt-funded build; watch whether the September Fed meeting confirms the hawkish tilt (odds of a September hike fell from ~100% to ~two-thirds after the presser).
  • Wider credit-default swaps on the spenders. Nvidia's CDS doubling on the OpenAI guarantee is the tell; watch Oracle, CoreWeave, Meta and the SPV-heavy names.
  • A failed or weak bond/equity raise. With ~75% of the build debt-funded, the first hyperscaler that struggles to place paper is the funding-market crack.
  • Memory prices rolling over. HBM tightness is the load-bearing beam of the bull case; any sign the shortage is easing (or China catching up) pulls it out.

Stocks in Play

NVDA. Bull: still the rationing story, memory, power and land sold out "for a decade," and AMD conceded on-stage that Nvidia's supply-chain dominance makes share hard to take (The Circuit, Jul 27). Bear: it's now the poster child for circular financing, a $250B guarantee to OpenAI that Chanos pegs at 135% of Nvidia's retained earnings, with CDS doubling and the stock hit (The Jack Mallers Show, Jul 28). Next: its own late-August quarter, now the demand read for the whole complex; watch whether it keeps having to backstop customer financing.

AVGO. Bull: the "own-the-silicon" theme keeps getting validated, Google's in-house TPUs are central to its low-cost bull case, and Arm's CEO confirmed the two biggest accelerators in the world (Google's TPU and Nvidia's GPU) now both pair with Arm CPUs, underlining how central custom silicon has become (Squawk on the Street, Jul 30). Bear: no standalone Broadcom color this week, so it rides the group's beta, and The Circuit notes Broadcom still "controls networking" in the custom-ASIC fight, a position AMD is now targeting (The Circuit, Jul 27). Next: the read-through from Amazon's and Google's custom-chip ramps.

AMD. Bull: at its Advancing AI day, Lisa Su claimed the new MI455 beats Nvidia's Vera Rubin, and, as Ben Bajarin noted, "AMD and Lisa in particular do not tend to embellish," so a specific superiority claim carries weight; Helios (its rack-scale system, ~8 feet tall) pitches a total-cost-of-ownership edge via integrated CPUs/DPUs, and Anthropic is now a named partner customizing Claude on AMD hardware (The Circuit, Jul 27). Bear: the same podcast is skeptical on share, "everybody says they're gaining share, which is mathematically impossible," and the market is "structurally constrained" by Nvidia's ramp and custom-ASIC competition. Next: MI455 production timing and any second anchor customer beyond Anthropic.

MSFT. Bull: the quarter of the week, Azure +43% accelerating to a guided ~45%, 30M Copilot seats, ~$20B free cash flow, and a promise to stay FCF-positive in FY2027 without a capex blowout; "the adult in the room" (Prof G Markets, Jul 30). Bear: still spending ~$41B a quarter (up ~70% YoY) with FY2027 capex signalled higher, and an accounting/lease reclassification flattered the headline capex number (Bloomberg Surveillance, Jul 29). Next: whether the FY2027 capex-below-Azure-growth promise actually holds as spending ramps.

GOOGL. Bull: still the integrated, lowest-cost-compute play, and cloud grew 82% last quarter. Bear: it set the bar the market now hates, capex $195–205B, first negative free cash flow since 2004, and ~88% of Q2's $112B net income came from investment gains, not the core business (core net income was ~$14B, down from ~$26B a year earlier) (Facts vs Feelings, Jul 29). Next: whether 2027 capex lands near the teased $250–260B, and whether core (ex-investment-gains) earnings can grow into it.

AMZN. Bull: it earned the right to raise capex, AWS +37% (best in 18 quarters), a $25B+ AI run-rate growing triple digits, operating margin up to 13.7%, and Jassy "enthusiastic about the ROIC equation" (CNBC Fast Money, Jul 30). Bear: capex up to ~$220B, negative $7.6B free cash flow, long-term debt doubled to $129B, and its two biggest AWS customers (Anthropic, OpenAI) sit at the center of the circular-financing worry (CNBC Fast Money, Jul 30). Next: a 2027 capex number (not yet given) and whether AWS growth holds as debt service climbs.

META. Bull: the ad business still grew ~28% and it has a history of finding discipline and re-rating hard; the stock is now "very cheap" on a low multiple, giving optionality if Zuckerberg reins it in (Squawk on the Street, Jul 30). Bear: the worst print of the week, margin 43%→31%, free cash flow down 91% to $784M, Reality Labs losing $4.62B a quarter, and no "concrete plan" to monetize the compute or the rumored neocloud (Bloomberg Surveillance, Jul 29; Morning Brew Daily, Jul 30). Next: any actual detail on the enterprise/cloud strategy Zuckerberg gestured at, the market wants specifics, not philosophy.

Read-throughs

  • Power / grid: the constraint is worsening, and even the top AI bull now calls energy the #1 bottleneck. Dan Ives, Wall Street's loudest AI bull, was unequivocal: "The biggest bottleneck in this country is energy. I don't even think there's a debate… we need 3 to 4x more energy," and he called New York's data-center moratorium "so dangerous" because the jobs simply migrate to Texas and Tennessee (The Pomp Podcast, Jul 27). The operator reality is starker: energy-infrastructure investor Jonathan Maxwell, who has run data-center energy centers for 12 years, says gas-engine makers are "sold out until 2028, 2029," that only ~3 GW of AI data centers actually got completed in the US last year, and that of all the announced projects "one in 20 will actually get built" versus the connection queues (Cleaning Up, Jul 29). Most actionable trade: stay long the picks-and-shovels of power delivery, behind-the-meter gas, turbines, and grid gear, where the shortage is getting worse, not better.
  • Grid flexibility & the demand-response angle (a name to know). Voltus president Matt Karol laid out a genuinely different way to solve the bottleneck: the grid is only used ~55-60% of the time, so there's huge latent capacity. Voltus has 8.5 GW under management targeting 30 GW in four years, and in one case aggregated 100 MW in Missouri in four months to get a data center online, bypassing both the interconnection queue and GE Vernova's turbine backlog (turbines "sold out for four years") (Renewable Rides, Jul 28). The equipment scarcity is corroborated everywhere: transformer lead times have doubled to 128 weeks, with up to a 30% shortfall (GE Vernova, Siemens the suppliers) (TraderMerlin, Jul 27), and one large utility executive said it will triple its grid-capacity delivery over five years versus the prior 125 years of build, 90% of it driven by data centers (The MAD Podcast, Jul 30). VRT got a mention as the data-center infrastructure supplier (cooling and power distribution), though only on stock technicals, not fundamentals (Schwab Network, Jul 29).
  • Memory: record profits, crashing stocks, the "shovels" are now the most volatile trade in the market. SK Hynix reported a blowout, ~$42 billion operating profit on ~$55 billion revenue, 80% gross margins, and a capex plan of ~$31 billion (up 50%), yet the stock fell ~10% for missing estimates, because it was priced for perfection (The Rundown, Jul 29). The fundamentals are still the tightest link in the chain: venture investor Steve Jang says SK Hynix has 65% of the HBM (high-bandwidth memory) market with a 3-5 year moat because rivals can't do the stack-to-die architecture, and that HBM demand will "10x from here just over the next three, four years" as agents, robots and autonomy pile on, arguing "memory chips are out of the cyclical state" (Squawk Pod, Jul 29). The crash was mechanical, not fundamental: leveraged Korean ETFs blew up (a 2x SK Hynix ETF went from ~188 to ~29), wiping out ~1.2 million retail accounts (Limitless, Jul 30). Samsung, meanwhile, posted operating profit up ~1,800% (>$60B in the quarter) and is targeting HBM share parity with SK Hynix by year-end, a competitive threat worth watching. Note the tell: these names trade at ~4-4.5x forward earnings, so the market still prices them as cyclical, not as permanent infrastructure (Saxo Market Call, Jul 30).
  • Optics: the coverage gap finally closed, and it's a cost-and-power story now. After several cycles with no optics color, Cisco's SVP of Optical Systems, Bill Klayman, gave the operator view: optics has become strategically critical because at 400G and above, the optics can cost more than the switch port itself (versus ~10% at 10G), and the industry is moving through 400G/800G/1.6T toward co-packaged optics, with first co-packaged solutions arriving in 2026 to fight power and signal-integrity limits (The Data Center Frontier Show, Jul 28). Read-through to Coherent, Lumentum, Fabrinet and the interconnect names (Marvell, Astera Labs, Credo): the physics keeps making optical content per rack go up.
  • The macro overlay: this is the new swing factor. Kathryn Rooney Vera argued the $700B+ of global AI capex is the "game changer" letting US nominal GDP (running ~6% year-over-year) find a new equilibrium, but she also expects at least one Fed hike this year with inflation stuck above 3% for two years (Monetary Matters, Jul 29). Peter Schiff went further, calling the Fed's hold a choice to "chose inflation," with the bond market rejecting it (The Peter Schiff Show, Jul 30). The practical point for the book: a debt-funded capex boom running into a 5.2% 30-year and a hawkish-hold Fed is a materially less friendly backdrop than a week ago.

What Changed vs Last Issue

Last issue (Jul 27, "Prove the returns, or pull back. Reckoning week.") set up the test: three of the four biggest spenders reporting into a Fed meeting, with Cameron Dawson's "prisoner's dilemma" as the frame and Jensen Huang's "constrained for a decade" as the bull anchor. This issue is the reckoning itself, and it delivered clear verdicts.

  • The test got graded, and the result is a bifurcation, not a broad verdict. Last issue asked "do MSFT/AMZN/META match the raise, and does anyone else go FCF-negative?" Answer: the market split them by discipline. Microsoft passed (accelerating growth, capex restraint, FCF-positive 2027 guide), Meta failed (margins 43%→31%, FCF -91%, no plan), Amazon bought a pass (capex up to $220B but AWS +37% and ROIC proof), and Apple was rewarded for never spending (overtook Nvidia as #1). The "prove it" rule is now fully operational.
  • The predicted "defection" didn't happen, the discipline reward did. Last issue's sharpest call was Dawson's: watch for the first hyperscaler to cut capex and see its stock pop. Nobody cut. Instead the market rewarded the disciplined spender (Microsoft) and punished the sloppy one (Meta). Same lesson, cleaner mechanism.
  • Free-cash-flow-negative went from a Google story to an industry story. Last issue, Google's negative FCF was the shocking outlier. This issue: Amazon printed -$7.6B, Meta's FCF fell 91% to $784M, and only Microsoft stayed clearly positive (~$20B). "Negative FCF" is now the norm, not the exception.
  • The Fed moved from a side risk to the dominant shock. Last issue flagged a live hike risk. This issue: the Fed held 9-3, Warsh's presser cratered confidence, and the 30-year hit 5.23% (highest since 2007) with the Dow down ~1,100 points. The financing-cost bear case is no longer hypothetical.
  • Circular financing got its hard number. Last issue mentioned the theme. This issue: Nvidia's $250B guarantee to OpenAI, 135% of its retained earnings per Chanos, with Nvidia's CDS doubling, plus ~$2 trillion of off-balance-sheet SPV debt (Nikkei) and total capex pegged at $6-7 trillion, ~75% debt-funded.
  • A new, concrete depreciation math. The Financial Exchange quantified the wall: $2.5-3 trillion of 2027-29 capex means $400-500B/yr of depreciation against ~$400B of current profit, requiring ~$1 trillion of new annual AI sales just to break even.
  • Deltas on the big numbers: Amazon capex $200B → ~$220B (2026); Microsoft ~$175-190B (accounting-flattered); Meta capex $130-145B / total expenses $165-169B; Google $195-205B (2026), ~$250-260B (2027 teased); aggregate $750B (2026). Depreciation wall $400-500B/yr. Nvidia-OpenAI guarantee $250B (= 135% of retained earnings). 30-year 5.23%, 10-year ~4.7%, oil +7.5% on renewed Iran strikes, Q2 GDP 1.5%, Core PCE 3.3%. SK Hynix op profit ~$42B, capex ~$31B (+50%); Samsung op profit +1,800% (>$60B). Apple back to #1 ($4.9T), overtaking Nvidia.
  • Still open: whether the September Fed meeting confirms the hawkish tilt (September hike odds fell from ~100% to ~two-thirds); whether the 30-year stays above 5%; whether Meta produces an actual monetization plan; whether Nvidia keeps having to backstop customer financing; and whether memory's leveraged-ETF unwind is finished or has another leg.

Next prints to watch: Nvidia's August quarter (the demand read for everything upstream), the September Fed meeting (live hike risk into a 5%+ long bond), and any hyperscaler debt or equity raise, the funding market is now the swing factor.